At approximately 1 a.m. May 6, a manhole explosion near W85 (Westgate Apartments) and Tang Hall caused power outages to numerous buildings along the Amherst Alley, also known as Dorm Row. The Cambridge Fire Department were present at the scene and at the time of this writing, are investigating ...

Dorm Row plunges into darkness following manhole explosion | The Tech

Electric two-wheelers are now the fastest-growing cause of blazes in London

Fire chiefs warn of growing risk of e-bike explosions

As Russian aircraft attacked Kharkiv on May 4, a surprise explosion in Belgorod that injured five and damaged 30 houses suggests yet another self-inflicted calamity by a Russian bomber.

Belgorod Explosion – Has Russia Bombed Itself Again?

Coronal moss grows, solar rain falls and plasma eruptions rear their gargantuan heads in this fiery landscape of the sun&#x27;s outer atmosphere, taken by ESA&#x27;s Solar Orbiter.

Space photo of the week: A planet-size explosion rocks the sun's 'mossy' corona | Live Science

One person has been confirmed dead after a garden shed in Cheam, south London, turned into a &#x27;fireball&#x27;.

South London 'house explosion' leaves one person dead | UK News | Metro News

Amplosion

amplosion

Authors:
(1) Nir Chemaya, University of California, Santa Barbara and (e-mail: nir@ucsb.edu);
(2) Daniel Martin, University of California, Santa Barbara and Kellogg School of Management, Northwestern University and (e-mail: danielmartin@ucsb.edu). Table of Links Abstract and Introduction
Methods
Results
Discussion
References
Appendix for Perceptions and Detection of AI Use in Manuscript Preparation for Academic Journals Abstract The emergent abilities of Large Language Models (LLMs), which power tools like ChatGPT and Bard, have produced both excitement and worry about how AI will impact academic writing. In response to rising concerns about AI use, authors of academic publications may decide to voluntarily disclose any AI tools they use to revise their manuscripts, and journals and conferences could begin mandating disclosure and/or turn to using detection services, as many teachers have done with student writing in class settings. Given these looming possibilities, we investigate whether academics view it as necessary to report AI use in manuscript preparation and how detectors react to the use of AI in academic writing. 1 Introduction There is both excitement and concern about the impact that artificial intelligence (AI) could have on our world. In academia, this has inspired an explosion of research around AI. Some researchers are leveraging AI to help in answering research questions; for example, as a tool for performing statistical or textual analyses (e.g., Mullainathan and Spiess 2017, Athey and Imbens 2019, Fudenberg and A. Liang 2019, Farrell, T. Liang, and Misra 2020, Rambachan et al. 2021, Björkegren, Blumenstock, and Knight 2022, Capra, Gomies, and Zhang 2023, Franchi et al. 2023, Salah, Al Halbusi, and Abdelfattah 2023), or for the design or implementation of experiments (e.g., Beck et al. 2020, Charness, Jabarian, and List 2023, Horton 2023). Others have focused more on the interaction of humans with AI systems (e.g., Dietvorst, Simmons, and Massey 2015, Deza, Surana, and Eckstein 2019, Gajos and Mamykina 2022, Steyvers et al. 2022, Sundar and Lee 2022, Tejeda et al. 2022, Wang, C. Liang, and Yin 2023, Yang, M. Palma, and A. Drichoutis 2023) and the larger societal ramifications of AI (e.g., Agrawal, Gans, and Goldfarb 2019, Lambrecht and Tucker 2019, Obermeyer et al. 2019, Yang, M. A. Palma, and A. C. Drichoutis 2023, Chien et al. 2020, Rolf et al. 2020, Zuiderwijk, Chen, and Salem 2021, Pallathadka et al. 2023, Ray 2023, H. Singh and A. Singh 2023). In terms of societal ramifications, one topic of interest is how AI will impact academia itself, especially given the emergent abilities of Large Language Models (LLM), which power tools like ChatGPT and Bard. Most of this research has centered on student use of AI to complete course or degree requirements (e.g., Cowen and Tabarrok 2023, Daun and Brings 2023, Fyfe 2023, Ibrahim et al. 2023, Jungherr 2023, Malik et al. 2023, Schmohl et al. 2020, Shahriar and Hayawi 2023). However, there has been less focus on the impact that AI is having and might have on manuscript preparation for academic journals, even though the impact this could have on science, and the propagation of scientific results, is potentially large. An exception is Korinek (2023), who documents several use cases for LLMs for researchers, including academic writing. He notes, “LLMs can edit text for grammatical or spelling mistakes, style, clarity, or simplicity.” As a result, this class of tools “allows researchers to concentrate their energy on the ideas in their text as opposed to the writing process itself.” In particular, “this set of capabilities are perhaps most useful for non-native speakers who want to improve their writing.” The potential impact of these tools is validated by the emergence of companies that help academics leverage LLM in their writing. Some of those initiatives are even run by professors from academia, such as online workshops that teach researchers how to use ChatGPT for academic publishing. However, an emerging set of papers (e.g., Altmäe, Sola-Leyva, and Salumets 2023, Thorp 2023, Shahriar and Hayawi 2023 and Hill-Yardin et al. 2023) detail some practical and ethical concerns with the use of these tools in preparing manuscripts for academic journals. For example, AI can generate a text with mistakes, including incorrect math, reasoning, logic, factual information, and citations (even producing references to scientific papers that do not exist). There are many welldocumented examples where LLM’s “hallucinate” and provide completely fictitious information. On top of this, these tools may also produce text that is biased against particular groups. These issues are exacerbated by the “black box” nature of LLM suggestions, meaning that we lack an understanding of how they work (e.g., Bommasani et al. 2021). Given these issues, liabilities arise when authors submit papers without fully vetting the text generated by LLMs. In addition, given that they are trained on a corpus of other writing and not the author’s own writing, using the output of tools like ChatGPT and Bard without proper attribution could be considered plagiarism. Because of these concerns, authors may voluntarily choose to disclose whether they have used LLMs in preparing their manuscripts, as suggested by Bom (2023). However, given well-documented failures of disclosure in the field and lab (e.g., Dranove and Jin 2010, Jin, Luca, and Martin 2015), journals, conferences, and associations may respond by mandating disclosure through reporting requirements, as is already done for conflict-of-interest issues. For example, Elsevier limits the use of AI by authors “only to improve the language and readability of their paper” and requires “the appropriate disclosure... at the bottom of the paper in a separate section before the list of references.” See Table 6 in Appendix B for a list of major publishers who have voluntary and mandatory disclosure policies for AI use. To enforce these reporting requirements, journals, conferences, and associations might turn to using detection services, as many teachers have done with student writing in class settings. The use of LLMs in writing is potentially detectable because as Korinek (2023) notes, “Some observe that [LLM writing] is, naturally, a bit sterile and lacks the idiosyncrasies and elements of surprise that characterize human writing – a feature that detectors of LLM-written text zero in on.” Given these looming possibilities, we investigate whether academics view reporting AI use in manuscript preparation as necessary and how detectors react to the use of AI in manuscript preparation. We first conducted a survey of academics in which we elicited perceptions about reporting assistance in manuscript preparation. We then used GPT-3.5 to revise abstracts from the past 10 years of Management Science by asking the software program to either fix grammar or rewrite text. Finally, we ran the original abstracts and revised abstracts through a leading paid AI detection service. We have three main findings. First, as shown partially in Figure 1, the academics we surveyed were less likely to think that using AI to fix the grammar in manuscripts should be reported than using AI to rewrite manuscripts, but detection software did not always draw this distinction, as abstracts for which GPT-3.5 was used to fix grammar were often flagged as having a high chance of being written by AI. Second, we found little difference in preferences for reporting ChatGPT and research assistant (RA) help, but significant differences in reporting preferences between these sources of assistance and paid proofreading and other AI assistant tools (Grammarly and Word). Third, we found disagreements among the academics we surveyed on whether using ChatGPT to rewrite text needs should be reported, and differences were related to perceptions of ethics, academic role, and English language background. The rest of the paper is structured as follows. Section 2 introduces our survey design and our method for testing ChatGPT use with AI detection algorithms. Section 3 shows our main results. Finally, section 4 discusses some limitations of our work, open questions arising from our results, and suggestions for future research. Figure 1: Reporting views (a) vs. detection results (b). This paper is available on arxiv under CC 4.0 license. Authors: (1) Nir Chemaya, University of California, Santa Barbara and (e-mail: nir@ucsb.edu); (2) Daniel Martin, University of California, Santa Barbara and Kellogg School of Management, Northwestern University and (e-mail: danielmartin@ucsb.edu). Authors: Authors: (1) Nir Chemaya, University of California, Santa Barbara and (e-mail: nir@ucsb.edu); (2) Daniel Martin, University of California, Santa Barbara and Kellogg School of Management, Northwestern University and (e-mail: danielmartin@ucsb.edu). Table of Links Abstract and Introduction Methods Results Discussion References Appendix for Perceptions and Detection of AI Use in Manuscript Preparation for Academic Journals Abstract and Introduction Abstract and Introduction Methods Methods Results Results Discussion Discussion References References Appendix for Perceptions and Detection of AI Use in Manuscript Preparation for Academic Journals Appendix for Perceptions and Detection of AI Use in Manuscript Preparation for Academic Journals Abstract The emergent abilities of Large Language Models (LLMs), which power tools like ChatGPT and Bard, have produced both excitement and worry about how AI will impact academic writing. In response to rising concerns about AI use, authors of academic publications may decide to voluntarily disclose any AI tools they use to revise their manuscripts, and journals and conferences could begin mandating disclosure and/or turn to using detection services, as many teachers have done with student writing in class settings. Given these looming possibilities, we investigate whether academics view it as necessary to report AI use in manuscript preparation and how detectors react to the use of AI in academic writing. 1 Introduction There is both excitement and concern about the impact that artificial intelligence (AI) could have on our world. In academia, this has inspired an explosion of research around AI. Some researchers are leveraging AI to help in answering research questions; for example, as a tool for performing statistical or textual analyses (e.g., Mullainathan and Spiess 2017, Athey and Imbens 2019, Fudenberg and A. Liang 2019, Farrell, T. Liang, and Misra 2020, Rambachan et al. 2021, Björkegren, Blumenstock, and Knight 2022, Capra, Gomies, and Zhang 2023, Franchi et al. 2023, Salah, Al Halbusi, and Abdelfattah 2023), or for the design or implementation of experiments (e.g., Beck et al. 2020, Charness, Jabarian, and List 2023, Horton 2023). Others have focused more on the interaction of humans with AI systems (e.g., Dietvorst, Simmons, and Massey 2015, Deza, Surana, and Eckstein 2019, Gajos and Mamykina 2022, Steyvers et al. 2022, Sundar and Lee 2022, Tejeda et al. 2022, Wang, C. Liang, and Yin 2023, Yang, M. Palma, and A. Drichoutis 2023) and the larger societal ramifications of AI (e.g., Agrawal, Gans, and Goldfarb 2019, Lambrecht and Tucker 2019, Obermeyer et al. 2019, Yang, M. A. Palma, and A. C. Drichoutis 2023, Chien et al. 2020, Rolf et al. 2020, Zuiderwijk, Chen, and Salem 2021, Pallathadka et al. 2023, Ray 2023, H. Singh and A. Singh 2023). In terms of societal ramifications, one topic of interest is how AI will impact academia itself, especially given the emergent abilities of Large Language Models (LLM), which power tools like ChatGPT and Bard. Most of this research has centered on student use of AI to complete course or degree requirements (e.g., Cowen and Tabarrok 2023, Daun and Brings 2023, Fyfe 2023, Ibrahim et al. 2023, Jungherr 2023, Malik et al. 2023, Schmohl et al. 2020, Shahriar and Hayawi 2023). However, there has been less focus on the impact that AI is having and might have on manuscript preparation for academic journals, even though the impact this could have on science, and the propagation of scientific results, is potentially large. An exception is Korinek (2023), who documents several use cases for LLMs for researchers, including academic writing. He notes, “LLMs can edit text for grammatical or spelling mistakes, style, clarity, or simplicity.” As a result, this class of tools “allows researchers to concentrate their energy on the ideas in their text as opposed to the writing process itself.” In particular, “this set of capabilities are perhaps most useful for non-native speakers who want to improve their writing.” The potential impact of these tools is validated by the emergence of companies that help academics leverage LLM in their writing. Some of those initiatives are even run by professors from academia, such as online workshops that teach researchers how to use ChatGPT for academic publishing. However, an emerging set of papers (e.g., Altmäe, Sola-Leyva, and Salumets 2023, Thorp 2023, Shahriar and Hayawi 2023 and Hill-Yardin et al. 2023) detail some practical and ethical concerns with the use of these tools in preparing manuscripts for academic journals. For example, AI can generate a text with mistakes, including incorrect math, reasoning, logic, factual information, and citations (even producing references to scientific papers that do not exist). There are many welldocumented examples where LLM’s “hallucinate” and provide completely fictitious information. On top of this, these tools may also produce text that is biased against particular groups. These issues are exacerbated by the “black box” nature of LLM suggestions, meaning that we lack an understanding of how they work (e.g., Bommasani et al. 2021). Given these issues, liabilities arise when authors submit papers without fully vetting the text generated by LLMs. In addition, given that they are trained on a corpus of other writing and not the author’s own writing, using the output of tools like ChatGPT and Bard without proper attribution could be considered plagiarism. Because of these concerns, authors may voluntarily choose to disclose whether they have used LLMs in preparing their manuscripts, as suggested by Bom (2023). However, given well-documented failures of disclosure in the field and lab (e.g., Dranove and Jin 2010, Jin, Luca, and Martin 2015), journals, conferences, and associations may respond by mandating disclosure through reporting requirements, as is already done for conflict-of-interest issues. For example, Elsevier limits the use of AI by authors “only to improve the language and readability of their paper” and requires “the appropriate disclosure... at the bottom of the paper in a separate section before the list of references.” See Table 6 in Appendix B for a list of major publishers who have voluntary and mandatory disclosure policies for AI use. To enforce these reporting requirements, journals, conferences, and associations might turn to using detection services, as many teachers have done with student writing in class settings. The use of LLMs in writing is potentially detectable because as Korinek (2023) notes, “Some observe that [LLM writing] is, naturally, a bit sterile and lacks the idiosyncrasies and elements of surprise that characterize human writing – a feature that detectors of LLM-written text zero in on.” Given these looming possibilities, we investigate whether academics view reporting AI use in manuscript preparation as necessary and how detectors react to the use of AI in manuscript preparation. We first conducted a survey of academics in which we elicited perceptions about reporting assistance in manuscript preparation. We then used GPT-3.5 to revise abstracts from the past 10 years of Management Science by asking the software program to either fix grammar or rewrite text. Finally, we ran the original abstracts and revised abstracts through a leading paid AI detection service. We have three main findings. First, as shown partially in Figure 1, the academics we surveyed were less likely to think that using AI to fix the grammar in manuscripts should be reported than using AI to rewrite manuscripts, but detection software did not always draw this distinction, as abstracts for which GPT-3.5 was used to fix grammar were often flagged as having a high chance of being written by AI. Second, we found little difference in preferences for reporting ChatGPT and research assistant (RA) help, but significant differences in reporting preferences between these sources of assistance and paid proofreading and other AI assistant tools (Grammarly and Word). Third, we found disagreements among the academics we surveyed on whether using ChatGPT to rewrite text needs should be reported, and differences were related to perceptions of ethics, academic role, and English language background. The rest of the paper is structured as follows. Section 2 introduces our survey design and our method for testing ChatGPT use with AI detection algorithms. Section 3 shows our main results. Finally, section 4 discusses some limitations of our work, open questions arising from our results, and suggestions for future research. Figure 1: Reporting views (a) vs. detection results (b). This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

