paint-brush
The Dark Matter of AI: Common Sense Is Not So Commonby@sparsh0427
841 reads
841 reads

The Dark Matter of AI: Common Sense Is Not So Common

by Sparsh November 19th, 2021
Read on Terminal Reader
Read this story w/o Javascript
tldt arrow

Too Long; Didn't Read

Artificial Intelligence has undoubtedly emerged as one of the technological successes in the past decade. With the amount of research and investment going into this domain, it is nowhere near an end. But there are still areas where AI lacks and causes problems — frustration, to the end-users, and these areas posses a great challenge for the researchers in trying to improve AI. Researchers have been trying, investing an admirable amount of time and resources to solve the problem of lack of common sense in artificial intelligence technologies.

Companies Mentioned

Mention Thumbnail
Mention Thumbnail
featured image - The Dark Matter of AI: Common Sense Is Not So Common
Sparsh  HackerNoon profile picture

“COMMON SENSE” is the Dark Matter of Artificial Intelligence.

In the present era of Artificial Intelligence, Deep Learning, advanced quantum computing, we humans are literally surrounded by machines, everywhere, everyday.


Many critics point to Artificial Intelligence as the main threat to humankind; while on the other hand, the supporters of AI claim that humans can never be replaced by machines, and would only ever compliment our abilities.


Over the past decade, Artificial Intelligence has undoubtedly emerged as one of the technological successes and with the amount of research and investment going into this domain, it is nowhere near an end.


AI has impacted our lives greatly, with so many services and products relying on it that it is irrevocably connected with our everyday world. Whether it be our smart home devices or a simple Google search, the impact of AI is everywhere. But there are still areas where AI lacks and causes problems — I would say frustration, to the end-users, and these areas pose a great challenge for researchers trying to improve AI.


Machines are dumb boxes — they can only perform the tasks they have been trained for. It is our dynamic thinking ability and common sense that makes us much superior to the machines.


No matter how much we train our models, how much test cases we train our machines for, there is still a room of uncertainty that could be in reality very simple to solve but AI would fail because it lacks the trait of Common Sense.


“Machines are dumb boxes — they can only perform the tasks they have been trained for.”


Many renowned researchers have been trying, investing an admirable amount of time and resources to solve the problem of lack of common sense in artificial intelligence technologies.

One such person was the renowned philanthropist, co-founder of Microsoft, Mr. Paul Allen. Before passing away in 2018, Allen had invested heavily to help incorporate common sense in AI technologies.


Allen founded the “Allen Institute of Artificial Intelligence” in 2014 whose main focus is to research and engineer artificial intelligence. Allen launched a project named “Alexandria” — a research program to help