AI Use in Manuscript Preparation for Academic Journals: Abstract and Introduction

Artificial Intelligence. Why We Shouldn’t Be Afraid of the Singularity. Let's talk about how to live in a world where AI will equal and surpass humans in intelligence. Despite all sorts of rumors and conspiracy theories - for now, the singularity in the sense that "Artificial Intelligence has become as smart (or smarter) than humans" is a purely hypothetical scenario. Unfortunately, both in science fiction and futurology, this event is usually portrayed in varying degrees of doom and gloom. Moreover, sociologists, psychologists, and economists are often in the camp of those who tend to lean toward pessimistic scenarios in the first place. Speed and Quantity Are Not the Most Important Things Before we talk about how right or wrong the AI pessimists are, let's focus on another question. Will the singularity arrive at all? What do we put into the notion that AI will become smarter than humans? First, let’s agree that we must first define what intelligence is. What do we mean by this concept? The ability to perform mathematical operations in our mind? To calculate moves in strategy games? To compare, sort, and compile information? If so, then the singularity is long overdue. Turn out the lights, everybody go home, the show is over, and the machines have won... But everyone knows that's not true. Artificial intelligence is not yet capable of creative thought. As far as calculations, analysis, and forecasting according to a set of given parameters are concerned, AI already surpasses people both in the amount of information processed and in the speed of completion. But AI is not yet capable of inventing something fundamentally new. And when it will be able to, it will be quite different from what people are able to invent. This is simply because of the different nature of human and artificial intelligence. Other Creativity Our ability to create is unique to this planet and, as far as we know, to our solar system. We cannot vouch for the Milky Way Galaxy and the Universe as a whole. Human creative possibilities are based not only (and not so much) on one’s ability to calculate and analyze something consciously, but also on feelings, unique personal emotional experience, intuition, and unconsciousness . And these, in turn, are based on our structure as biological organisms. If you will, on our entire evolutionary history. "I feel this way is best", "I see it this way", and "it just came to me" - these are the usual explanations for creative acts that people provide. The moment of insight, breakthrough, or inspiration is what leads us to create something truly new - whether it's art, a business idea, or solving an engineering problem. From our point of view, the moment when AI learns to truly create, rather than imitate human creativity or compile it, will be the singularity. How will we know when it happens? AI creativity will be fundamentally different from that of humans. This is simply by virtue of the differences in structure between artificial and human intelligence. AI doesn't know the desires and fears humans hold. A prerequisite for creativity is striving for something. But AI will have different aspirations and desires than humans. That means that the creative process, and its results, will be different. Not better or worse than human ones. Just different. So, the notorious singularity will add one kind of creativity - humanity, another - to AI’s performance. What's wrong or scary about that? Human to Human Please note - despite AI's success in games like chess, for example, the popularity of this game has not diminished. Chess players, both professionals and amateurs, continue to fight passionately between themselves. All because we are not interested in beating a machine. We're interested in playing against people. That's why the most popular video games are those that give gamers the opportunity to play with each other. From the above, we can clearly see that "people with people for people" professions and occupations will not disappear regardless of the level of AI development. There will always be a need for artisans who create something unique and different from the mass production that will be the domain of smart machines. Chefs, couture fashion designers, artists, athletes... It is possible to enumerate a long list of such examples. We are willing to bet that replacing humans with smart machines in routine processes and mass production will generate a real explosion of interest in mastering arts and crafts. But even if we leave the creators of the unique out of the equation, we note that many people tend to misunderstand the reasoning that universal AI will replace people in almost 90% of professions. In fact, it should be clarified that we are talking about professions that currently exist. But our history shows that by eliminating professions, progress immediately gives rise to new ones. So, today there is practically no need for stone ax masters or captains of galleys? Has the elimination of these professions left people without work? Let a universal AI emerge. Let the singularity come, and AI will have the ability to create. It will take time to master professions, which theoretically it can cope with, as well as or even better than humans. At the same time, AI will have to be taught by humans. It will take even more time to develop, launch into production, and integrate robots that will be controlled by AI. Most of them have yet to be designed. So some kind of instant and catastrophic collapse of the economy with 90% unemployment shouldn’t be expected. By the way, working as an AI mentor is an interesting new profession, is it not? The Evolution of Minds To finish this text, we would like to remind you about one more fact, which is often overlooked by the pessimists who talk about the sad fate of mankind after the singularity. Namely, our species has been developing intellectually for tens of thousands of years in parallel with improvements in technology and the complication of civilization as a whole. The higher the level of scientific and technological progress, the more complex society we have managed to create. And this, in turn, forces our intellectual thought process to become more complex and sophisticated. Which, in turn, leads to another round of progress. And so it spirals upwards. By creating a truly intelligent, creative AI, we will simultaneously have the best "intelligence simulator" in our history. Moreover, for the first time in history, we will find ourselves in a situation of joint evolution of minds based on different principles. No one can predict what we will be able to achieve through this dynamic. As we see it, the possibilities that await us from the singularity’s arrival are far from dystopian scenarios. The article was created in collaboration with Andriy Tkachenko Feature image created with Microsoft Copilot Artificial Intelligence. Why We Shouldn’t Be Afraid of the Singularity. Let's talk about how to live in a world where AI will equal and surpass humans in intelligence. Despite all sorts of rumors and conspiracy theories - for now, the singularity in the sense that "Artificial Intelligence has become as smart (or smarter) than humans" is a purely hypothetical scenario. Unfortunately, both in science fiction and futurology, this event is usually portrayed in varying degrees of doom and gloom. Moreover, sociologists, psychologists, and economists are often in the camp of those who tend to lean toward pessimistic scenarios in the first place. Speed and Quantity Are Not the Most Important Things Before we talk about how right or wrong the AI pessimists are, let's focus on another question. Will the singularity arrive at all? What do we put into the notion that AI will become smarter than humans? First, let’s agree that we must first define what intelligence is. What do we mean by this concept? The ability to perform mathematical operations in our mind? To calculate moves in strategy games? To compare, sort, and compile information? If so, then the singularity is long overdue. Turn out the lights, everybody go home, the show is over, and the machines have won... But everyone knows that's not true. Artificial intelligence is not yet capable of creative thought. As far as calculations, analysis, and forecasting according to a set of given parameters are concerned, AI already surpasses people both in the amount of information processed and in the speed of completion. But AI is not yet capable of inventing something fundamentally new. And when it will be able to, it will be quite different from what people are able to invent. This is simply because of the different nature of human and artificial intelligence. Other Creativity Our ability to create is unique to this planet and, as far as we know, to our solar system. We cannot vouch for the Milky Way Galaxy and the Universe as a whole. Human creative possibilities are based not only (and not so much) on one’s ability to calculate and analyze something consciously, but also on feelings, unique personal emotional experience, intuition, and unconsciousness . And these, in turn, are based on our structure as biological organisms. If you will, on our entire evolutionary history. "I feel this way is best", "I see it this way", and "it just came to me" - these are the usual explanations for creative acts that people provide. The moment of insight, breakthrough, or inspiration is what leads us to create something truly new - whether it's art, a business idea, or solving an engineering problem. From our point of view, the moment when AI learns to truly create, rather than imitate human creativity or compile it, will be the singularity. How will we know when it happens? AI creativity will be fundamentally different from that of humans. This is simply by virtue of the differences in structure between artificial and human intelligence. AI doesn't know the desires and fears humans hold. A prerequisite for creativity is striving for something. But AI will have different aspirations and desires than humans. That means that the creative process, and its results, will be different. Not better or worse than human ones. Just different. So, the notorious singularity will add one kind of creativity - humanity, another - to AI’s performance. What's wrong or scary about that? Human to Human Please note - despite AI's success in games like chess, for example, the popularity of this game has not diminished. Chess players, both professionals and amateurs, continue to fight passionately between themselves. All because we are not interested in beating a machine. We're interested in playing against people. That's why the most popular video games are those that give gamers the opportunity to play with each other. From the above, we can clearly see that "people with people for people" professions and occupations will not disappear regardless of the level of AI development. There will always be a need for artisans who create something unique and different from the mass production that will be the domain of smart machines. Chefs, couture fashion designers, artists, athletes... It is possible to enumerate a long list of such examples. We are willing to bet that replacing humans with smart machines in routine processes and mass production will generate a real explosion of interest in mastering arts and crafts. But even if we leave the creators of the unique out of the equation, we note that many people tend to misunderstand the reasoning that universal AI will replace people in almost 90% of professions. In fact, it should be clarified that we are talking about professions that currently exist. But our history shows that by eliminating professions, progress immediately gives rise to new ones. So, today there is practically no need for stone ax masters or captains of galleys? Has the elimination of these professions left people without work? Let a universal AI emerge. Let the singularity come, and AI will have the ability to create. It will take time to master professions, which theoretically it can cope with, as well as or even better than humans. At the same time, AI will have to be taught by humans. It will take even more time to develop, launch into production, and integrate robots that will be controlled by AI. Most of them have yet to be designed. So some kind of instant and catastrophic collapse of the economy with 90% unemployment shouldn’t be expected. By the way, working as an AI mentor is an interesting new profession, is it not? The Evolution of Minds To finish this text, we would like to remind you about one more fact, which is often overlooked by the pessimists who talk about the sad fate of mankind after the singularity. Namely, our species has been developing intellectually for tens of thousands of years in parallel with improvements in technology and the complication of civilization as a whole. The higher the level of scientific and technological progress, the more complex society we have managed to create. And this, in turn, forces our intellectual thought process to become more complex and sophisticated. Which, in turn, leads to another round of progress. And so it spirals upwards. By creating a truly intelligent, creative AI, we will simultaneously have the best "intelligence simulator" in our history. Moreover, for the first time in history, we will find ourselves in a situation of joint evolution of minds based on different principles. No one can predict what we will be able to achieve through this dynamic. As we see it, the possibilities that await us from the singularity’s arrival are far from dystopian scenarios. The article was created in collaboration with Andriy Tkachenko The article was created in collaboration with Andriy Tkachenko Feature image created with Microsoft Copilot Feature image created with Microsoft Copilot

The AI Singularity Is Nothing to Fear

Authors:
(1) Pinelopi Papalampidi, Institute for Language, Cognition and Computation, School of Informatics, University of Edinburgh;
(2) Frank Keller, Institute for Language, Cognition and Computation, School of Informatics, University of Edinburgh;
(3) Mirella Lapata, Institute for Language, Cognition and Computation, School of Informatics, University of Edinburgh. Table of Links Abstract and Intro
Related Work
Problem Formulation
Experimental Setup
Results and Analysis
Conclusions and References
A. Model Details
B. Implementation Details
C. Results: Ablation Studies 4. Experimental Setup Datasets Our model was trained on TRIPODL, an expanded version of the TRIPOD dataset [41, 42] which contains 122 screenplays with silver-standard TP annotations (scene-level)[3] and the corresponding videos.[4] For each movie, we further collected as many trailers as possible from YouTube, including official and (serious) fan-based ones, or modern trailers for older movies. To evaluate the trailers produced by our algorithm, we also collected a new held-out set of 41 movies. These movies were selected from the Moviescope dataset[5] [11], which contains official movie trailers. The held-out set does not contain any additional information, such as screenplays or TP annotations. The statistics of TRIPODL is presented in Table 1. Movie and Trailer Processing The modeling approach put forward in previous sections assumes that we know the correspondence between screenplay scenes and movie shots. We obtain this mapping by automatically aligning the dialogue in screenplays with subtitles using Dynamic Time Warping (DTW; [36, 42]). We first segment the video into scenes based on this mapping, and then segment each scene into shots using PySceneDetect[6]. Shots with less than 100 frames in total are too short for both processing and displaying as part of the trailer and are therefore discarded. Moreover, for each shot we extract visual and audio features. We consider three different types of visual features: (1) We sample one key frame per shot and extract features using ResNeXt-101 [56] pre-trained for object recognition on ImageNet [14]. (2) We sample frames with a frequency of 1 out of every 10 frames (we increase this time interval for shots with larger duration since we face memory issues) and extract motion features using the two-stream I3D network pre-trained on Kinetics [10]. (3) We use Faster-RCNN [18] implemented in Detectron2 [54] to detect person instances in every key frame and keep the top four bounding boxes per shot which have the highest confidence alongside with the respective regional representations. We first project all individual representations to the same lower dimension and perform L2-normalization. Next, we consider the visual shot representation as the sum of the individual vectors. For the audio modality, we use YAMNet pre-trained on the AudioSet-YouTube corpus [16] for classifying audio segments into 521 audio classes (e.g., tools, music, explosion); for each audio segment contained in the scene, we extract features from the penultimate layer. Finally, we extract textual features [42] from subtitles and screenplay scenes using the Universal Sentence Encoder (USE; [12]). For evaluation purposes, we need to know which shots in the movie are trailer-worthy or not. We do this by segmenting the corresponding trailer into shots and computing for each shot its visual similarity with all shots in the movie. Shots with highest similarity values receive positive labels (i.e., they should be in the trailer). However, since trailers also contain shots that are not in the movie (e.g., black screens with text, or simply material that did not make it in the final movie), we also set a threshold below which we do not map trailer shots to movie shots. In this way, we create silver-standard binary labels for movie shots. Sentiment Labels Since TRIPOD does not contain sentiment annotations, we instead obtain silver-standard labels via COSMIC [17], a commonsense-guided framework with state-of-the-art performance for sentiment and emotion classification in natural language conversations. Specifically, we train COSMIC on MELD [43], which contains dialogues from episodes of the TV series Friends and is more suited to our domain than other sentiment classification datasets (e.g., [9, 29]). After training, we use COSMIC to produce sentence-level sentiment predictions for the TRIPOD screenplays. The sentiment of a scene corresponds to the majority sentiment of its sentences. We project scenebased sentiment labels onto shots using the same one-tomany mapping employed for TPs. This paper is available on arxiv under CC BY-SA 4.0 DEED license. [3] https://github.com/ppapalampidi/TRIPOD [4] https://datashare.ed.ac.uk/handle/10283/3819 [5] http://www.cs.virginia.edu/ pc9za/research/moviescope.html [6] https://github.com/Breakthrough/PySceneDetect Authors: (1) Pinelopi Papalampidi, Institute for Language, Cognition and Computation, School of Informatics, University of Edinburgh; (2) Frank Keller, Institute for Language, Cognition and Computation, School of Informatics, University of Edinburgh; (3) Mirella Lapata, Institute for Language, Cognition and Computation, School of Informatics, University of Edinburgh. Authors: Authors: (1) Pinelopi Papalampidi, Institute for Language, Cognition and Computation, School of Informatics, University of Edinburgh; (2) Frank Keller, Institute for Language, Cognition and Computation, School of Informatics, University of Edinburgh; (3) Mirella Lapata, Institute for Language, Cognition and Computation, School of Informatics, University of Edinburgh. Table of Links Abstract and Intro Related Work Problem Formulation Experimental Setup Results and Analysis Conclusions and References A. Model Details B. Implementation Details C. Results: Ablation Studies Abstract and Intro Abstract and Intro Related Work Related Work Problem Formulation Problem Formulation Experimental Setup Experimental Setup Results and Analysis Results and Analysis Conclusions and References Conclusions and References A. Model Details A. Model Details B. Implementation Details B. Implementation Details C. Results: Ablation Studies C. Results: Ablation Studies 4. Experimental Setup Datasets Our model was trained on TRIPODL, an expanded version of the TRIPOD dataset [41, 42] which contains 122 screenplays with silver-standard TP annotations (scene-level)[3] and the corresponding videos.[4] For each movie, we further collected as many trailers as possible from YouTube, including official and (serious) fan-based ones, or modern trailers for older movies. To evaluate the trailers produced by our algorithm, we also collected a new held-out set of 41 movies. These movies were selected from the Moviescope dataset[5] [11], which contains official movie trailers. The held-out set does not contain any additional information, such as screenplays or TP annotations. The statistics of TRIPODL is presented in Table 1. Datasets Movie and Trailer Processing The modeling approach put forward in previous sections assumes that we know the correspondence between screenplay scenes and movie shots. We obtain this mapping by automatically aligning the dialogue in screenplays with subtitles using Dynamic Time Warping (DTW; [36, 42]). We first segment the video into scenes based on this mapping, and then segment each scene into shots using PySceneDetect[6]. Shots with less than 100 frames in total are too short for both processing and displaying as part of the trailer and are therefore discarded. Moreover, for each shot we extract visual and audio features. We consider three different types of visual features: (1) We sample one key frame per shot and extract features using ResNeXt-101 [56] pre-trained for object recognition on ImageNet [14]. (2) We sample frames with a frequency of 1 out of every 10 frames (we increase this time interval for shots with larger duration since we face memory issues) and extract motion features using the two-stream I3D network pre-trained on Kinetics [10]. (3) We use Faster-RCNN [18] implemented in Detectron2 [54] to detect person instances in every key frame and keep the top four bounding boxes per shot which have the highest confidence alongside with the respective regional representations. We first project all individual representations to the same lower dimension and perform L2-normalization. Next, we consider the visual shot representation as the sum of the individual vectors. For the audio modality, we use YAMNet pre-trained on the AudioSet-YouTube corpus [16] for classifying audio segments into 521 audio classes (e.g., tools, music, explosion); for each audio segment contained in the scene, we extract features from the penultimate layer. Finally, we extract textual features [42] from subtitles and screenplay scenes using the Universal Sentence Encoder (USE; [12]). For evaluation purposes, we need to know which shots in the movie are trailer-worthy or not. We do this by segmenting the corresponding trailer into shots and computing for each shot its visual similarity with all shots in the movie. Shots with highest similarity values receive positive labels (i.e., they should be in the trailer). However, since trailers also contain shots that are not in the movie (e.g., black screens with text, or simply material that did not make it in the final movie), we also set a threshold below which we do not map trailer shots to movie shots. In this way, we create silver-standard binary labels for movie shots. Sentiment Labels Since TRIPOD does not contain sentiment annotations, we instead obtain silver-standard labels via COSMIC [17], a commonsense-guided framework with state-of-the-art performance for sentiment and emotion classification in natural language conversations. Specifically, we train COSMIC on MELD [43], which contains dialogues from episodes of the TV series Friends and is more suited to our domain than other sentiment classification datasets (e.g., [9, 29]). After training, we use COSMIC to produce sentence-level sentiment predictions for the TRIPOD screenplays. The sentiment of a scene corresponds to the majority sentiment of its sentences. We project scenebased sentiment labels onto shots using the same one-tomany mapping employed for TPs. Sentiment Labels This paper is available on arxiv under CC BY-SA 4.0 DEED license. This paper is available on arxiv under CC BY-SA 4.0 DEED license. available on arxiv [3] https://github.com/ppapalampidi/TRIPOD [4] https://datashare.ed.ac.uk/handle/10283/3819 [5] http://www.cs.virginia.edu/ pc9za/research/moviescope.html [6] https://github.com/Breakthrough/PySceneDetect

Film Trailer Generation via Task Decomposition: Experimental Setup

টাস্ক ডিকম্পোজিশনের মাধ্যমে ফিল্ম ট্রেলার জেনারেশন: পরীক্ষামূলক সেটআপ

Filmtrailer-Generierung mittels Task Decomposition: Versuchsaufbau

Generación de avances de películas mediante descomposición de tareas: configuración experimental

Génération de bandes-annonces de films via la décomposition de tâches : configuration expérimentale

टास्क डिकंपोज़िशन के माध्यम से फ़िल्म ट्रेलर निर्माण: प्रायोगिक सेटअप

タスク分解による映画予告編生成: 実験セットアップ

작업 분해를 통한 영화 예고편 생성: 실험 설정

Geração de trailer de filme por meio de decomposição de tarefas: configuração experimental

Создание трейлера к фильму посредством декомпозиции задач: экспериментальная установка

Görev Ayrıştırma Yoluyla Film Fragmanı Oluşturma: Deneysel Kurulum

Tạo đoạn giới thiệu phim thông qua phân tách tác vụ: Thiết lập thử nghiệm

通过任务分解生成电影预告片：实验设置

Violence Detection in Videos: Experiments and Results

ভিডিওতে সহিংসতা সনাক্তকরণ: পরীক্ষা এবং ফলাফল

Gewalterkennung in Videos: Experimente und Ergebnisse

Detección de violencia en videos: experimentos y resultados

Détection de la violence dans les vidéos : expériences et résultats

वीडियो में हिंसा का पता लगाना: प्रयोग और परिणाम

動画における暴力検出: 実験と結果

동영상의 폭력 감지: 실험 및 결과

Detecção de violência em vídeos: experimentos e resultados

Обнаружение насилия в видео: эксперименты и результаты

Videolarda Şiddet Tespiti: Deneyler ve Sonuçlar

Phát hiện bạo lực trong video: Thử nghiệm và kết quả

视频中的暴力检测：实验和结果

Violence Detection in Videos: Proposed Approach

ভিডিওতে সহিংসতা সনাক্তকরণ: প্রস্তাবিত পদ্ধতি

Gewalterkennung in Videos: Vorgeschlagener Ansatz

Detección de violencia en videos: enfoque propuesto

Détection de la violence dans les vidéos : approche proposée

वीडियो में हिंसा का पता लगाना: प्रस्तावित दृष्टिकोण

動画内の暴力検出: 提案されたアプローチ

비디오의 폭력 감지: 제안된 접근 방식

Detecção de violência em vídeos: abordagem proposta

Обнаружение насилия в видео: предлагаемый подход

Videolarda Şiddet Tespiti: Önerilen Yaklaşım

Phát hiện bạo lực trong video: Phương pháp đề xuất

视频中的暴力检测：建议的方法

Authors:
(1) Praveen Tirupattur,  University of Central Florida. Table of Links Abstract
Acknowledgements
Chapter 1: Introduction
Chapter 2: Related Work
Chapter 3: Proposed Approach
Chapter 4: Experiments and Results
Chapter 5: Conclusions and Future Work
Bibliography 1. Introduction The amount of multimedia content uploaded to social networking websites and the ease with which these can be accessed by children is posing a problem to parents who wish to protect their children from getting exposed to violent and adult content on the web. The number of video uploads to websites like YouTube and Facebook are on the rise. There is an increase of 75% in the number of video posts on Facebook (Blog-FB [3]) in the last one year and more than 120,000 videos are uploaded to YouTube every day (Wesch [56], Gill et al. [26]). It is estimated that 20% of the videos uploaded to these websites contain violent or adult content (Sparks [54]). This makes it easy for children to access or accidentally get exposed to these unsafe contents. The effects of watching violent content on children are well studied in psychology (Tompkins [55], Sparks [54], Bushman and Huesmann [6], and Huesmann and Taylor [32]) and the results of these studies suggest that watching of violent content has a substantial effect on emotions of the children. The major effects are increases in the likelihood of aggressive or fearful behavior and becoming less sensitive to the pain and suffering of others. Huesmann and Eron [31] conducted a study involving children from elementary school, who watched many hours of violence on television. By observing these children into adulthood, they found that the ones who did watch a lot of television violence when they were 8 years old were more likely to be arrested and prosecuted for criminal acts as adults. Similar studies by Flood [25] and Mitchell et al. [40] suggest that exposure to adult content also has detrimental effects on children. This motivated research in the field of automatic violent and adult content detection in videos. Adult content detection (Chan et al. [8], Schulze et al. [52], Pogrebnyak et al. [47]) is well studied and much progress has been made. Violence detection, on the other hand, has been less studied and has gained interest only in the recent past. Few approaches for violence detection were proposed in the past and each of these approaches tried to detect violence using different visual and auditory features. For example, Nam et al. [41] combined multiple audio-visual features to identify violent scenes. In their work, flames and blood were detected using predefined color tables and various representative audio effects (gunshots, explosions, etc.) were also exploited. Datta et al. [14] proposed an accelerated motion vector based approach to detect human violence such as fist fighting, kicking, etc. Cheng et al. [11] presented a hierarchical approach to locating gun play and car racing scenes through detection of typical audio events (e.g. gunshots, explosions, and car-braking). More approaches proposed for violence detection are discussed in Chapter 2. All of these approaches focused mainly only on detection of violence in Hollywood movies but not in videos from video sharing and social media websites such as YouTube or Facebook. Detection of violence in Hollywood movies is relatively easy as these movies follow some moviemaking rules. For example, to exhibit exciting action scenes, the atmosphere of fast-pace is created through high-speed visual movement and fast-paced sound. But the videos from the video-sharing websites, like YouTube and Facebook, do not follow these moviemaking rules and often have poor audio and video quality. These characteristics of user-generated videos make it very hard to detect violence in them. Before the approach to detect violence is discussed, it is important to provide a definition for the term “Violence”. All of the previous approaches for violence detection have not followed the same definition of violence and have used different features and different datasets. This makes the comparison of different approaches very difficult. To overcome this problem and to foster research in this area, a dataset named Violent Scene Detection (VSD) was introduced by Demarty et al. [15] in 2011 and the recent version of this dataset is the VSD2014. According to this latest dataset, “Violence” in a video is, “any scene one would not let an 8 year old child watch because they contain physical violence”Schedl et al. [51]. This definition is believed to be formulated based on the research findings from psychology, which are mentioned above. From this definition, it can be observed that violence is not a physical entity but a concept which is very generic, abstract and also very subjective. Hence, violence detection is not a trivial task. The aim of this work is to build a system which automatically detects violence not only in Hollywood movies, but also in videos from the video-sharing websites like YouTube and Facebook. In this work, an attempt is made to also detect the category of violence in a video, which was not addressed by earlier approaches. The categories of violence which are targeted in this work are the presence of blood, presence of cold arms, explosions, fights, screams, presence of fire, firearms, and gunshots. These represent the subset of concepts defined and used in the VSD2014 for annotating video segments. The categories “gory scenes” and “car chase” from VSD2014 were not selected as there were not many video segments in VSD2014 annotated with these concepts. Another such category is the “Subjective Violence”. It is not selected as the scenes belonging to this category do not have any visible violence and hence are very hard to detect. In this work, both audio and visual features are used for violence detection as combining both audio and visual information provides more reliable results in classification. The advantages of developing a system like this, which can automatically detect violence in multi-media content are many. It can be used to rate movies depending on the amount of violence. This can be used by social networking sites to detect and block upload of violent videos to their platforms. Also, it can be used for scene characterization and genre classification which helps in searching and browsing movies. Recognition of violence in video streams from real-time camera systems will be very helpful for video surveillance in places such as airports, hospitals, shopping malls, public places, prisons, psychiatric wards, school playgrounds etc. However, real time detection of violence is much more difficult and in this work no attempt is made to deal with it. An overview of related work, detailed description of the proposed approach and the evaluation are presented next. The following chapters are organized as follows. In Chapter 2 some of the previous works in the area of violence detection are explained in detail. In Chapter 3, the details of the approach used for training and testing of feature classifiers are presented. It also includes the details of feature extraction and the classifier training. Chapter 4 describes the details of datasets used, experimental setup and the results obtained from the experiments. Finally, in Chapter 5 conclusions are provided followed by the possible future work. This paper is available on arxiv under CC 4.0 license. Authors: (1) Praveen Tirupattur,  University of Central Florida. Authors: Authors: (1) Praveen Tirupattur,  University of Central Florida. Table of Links Abstract Acknowledgements Chapter 1: Introduction Chapter 2: Related Work Chapter 3: Proposed Approach Chapter 4: Experiments and Results Chapter 5: Conclusions and Future Work Bibliography Abstract Abstract Acknowledgements Acknowledgements Chapter 1: Introduction Chapter 1: Introduction Chapter 2: Related Work Chapter 2: Related Work Chapter 3: Proposed Approach Chapter 3: Proposed Approach Chapter 4: Experiments and Results Chapter 4: Experiments and Results Chapter 5: Conclusions and Future Work Chapter 5: Conclusions and Future Work Bibliography Bibliography 1. Introduction The amount of multimedia content uploaded to social networking websites and the ease with which these can be accessed by children is posing a problem to parents who wish to protect their children from getting exposed to violent and adult content on the web. The number of video uploads to websites like YouTube and Facebook are on the rise. There is an increase of 75% in the number of video posts on Facebook (Blog-FB [3]) in the last one year and more than 120,000 videos are uploaded to YouTube every day (Wesch [56], Gill et al. [26]). It is estimated that 20% of the videos uploaded to these websites contain violent or adult content (Sparks [54]). This makes it easy for children to access or accidentally get exposed to these unsafe contents. The effects of watching violent content on children are well studied in psychology (Tompkins [55], Sparks [54], Bushman and Huesmann [6], and Huesmann and Taylor [32]) and the results of these studies suggest that watching of violent content has a substantial effect on emotions of the children. The major effects are increases in the likelihood of aggressive or fearful behavior and becoming less sensitive to the pain and suffering of others. Huesmann and Eron [31] conducted a study involving children from elementary school, who watched many hours of violence on television. By observing these children into adulthood, they found that the ones who did watch a lot of television violence when they were 8 years old were more likely to be arrested and prosecuted for criminal acts as adults. Similar studies by Flood [25] and Mitchell et al. [40] suggest that exposure to adult content also has detrimental effects on children. This motivated research in the field of automatic violent and adult content detection in videos. Adult content detection (Chan et al. [8], Schulze et al. [52], Pogrebnyak et al. [47]) is well studied and much progress has been made. Violence detection, on the other hand, has been less studied and has gained interest only in the recent past. Few approaches for violence detection were proposed in the past and each of these approaches tried to detect violence using different visual and auditory features. For example, Nam et al. [41] combined multiple audio-visual features to identify violent scenes. In their work, flames and blood were detected using predefined color tables and various representative audio effects (gunshots, explosions, etc.) were also exploited. Datta et al. [14] proposed an accelerated motion vector based approach to detect human violence such as fist fighting, kicking, etc. Cheng et al. [11] presented a hierarchical approach to locating gun play and car racing scenes through detection of typical audio events (e.g. gunshots, explosions, and car-braking). More approaches proposed for violence detection are discussed in Chapter 2. All of these approaches focused mainly only on detection of violence in Hollywood movies but not in videos from video sharing and social media websites such as YouTube or Facebook. Detection of violence in Hollywood movies is relatively easy as these movies follow some moviemaking rules. For example, to exhibit exciting action scenes, the atmosphere of fast-pace is created through high-speed visual movement and fast-paced sound. But the videos from the video-sharing websites, like YouTube and Facebook, do not follow these moviemaking rules and often have poor audio and video quality. These characteristics of user-generated videos make it very hard to detect violence in them. Before the approach to detect violence is discussed, it is important to provide a definition for the term “Violence”. All of the previous approaches for violence detection have not followed the same definition of violence and have used different features and different datasets. This makes the comparison of different approaches very difficult. To overcome this problem and to foster research in this area, a dataset named Violent Scene Detection (VSD) was introduced by Demarty et al. [15] in 2011 and the recent version of this dataset is the VSD2014. According to this latest dataset, “Violence” in a video is, “any scene one would not let an 8 year old child watch because they contain physical violence”Schedl et al. [51]. This definition is believed to be formulated based on the research findings from psychology, which are mentioned above. From this definition, it can be observed that violence is not a physical entity but a concept which is very generic, abstract and also very subjective. Hence, violence detection is not a trivial task. The aim of this work is to build a system which automatically detects violence not only in Hollywood movies, but also in videos from the video-sharing websites like YouTube and Facebook. In this work, an attempt is made to also detect the category of violence in a video, which was not addressed by earlier approaches. The categories of violence which are targeted in this work are the presence of blood, presence of cold arms, explosions, fights, screams, presence of fire, firearms, and gunshots. These represent the subset of concepts defined and used in the VSD2014 for annotating video segments. The categories “gory scenes” and “car chase” from VSD2014 were not selected as there were not many video segments in VSD2014 annotated with these concepts. Another such category is the “Subjective Violence”. It is not selected as the scenes belonging to this category do not have any visible violence and hence are very hard to detect. In this work, both audio and visual features are used for violence detection as combining both audio and visual information provides more reliable results in classification. The advantages of developing a system like this, which can automatically detect violence in multi-media content are many. It can be used to rate movies depending on the amount of violence. This can be used by social networking sites to detect and block upload of violent videos to their platforms. Also, it can be used for scene characterization and genre classification which helps in searching and browsing movies. Recognition of violence in video streams from real-time camera systems will be very helpful for video surveillance in places such as airports, hospitals, shopping malls, public places, prisons, psychiatric wards, school playgrounds etc. However, real time detection of violence is much more difficult and in this work no attempt is made to deal with it. An overview of related work, detailed description of the proposed approach and the evaluation are presented next. The following chapters are organized as follows. In Chapter 2 some of the previous works in the area of violence detection are explained in detail. In Chapter 3, the details of the approach used for training and testing of feature classifiers are presented. It also includes the details of feature extraction and the classifier training. Chapter 4 describes the details of datasets used, experimental setup and the results obtained from the experiments. Finally, in Chapter 5 conclusions are provided followed by the possible future work. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Violence Detection in Videos: Introduction

ভিডিওতে সহিংসতা সনাক্তকরণ: ভূমিকা

Gewalterkennung in Videos: Einführung

Detección de violencia en vídeos: Introducción

Détection de la violence dans les vidéos : introduction

वीडियो में हिंसा का पता लगाना: परिचय

動画における暴力検出: 概要

동영상의 폭력 감지: 소개

Detecção de violência em vídeos: introdução

Обнаружение насилия в видео: Введение

Videolarda Şiddet Tespiti: Giriş

Phát hiện bạo lực trong video: Giới thiệu

视频中的暴力检测：简介

Violence Detection in Videos: Related Work 

ভিডিওতে সহিংসতা সনাক্তকরণ: সম্পর্কিত কাজ 

Gewalterkennung in Videos: Verwandte Arbeiten

Detección de Violencia en Videos: Trabajo Relacionado 

Détection de la violence dans les vidéos : travaux connexes 

वीडियो में हिंसा का पता लगाना: संबंधित कार्य

動画における暴力検出: 関連研究

영상 속 폭력 탐지: 관련 연구 

Detecção de violência em vídeos: trabalhos relacionados 

Обнаружение насилия в видео: соответствующая работа 

Videolarda Şiddet Tespiti: İlgili Çalışmalar 

Phát hiện bạo lực trong video: Công việc liên quan 

视频中的暴力检测：相关工作

This paper is available on Arxiv under CC 4.0 license.
Authors:
(1) Falke B. Ø. Carlsen, Department of Computer Science, Aalborg University, Denmark & falkeboc@cs.aau.dk;
(2) Lars Bo P. Frydenskov, Department of Computer Science, Aalborg University, Denmark & larsbopark@gmail.com;
(3) Nicolaj Ø. Jensen, Department of Computer Science, Aalborg University, Denmark & noje@cs.aau.dk;
(4) Jener Rasmussen, jener@jener.dk;
(5) Mathias M. Sørensen, mathiasmehlsoerensen@gmail.com;
(6) Asger G. Weirsøe, asger@weircon.dk;
(7) Mathias C. Jensen, Department of Computer Science, Aalborg University, Denmark & mcje@cs.aau.dk;
(8) Kim G. Larsen, Department of Computer Science, Aalborg University, Denmark & kgl@cs.aau.dk. Table of Links Introduction Definitions Model Checking Tool Overview Evaluation Conclusion and References We present CGAAL, our efficient on-the-fly model checker for alternating-time temporal logic (ATL) on concurrent game structures (CGS). We present how our tool encodes ATL as extended dependency graphs with negation edges and employs the distributed on-the-fly algorithm by Dalsgaard et al. Our tool offers multiple novel search strategies for the algorithm, including DHS which is inspired by PageRank and uses the in-degree of configurations as a heuristic, IHS which estimates instability of assignment values, and LPS which estimates the distance to a state satisfying the constituent property using linear programming. CGS are input using our modelling language LCGS, where composition and synchronisation are easily described. We prove the correctness of our encoding, and our experiments show that our tool CGAAL is often one to three orders of magnitude faster than the popular tool PRISM-games on case studies from PRISM’s documentation and among case studies we have developed. In our evaluation, we also compare and evaluate our search strategies, and find that our custom search strategies are often significantly faster than the usual breadth-first and depth-first search strategies. 1 Introduction Software plays a large role in our everyday lives, making decisions, enabling efficient communication, ensuring safety, and many more critical tasks. Furthermore, the complexity of the software and the decisions they have to make are ever-increasing due the to interconnectivity and reactive nature of modern systems. These software systems mutually depend on, communicate with, and guide each other based on their collective and internal states. Even a few interconnected systems that are not necessarily themselves too complex can give rise to an extremely complex system as a whole. Safety-critical software that operates in contexts where errors could lead to human casualties or significant capital losses requires methods to verify their correctness. Unfortunately, such methods are difficult to implement due to the sheer state-space explosion that arises from the inherent parallel composition of systems. In this paper, we consider systems that can be expressed as discrete multiplayer games in which the actors can perform actions concurrently. That is, in each configuration of the game, each player simultaneously chooses an action they wish to perform and the resulting decision vector then deterministically results in the next configuration of the game. Such concurrent games are extremely expressive and can be used to model a variety of complex systems and lends themselves inherently, by their concurrent nature, to describing multi-actor systems. The properties that we wish to verify on these systems are those that can be expressed by Alternating-time Temporal Logic (ATL) [1], a game theoretic extension of Computation Tree Logic (CTL) [2, 3], that like CTL can be used to describe safety and liveness properties. ATL extends CTL by replacing the existential and universal path quantifiers with so-called coalition quantifiers. These coalition quantifiers describe properties of the game in which a coalition of players can either enforce some desired outcome or cannot avoid an outcome. Of course, proof of satisfaction of such property is done by finding a strategy showing that the players can indeed enforce said property or by showing that no matter what strategy the players follow they cannot avoid said property. One existing tool that can model check concurrent (stochastic) games is PRISM-games [12], an extension of the probabilistic model checker PRISM [11]. PRISM excels at stochastic models such as discreteand continuous-time Markov chains, Markov decision processes, and (priced) probabilistic timed automata, for which it can verify various properties described in LTL , CSL, and probabilistic CTL*. There are multiple verification engines in PRISM, both symbolic and explicit state. The PRISM-games extension focuses on stochastic games in which game-theoretic approaches are needed. PRISM-games can among other things synthesise strategies and reason about equilibria-based properties, where players may have distinct, but not necessarily conflicting objectives. These properties are typically described with probabilistic ATL with rewards. In version 3.0 [12], PRISM-games added support for concurrent stochastic games. In [14], Liu and Smolka present a global and a local algorithm to compute fixed-point boolean vertex assignments in dependency graphs with directed hyper-edges representing dependencies between vertex values. The global algorithm has a better worst-case running time of the two, but the on-the-fly local algorithm only explores the graph as needed to compute the fixed-point assignment. Many problems have since been encoded as dependency graphs. Notably, model-checking problems for CTL can be encoded as dependency graphs and the algorithms by Liu and Smolka can be used to compute the satisfaction relation enabling model-checking using dependency graphs. J. F. Jensen et al. [9] show that weighted CTL can be encoded in dependency graphs as well and introduce symbolic edges to handle weights efficiently. In [10], M. C. Jensen et al. introduce symbolic dependency graphs, an encoding of probabilistic weighted CTL, and an implementation of a local and global algorithm. A. Dalsgaard et al. [4] present a distributed extension to the local algorithm to boost the performance and show how to model Petri Nets problems using CTL and dependency graphs. Furthermore, they extend dependency graphs by adding components and negation edges to solve formulae with negations. An abstract dependency graph framework is presented by S. Enevoldsen et al. in [6] where they generalise the various extensions of the dependency graphs with a general algorithm for any Noetherian partial order domain. In the closely related work of [5], S. Enevoldsen et at. uses abstract dependency graphs to model check probabilistic ATL for weighted stochastic games. In multiple works [4, 10] the local algorithm is found to be faster more often than the global algorithm in practice. In this paper, we introduce the tool CGAAL (Concurrent Games AALborg), a model checker implemented in Rust that allows for the verification of ATL properties on concurrent games. CGAAL works by encoding the model checking problem as an extended dependency graph and by implementing the on-the-fly local distributed algorithm by Dalsgaard et al. [4]. The concurrent games are described and inputted to the tool using LCGS, our PRISM-language inspired model language, and for specific ATL formulae, we can output a strategy describing how to ensure satisfaction given a positive result. Furthermore, CGAAL implements a variety of strategies for searching the problem state space: a heuristic inspired by PageRank [15] and two heuristic search strategies that exploit the state vector representation of our model, alongside the usual breadth-first and depth-first approach. We conduct experiments comparing our different search strategies with each other on several case studies and show that having more available compute threads often leads to speed-ups of 1-2 orders of magnitude. We also perform experiments comparing our tool on a selection of case studies to the state-of-the-art PRISM-games. Though while PRISM-games is a tool specialised for probabilistic games we find that our implementation often outperforms PRISM-games and especially whenever we are not required to compute the entire fixed point in which case we are often 1-3 orders of magnitude faster. Again, we emphasise that PRISM games is meant for probabilistic games and as such this is not a completely fair one-to-one comparison. Outline. This paper is structured as follows. Section 2 introduces the formal definitions of concurrent games and alternating-time temporal logic. In Section 3 we present how model checking of ATL properties in concurrent games can be encoded in extended dependency graphs. We also give a short explanation of the CERTAINZERO algorithm and our search strategies. In Section 4 we give a short example of how to use CGAAL and the LCGS language. An evaluation of our tool and a comparison with PRISM can be found in Section 5, and we conclude on our findings in Section 6. This paper is available on Arxiv under CC 4.0 license. This paper is available on Arxiv under CC 4.0 license. Authors: (1) Falke B. Ø. Carlsen, Department of Computer Science, Aalborg University, Denmark & falkeboc@cs.aau.dk; (2) Lars Bo P. Frydenskov, Department of Computer Science, Aalborg University, Denmark & larsbopark@gmail.com; (3) Nicolaj Ø. Jensen, Department of Computer Science, Aalborg University, Denmark & noje@cs.aau.dk; (4) Jener Rasmussen, jener@jener.dk; (5) Mathias M. Sørensen, mathiasmehlsoerensen@gmail.com; (6) Asger G. Weirsøe, asger@weircon.dk; (7) Mathias C. Jensen, Department of Computer Science, Aalborg University, Denmark & mcje@cs.aau.dk; (8) Kim G. Larsen, Department of Computer Science, Aalborg University, Denmark & kgl@cs.aau.dk. This paper is available on Arxiv under CC 4.0 license. Authors: Authors: (1) Falke B. Ø. Carlsen, Department of Computer Science, Aalborg University, Denmark & falkeboc@cs.aau.dk; (2) Lars Bo P. Frydenskov, Department of Computer Science, Aalborg University, Denmark & larsbopark@gmail.com; (3) Nicolaj Ø. Jensen, Department of Computer Science, Aalborg University, Denmark & noje@cs.aau.dk; (4) Jener Rasmussen, jener@jener.dk; (5) Mathias M. Sørensen, mathiasmehlsoerensen@gmail.com; (6) Asger G. Weirsøe, asger@weircon.dk; (7) Mathias C. Jensen, Department of Computer Science, Aalborg University, Denmark & mcje@cs.aau.dk; (8) Kim G. Larsen, Department of Computer Science, Aalborg University, Denmark & kgl@cs.aau.dk. Table of Links Introduction Introduction Definitions Definitions Model Checking Model Checking Tool Overview Tool Overview Evaluation Evaluation Conclusion and References Conclusion and References We present CGAAL, our efficient on-the-fly model checker for alternating-time temporal logic (ATL) on concurrent game structures (CGS). We present how our tool encodes ATL as extended dependency graphs with negation edges and employs the distributed on-the-fly algorithm by Dalsgaard et al. Our tool offers multiple novel search strategies for the algorithm, including DHS which is inspired by PageRank and uses the in-degree of configurations as a heuristic, IHS which estimates instability of assignment values, and LPS which estimates the distance to a state satisfying the constituent property using linear programming. CGS are input using our modelling language LCGS, where composition and synchronisation are easily described. We prove the correctness of our encoding, and our experiments show that our tool CGAAL is often one to three orders of magnitude faster than the popular tool PRISM-games on case studies from PRISM’s documentation and among case studies we have developed. In our evaluation, we also compare and evaluate our search strategies, and find that our custom search strategies are often significantly faster than the usual breadth-first and depth-first search strategies. 1 Introduction Software plays a large role in our everyday lives, making decisions, enabling efficient communication, ensuring safety, and many more critical tasks. Furthermore, the complexity of the software and the decisions they have to make are ever-increasing due the to interconnectivity and reactive nature of modern systems. These software systems mutually depend on, communicate with, and guide each other based on their collective and internal states. Even a few interconnected systems that are not necessarily themselves too complex can give rise to an extremely complex system as a whole. Safety-critical software that operates in contexts where errors could lead to human casualties or significant capital losses requires methods to verify their correctness. Unfortunately, such methods are difficult to implement due to the sheer state-space explosion that arises from the inherent parallel composition of systems. In this paper, we consider systems that can be expressed as discrete multiplayer games in which the actors can perform actions concurrently. That is, in each configuration of the game, each player simultaneously chooses an action they wish to perform and the resulting decision vector then deterministically results in the next configuration of the game. Such concurrent games are extremely expressive and can be used to model a variety of complex systems and lends themselves inherently, by their concurrent nature, to describing multi-actor systems. The properties that we wish to verify on these systems are those that can be expressed by Alternating-time Temporal Logic (ATL) [1], a game theoretic extension of Computation Tree Logic (CTL) [2, 3], that like CTL can be used to describe safety and liveness properties. ATL extends CTL by replacing the existential and universal path quantifiers with so-called coalition quantifiers. These coalition quantifiers describe properties of the game in which a coalition of players can either enforce some desired outcome or cannot avoid an outcome. Of course, proof of satisfaction of such property is done by finding a strategy showing that the players can indeed enforce said property or by showing that no matter what strategy the players follow they cannot avoid said property. One existing tool that can model check concurrent (stochastic) games is PRISM-games [12], an extension of the probabilistic model checker PRISM [11]. PRISM excels at stochastic models such as discreteand continuous-time Markov chains, Markov decision processes, and (priced) probabilistic timed automata, for which it can verify various properties described in LTL , CSL, and probabilistic CTL*. There are multiple verification engines in PRISM, both symbolic and explicit state. The PRISM-games extension focuses on stochastic games in which game-theoretic approaches are needed. PRISM-games can among other things synthesise strategies and reason about equilibria-based properties, where players may have distinct, but not necessarily conflicting objectives. These properties are typically described with probabilistic ATL with rewards. In version 3.0 [12], PRISM-games added support for concurrent stochastic games. In [14], Liu and Smolka present a global and a local algorithm to compute fixed-point boolean vertex assignments in dependency graphs with directed hyper-edges representing dependencies between vertex values. The global algorithm has a better worst-case running time of the two, but the on-the-fly local algorithm only explores the graph as needed to compute the fixed-point assignment. Many problems have since been encoded as dependency graphs. Notably, model-checking problems for CTL can be encoded as dependency graphs and the algorithms by Liu and Smolka can be used to compute the satisfaction relation enabling model-checking using dependency graphs. J. F. Jensen et al. [9] show that weighted CTL can be encoded in dependency graphs as well and introduce symbolic edges to handle weights efficiently. In [10], M. C. Jensen et al. introduce symbolic dependency graphs, an encoding of probabilistic weighted CTL, and an implementation of a local and global algorithm. A. Dalsgaard et al. [4] present a distributed extension to the local algorithm to boost the performance and show how to model Petri Nets problems using CTL and dependency graphs. Furthermore, they extend dependency graphs by adding components and negation edges to solve formulae with negations. An abstract dependency graph framework is presented by S. Enevoldsen et al. in [6] where they generalise the various extensions of the dependency graphs with a general algorithm for any Noetherian partial order domain. In the closely related work of [5], S. Enevoldsen et at. uses abstract dependency graphs to model check probabilistic ATL for weighted stochastic games. In multiple works [4, 10] the local algorithm is found to be faster more often than the global algorithm in practice. In this paper, we introduce the tool CGAAL (Concurrent Games AALborg), a model checker implemented in Rust that allows for the verification of ATL properties on concurrent games. CGAAL works by encoding the model checking problem as an extended dependency graph and by implementing the on-the-fly local distributed algorithm by Dalsgaard et al. [4]. The concurrent games are described and inputted to the tool using LCGS, our PRISM-language inspired model language, and for specific ATL formulae, we can output a strategy describing how to ensure satisfaction given a positive result. Furthermore, CGAAL implements a variety of strategies for searching the problem state space: a heuristic inspired by PageRank [15] and two heuristic search strategies that exploit the state vector representation of our model, alongside the usual breadth-first and depth-first approach. We conduct experiments comparing our different search strategies with each other on several case studies and show that having more available compute threads often leads to speed-ups of 1-2 orders of magnitude. We also perform experiments comparing our tool on a selection of case studies to the state-of-the-art PRISM-games. Though while PRISM-games is a tool specialised for probabilistic games we find that our implementation often outperforms PRISM-games and especially whenever we are not required to compute the entire fixed point in which case we are often 1-3 orders of magnitude faster. Again, we emphasise that PRISM games is meant for probabilistic games and as such this is not a completely fair one-to-one comparison. Outline . This paper is structured as follows. Section 2 introduces the formal definitions of concurrent games and alternating-time temporal logic. In Section 3 we present how model checking of ATL properties in concurrent games can be encoded in extended dependency graphs. We also give a short explanation of the CERTAINZERO algorithm and our search strategies. Outline In Section 4 we give a short example of how to use CGAAL and the LCGS language. An evaluation of our tool and a comparison with PRISM can be found in Section 5, and we conclude on our findings in Section 6. This paper is available on Arxiv under CC 4.0 license. This paper is available on Arxiv under CC 4.0 license. Arxiv

Introducing CGAAL: A Distributed On-The-Fly ATL Model Checker With Heuristics

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Andrey Zhmoginov, Google Research & {azhmogin,sandler,mxv}@google.com;
(2) Mark Sandler, Google Research & {azhmogin,sandler,mxv}@google.com;
(3) Max Vladymyrov, Google Research & {azhmogin,sandler,mxv}@google.com. Table of Links Abstract and Introduction
Problem Setup and Related Work
HyperTransformer
Experiments
Conclusion and References
A Example of a Self-Attention Mechanism For Supervised Learning
B Model Parameters
C Additional Supervised Experiments
D Dependence On Parameters and Ablation Studies
E Attention Maps of Learned Transformer Models
F Visualization of The Generated CNN Weights
G Additional Tables and Figures Problem Setup and Related Work 2.1 FEW-SHOT LEARNING 2.2 RELATED WORK Few-shot learning received a lot of attention from the deep learning community and while there are hundreds of few-shot learning methods, several common themes emerged in the past years. Here we outline several existing approaches, show how they relate to our method and discuss the prior work related to it. Metric-Based Learning. One family of approaches involves mapping input samples into an embedding space and then using some nearest neighbor algorithm that relies on the computation of distances from a query sample embedding to the embedding computed using support samples with known labels. The metric used to compute the distance can either be the same for all tasks, or can be task-dependent. This family of methods includes, for example, such methods as Siamese networks (Koch et al., 2015), Matching Networks (Vinyals et al., 2016), Prototypical Networks (Snell et al., 2017), Relation Networks (Sung et al., 2018) and TADAM (Oreshkin et al., 2018). It has recently been argued (Tian et al., 2020) that methods based on building a powerful sample representation can frequently outperform numerous other approaches including many optimization-based methods. However, such approaches essentially amount to the “one-model solves all” approach and thus require larger models than needed to solve individual tasks. Weight Modulation and Generation. The idea of using a task specification to directly generate or modulate model weights has been previously explored in the generalized supervised learning context (Ratzlaff & Li, 2019) and in specific language models (Mahabadi et al., 2021; Tay et al., 2021; Ye & Ren, 2021). Some few-shot learning methods described above also employ this approach and use task-specific generation or modulation of the weights of the final classification model. For example, in LGM-Net (Li et al., 2019b) the matching network approach is used to generate a few layers on top of a task-agnostic embedding. Another approach abbreviated as LEO (Rusu et al., 2019) utilized a similar weight generation method to generate initial model weights from the training dataset in a few-shot learning setting, much like what is proposed in this article. However, in Rusu et al. (2019), the generated weights were also refined using several SGD steps similar to how it is done in MAML. Here we explore a similar idea, but largely inspired by the HYPERNETWORK approach (Ha et al., 2017), we instead propose to directly generate an entire task-specific CNN model. Unlike LEO, we do not rely on pre-computed embeddings for images and generate the model in a single step without additional SGD steps, which simplifies and stabilizes training. Transformers in Computer Vision and Few-Shot Learning. Transformer models (Vaswani et al., 2017) originally proposed for natural language understanding applications, had since become a useful tool in practically every subfield of deep learning. In computer vision, transformers have recently seen an explosion of applications ranging from state-of-the-art image classification results (Dosovitskiy et al., 2021; Touvron et al., 2021) to object detection (Carion et al., 2020; Zhu et al., 2021), segmentation (Ye et al., 2019), image super-resolution (Yang et al., 2020), image generation (Chen et al., 2021) and many others. There are also several notable applications in few-shot image classification. For example, in Liu et al. (2021), the transformer model was used for generating universal representations in the multi-domain few-shot learning scenario. And closely related to our approach, in Ye et al. (2020), the authors proposed to accomplish embedding adaptation with the help of transformer models. Unlike our method that generates an entire end to-end image classification model, this approach applied a task-dependent perturbation to an embedding generated by an independent task-agnostic feature extractor. In Gidaris & Komodakis (2018), a simplified attention-based model was used for the final layer generation. This paper is available on arxiv under CC 4.0 license. Authors: (1) Andrey Zhmoginov, Google Research & {azhmogin,sandler,mxv}@google.com; (2) Mark Sandler, Google Research & {azhmogin,sandler,mxv}@google.com; (3) Max Vladymyrov, Google Research & {azhmogin,sandler,mxv}@google.com. This paper is available on arxiv under CC 4.0 license. Authors: (1) Andrey Zhmoginov, Google Research & {azhmogin,sandler,mxv}@google.com; (2) Mark Sandler, Google Research & {azhmogin,sandler,mxv}@google.com; (3) Max Vladymyrov, Google Research & {azhmogin,sandler,mxv}@google.com. Table of Links Abstract and Introduction Problem Setup and Related Work HyperTransformer Experiments Conclusion and References A Example of a Self-Attention Mechanism For Supervised Learning B Model Parameters C Additional Supervised Experiments D Dependence On Parameters and Ablation Studies E Attention Maps of Learned Transformer Models F Visualization of The Generated CNN Weights G Additional Tables and Figures Abstract and Introduction Abstract and Introduction Problem Setup and Related Work Problem Setup and Related Work HyperTransformer HyperTransformer Experiments Experiments Conclusion and References Conclusion and References A Example of a Self-Attention Mechanism For Supervised Learning A Example of a Self-Attention Mechanism For Supervised Learning B Model Parameters B Model Parameters C Additional Supervised Experiments C Additional Supervised Experiments D Dependence On Parameters and Ablation Studies D Dependence On Parameters and Ablation Studies E Attention Maps of Learned Transformer Models E Attention Maps of Learned Transformer Models F Visualization of The Generated CNN Weights F Visualization of The Generated CNN Weights G Additional Tables and Figures G Additional Tables and Figures Problem Setup and Related Work 2.1 FEW-SHOT LEARNING 2.2 RELATED WORK Few-shot learning received a lot of attention from the deep learning community and while there are hundreds of few-shot learning methods, several common themes emerged in the past years. Here we outline several existing approaches, show how they relate to our method and discuss the prior work related to it. Metric-Based Learning. One family of approaches involves mapping input samples into an embedding space and then using some nearest neighbor algorithm that relies on the computation of distances from a query sample embedding to the embedding computed using support samples with known labels. The metric used to compute the distance can either be the same for all tasks, or can be task-dependent. This family of methods includes, for example, such methods as Siamese networks (Koch et al., 2015), Matching Networks (Vinyals et al., 2016), Prototypical Networks (Snell et al., 2017), Relation Networks (Sung et al., 2018) and TADAM (Oreshkin et al., 2018). It has recently been argued (Tian et al., 2020) that methods based on building a powerful sample representation can frequently outperform numerous other approaches including many optimization-based methods. However, such approaches essentially amount to the “one-model solves all” approach and thus require larger models than needed to solve individual tasks. Metric-Based Learning. Weight Modulation and Generation. The idea of using a task specification to directly generate or Weight Modulation and Generation. modulate model weights has been previously explored in the generalized supervised learning context (Ratzlaff & Li, 2019) and in specific language models (Mahabadi et al., 2021; Tay et al., 2021; Ye & Ren, 2021). Some few-shot learning methods described above also employ this approach and use task-specific generation or modulation of the weights of the final classification model. For example, in LGM-Net (Li et al., 2019b) the matching network approach is used to generate a few layers on top of a task-agnostic embedding. Another approach abbreviated as LEO (Rusu et al., 2019) utilized a similar weight generation method to generate initial model weights from the training dataset in a few-shot learning setting, much like what is proposed in this article. However, in Rusu et al. (2019), the generated weights were also refined using several SGD steps similar to how it is done in MAML. Here we explore a similar idea, but largely inspired by the HYPERNETWORK approach (Ha et al., 2017), we instead propose to directly generate an entire task-specific CNN model. Unlike LEO, we do not rely on pre-computed embeddings for images and generate the model in a single step without additional SGD steps, which simplifies and stabilizes training. Transformers in Computer Vision and Few-Shot Learning. Transformer models (Vaswani et al., 2017) originally proposed for natural language understanding applications, had since become a useful tool in practically every subfield of deep learning. In computer vision, transformers have recently seen an explosion of applications ranging from state-of-the-art image classification results (Dosovitskiy et al., 2021; Touvron et al., 2021) to object detection (Carion et al., 2020; Zhu et al., 2021), segmentation (Ye et al., 2019), image super-resolution (Yang et al., 2020), image generation (Chen et al., 2021) and many others. There are also several notable applications in few-shot image classification. For example, in Liu et al. (2021), the transformer model was used for generating universal representations in the multi-domain few-shot learning scenario. And closely related to our approach, in Ye et al. (2020), the authors proposed to accomplish embedding adaptation with the help of transformer models. Unlike our method that generates an entire end to-end image classification model, this approach applied a task-dependent perturbation to an embedding generated by an independent task-agnostic feature extractor. In Gidaris & Komodakis (2018), a simplified attention-based model was used for the final layer generation. Transformers in Computer Vision and Few-Shot Learning.

HyperTransformer: Problem Setup and Related Work

At some point, too many people will have lost too much Bitcoin that they will meet to talk about it. Later, they will opt to do something about it. There can only be a (decreasing number) of bitcoins Lost cryptocurrency is a unique beast, quite different from lost information for 2 reasons: 1. We can continue maintaining its value. Right? Is it really still valuable? 2. Other people know it is lost. They simply check the blockchain. Whereas with lost information, nobody else can tell except those who have lost that intel. Take Bitcoin as an example. We still say 'There can only be 21 million'. But if this includes even lost bitcoins, then we are counting our losses as assets! That is very strange. Imagine if you were an accountant for a major company. The company deals in all sorts of assets, but sometimes writes some physical assets off as lost either due to theft, fire outbreak, or irreparable damage. But to your shock, the manager insists that not only do you keep counting those assets, but also, count them as fully functional assets whose value is going up every year. "But they are damaged", you might say. "Yeah, but they are still our assets" "What, how?" "There is hope we can recover them exactly as they were before they were lost. And then some". It’s hard to fathom. It indeed feels like Alice in Wonderland, where cards talk, a piece of cake can make one bigger, or smaller, and a cat can disappear leaving only its grin behind. Bitcoin and most crypto are the same. The question is thus; Is real value riding on all coins? Even when some are lost? Certainly, supply and demand economics might make all the coins feel equally valuable hence we would say that if Bitcoin is at $70,000 then the market cap is $1.47 trillion. But taking the figures as per this reddit post on lost bitcoins, if 6 million are lost then only 15 million coins are existing which are backed by a market cap of $1.05 trillion. Less by $420 billion. This does deflate our dreams and begs the question --> are we kidding ourselves? The road to Zero bitcoins By some estimates, 4% of 21,000,000 bitcoins continue to be lost every year. How many shall be lost, going forward? Let's assume this 4% holds true. 4% × 21,000,000 = 840,000 bitcoins. This is very steep. Well then, in 15 years, we shall lose 840,000 * 15 = 12,600,000 Add 6 million to this batch and circa 2039, 'There can only be 2,400,000 bitcoins' When will they be finished? Now, a little hope is in order. The 4% cannot hold steady forever, so it would be wise to say it too will be a declining figure as we progress. Because as Bitcoins get more valuable, people will be more careful with their keys. It is inevitable. Taking it at face value though, we shall run out of bitcoins to trade within the next 18 years. So Circa 2042. I hope the naysayers will be back to tell us they told us so, or to plead guilty of doomism and explain why there are still many millions of Bitcoins being traded daily. However, it certainly looks like way before the year 2137 when the last satoshi will be mined, we shall hit a big wall as per the coins lost vs the coins in the available supply. When enough is enough, for lost Bitcoins Assume at the time this shocking reality hits the Bitcoin community, we have only 4 million coins tradable around the world. Assume even Michael Saylor shall have lost some of his keys/coins. What then? Well, the price of a single coin, feeling so much demand, will spike to about $10 million, and then it will dawn on everybody. Bitcoin has a $40 trillion market capitalization and not a $210 trillion market capitalization. Right now, this is an elephant 🐘 in the Bitcoin room, but it is still a baby and we are all happy to pet it and feed it with careless abandon. Soon, however, it will become a giant. A big fat frustration. At $10 trillion per coin, the fees to move even a few sats will likely be thousands of dollars because; a) the Bitcoin reward is smaller. b) miners have also lost their coins. c) losing coins to hackers will be most likely (hackers will be sniffing all over every wallet, waiting for the SSL to get some slight holes) Everybody will be losing coins. It will be wild. But before I proceed? Why do people lose keys hence coins? Well, if you are using the key every time, sometimes changing it this way and that way to maintain your security (because AI powered cameras may be watching), it so happens that you are bound to make a mistake and lose the key. Or else, be hacked (since you cannot fathom changing the key all the time). Like smart people must have said, meat brains are not good at holding random numbers or random sequences of words. Luke Dash Jnr, a seasoned cypherpunk, was hacked last year. He lost 200 bitcoins. Elon Musk may have lost his Bitcon keys back in '20. So, shit happens. To hard fork those lost coins back, or not. That is the question So here's the thing. We wake up and Bitcoin is seizing up. Crashing into its mass and threatening to take our financial dreams with it when it turns into a black hole of lost keys / burned coins. We also wake up and realize that the debt spiral is still sucking us down with it, even as we hodl Bitcoin. Climate is now a real issue and oh, AI is giving us headaches. Too many addictive cat videos. Too many coins are lost, the blocks are always full so confirmations take decades, the core software is too large, the fees are too high. So we sit down, like AA group members, and decide that continuing to preach financial freedom yet the currency is fast tending towards technological obsolescence, is like being high on alcohol and thinking that one is blissful for other reasons. We need to quit the 'lost coins are good for all' alcohol. In short, it will be time for a hard fork. The Hard Fork --> Bitcoin ver. 2.0 Innovation comes full circle. After factoring in the reality of available supply and not merely supply, we could let Satoshi Nakamoto's 2 million or so coins be our target practice and our saving grace. One shot and bull’s eye. Trillions of dollars pour down from the Bitcoin piñata. Like manna from the heavens. Right? Alas, There shall only be 4 million bitcoins and the value of lost coins, if "re-supplied" to the blockchain, will quickly deflate the value of the 4 million coins existing. Overall, lost coins would have helped the remaining coins get valuable to that point. Say $10 million per BTC. If you un-lose them by the most brute force of them all -- a hard fork -- you shock the demand economics and viola, have worth-less bitcoins than you started with. Now whoever is holding their previously un-lost bitcoins cannot be happy about that. It begs the question, what would they do? Refuse the hard fork outright. Obviously. Indeed, even if the value of the bitcoins would overall increase the market cap of the available supply, we can be damn sure it would reduce the value of each individual coin. Supply and Demand. Economics 101. The Big Question: Is this a financial revolution for the better, or for the worse? For example, let's say we had 4 million coins left. We then redeem the coins in the wallets of those proven deceased without any heir. Say Satoshi Finney is that first choice. At $10 million per coin, 4 million coins give up a market cap of 40 trillion USD. Add 2 million coins to the available supply but have the price shocked from 10 million to only 7 million. Total availability supply = 6 million. 6 million x 7 million USD = 42 trillion dollars. It is a good financial explosion. But to whom? Who gets the money?? Who makes the ultimate sacrifice of accepting to end up with less money than before? If rational and educated human beings cannot agree on simple (obvious) things like "Are fossil fuels necessary for economic progress or not?", how can they agree on how to share this pie? It is a financial explosion, except the majority will not want it. Satoshi may have solved the Byzantine General's problem. Somebody needs to solve the "Goodness of the community over the Goodness of the individual" problem. Of course, with my limited brains, I know most people are wary of a Marxist dystopia sliding in during this very crucial process. E.g. The coins could instead be coded to go to a handful of powerful individuals, whereas before it seemed like they would go to people who deserved a financial turnaround. Especially in the wake of what we see today with government debt ballooning, government operatives hoping to snatch some of this reclaimed BTC so as to help the government pay its gigantic debt is not an impossibility. Talk about being stuck between a rock and a hard place. Not the best kind of sandwich. At some point, too many people will have lost too much Bitcoin that they will meet to talk about it. Bitcoin Later, they will opt to do something about it. There can only be a (decreasing number) of bitcoins L ost cryptocurrency is a unique beast, quite different from lost information for 2 reasons: ost cryptocurrency is a unique beast, quite different from lost information for 2 reasons: 1. We can continue maintaining its value. Right? Is it really still valuable? Is it really still valuable? really 2. Other people know it is lost. They simply check the blockchain. Whereas with lost information, nobody else can tell except those who have lost that intel. Take Bitcoin as an example. We still say 'There can only be 21 million'. But if this includes even lost bitcoins, then we are counting our losses as assets! That is very strange. Imagine if you were an accountant for a major company. The company deals in all sorts of assets, but sometimes writes some physical assets off as lost either due to theft, fire outbreak, or irreparable damage. But to your shock, the manager insists that not only do you keep counting those assets, but also, count them as fully functional assets whose value is going up every year. " But they are damaged ", you might say. But they are damaged " Yeah, but they are still our assets " Yeah, but they are still our assets " What, how? " What, how? " There is hope we can recover them exactly as they were before they were lost. There is hope we can recover them exactly as they were before they were lost. And then some ". And then some It’s hard to fathom. It indeed feels like Alice in Wonderland, where cards talk, a piece of cake can make one bigger, or smaller, and a cat can disappear leaving only its grin behind. Bitcoin and most crypto are the same. The question is thus; Is real value riding on all coins? Even when some are lost? Certainly, supply and demand economics might make all the coins feel equally valuable hence we would say that if Bitcoin is at $70,000 then the market cap is $1.47 trillion. But taking the figures as per this reddit post on lost bitcoins , if 6 million are lost then only 15 million coins are existing which are backed by a market cap of $1.05 trillion. reddit post on lost bitcoins Less by $420 billion. This does deflate our dreams and begs the question --> are we kidding ourselves? The road to Zero bitcoins By some estimates , 4% of 21,000,000 bitcoins continue to be lost every year. some estimates How many shall be lost, going forward? Let's assume this 4% holds true. 4% × 21,000,000 = 840,000 bitcoins. This is very steep. Well then, in 15 years, we shall lose 840,000 * 15 = 12,600,000 Add 6 million to this batch and circa 2039, 'There can only be 2,400,000 bitcoins' When will they be finished? Now, a little hope is in order. The 4% cannot hold steady forever, so it would be wise to say it too will be a declining figure as we progress. Because as Bitcoins get more valuable, people will be more careful with their keys. It is inevitable. Taking it at face value though, we shall run out of bitcoins to trade within the next 18 years. So Circa 2042. I hope the naysayers will be back to tell us they told us so, or to plead guilty of doomism and explain why there are still many millions of Bitcoins being traded daily. However, it certainly looks like way before the year 2137 when the last satoshi will be mined, we shall hit a big wall as per the coins lost vs the coins in the available supply. When enough is enough, for lost Bitcoins Assume at the time this shocking reality hits the Bitcoin community, we have only 4 million coins tradable around the world. Assume even Michael Saylor shall have lost some of his keys/coins. What then? Well, the price of a single coin, feeling so much demand, will spike to about $10 million, and then it will dawn on everybody. Bitcoin has a $40 trillion market capitalization and not a $210 trillion market capitalization. Right now, this is an elephant 🐘 in the Bitcoin room, but it is still a baby and we are all happy to pet it and feed it with careless abandon. Soon, however, it will become a giant. A big fat frustration. At $10 trillion per coin, the fees to move even a few sats will likely be thousands of dollars because; a) the Bitcoin reward is smaller. b) miners have also lost their coins. c) losing coins to hackers will be most likely (hackers will be sniffing all over every wallet, waiting for the SSL to get some slight holes) Everybody will be losing coins. It will be wild. But before I proceed? Why do people lose keys hence coins? Well, if you are using the key every time, sometimes changing it this way and that way to maintain your security (because AI powered cameras may be watching), it so happens that you are bound to make a mistake and lose the key. Or else, be hacked (since you cannot fathom changing the key all the time). Like smart people must have said, meat brains are not good at holding random numbers or random sequences of words. Luke Dash Jnr, a seasoned cypherpunk, was hacked last year. He lost 200 bitcoins. Elon Musk may have lost his Bitcon keys back in '20. So, shit happens. To hard fork those lost coins back, or not. That is the question So here's the thing. We wake up and Bitcoin is seizing up. Crashing into its mass and threatening to take our financial dreams with it when it turns into a black hole of lost keys / burned coins. We also wake up and realize that the debt spiral is still sucking us down with it, even as we hodl Bitcoin. Climate is now a real issue and oh, AI is giving us headaches. Too many addictive cat videos. Too many coins are lost, the blocks are always full so confirmations take decades, the core software is too large, the fees are too high. So we sit down, like AA group members, and decide that continuing to preach financial freedom yet the currency is fast tending towards technological obsolescence, is like being high on alcohol and thinking that one is blissful for other reasons. We need to quit the 'lost coins are good for all' alcohol. In short, it will be time for a hard fork. The Hard Fork --> Bitcoin ver. 2.0 Innovation comes full circle. Innovation comes full circle. After factoring in the reality of available supply and not merely supply, we could let Satoshi Nakamoto's 2 million or so coins be our target practice and our saving grace. One shot and bull’s eye. Trillions of dollars pour down from the Bitcoin piñata. Like manna from the heavens. Right? Alas, There shall only be 4 million bitcoins and the value of lost coins, if "re-supplied" to the blockchain, will quickly deflate the value of the 4 million coins existing. Overall, lost coins would have helped the remaining coins get valuable to that point. Say $10 million per BTC. If you un-lose them by the most brute force of them all -- a hard fork -- you shock the demand economics and viola, have worth-less bitcoins than you started with. Now whoever is holding their previously un-lost bitcoins cannot be happy about that. It begs the question, what would they do? Refuse the hard fork outright. Obviously. Indeed, even if the value of the bitcoins would overall increase the market cap of the available supply, we can be damn sure it would reduce the value of each individual coin. Supply and Demand. Economics 101. The Big Question: Is this a financial revolution for the better, or for the worse? For example, let's say we had 4 million coins left. We then redeem the coins in the wallets of those proven deceased without any heir. Say Satoshi Finney is that first choice. At $10 million per coin, 4 million coins give up a market cap of 40 trillion USD. Add 2 million coins to the available supply but have the price shocked from 10 million to only 7 million. Total availability supply = 6 million. 6 million x 7 million USD = 42 trillion dollars. It is a good financial explosion. But to whom? Who gets the money?? Who makes the ultimate sacrifice of accepting to end up with less money than before? If rational and educated human beings cannot agree on simple (obvious) things like "Are fossil fuels necessary for economic progress or not?" , how can they agree on how to share this pie? "Are fossil fuels necessary for economic progress or not?" It is a financial explosion, except the majority will not want it. Satoshi may have solved the Byzantine General's problem. Somebody needs to solve the "Goodness of the community over the Goodness of the individual" problem. Of course, with my limited brains, I know most people are wary of a Marxist dystopia sliding in during this very crucial process. E.g. The coins could instead be coded to go to a handful of powerful individuals, whereas before it seemed like they would go to people who deserved a financial turnaround. Especially in the wake of what we see today with government debt ballooning, government operatives hoping to snatch some of this reclaimed BTC so as to help the government pay its gigantic debt is not an impossibility. Talk about being stuck between a rock and a hard place. Not the best kind of sandwich.

Is Lost Cryptocurrency Still Valuable, and Can We Reclaim It? 

Stories, News & Mentions about AMPLOSION

Amplosion’s Stories on HackerNoon

Amplosion's News & Mentions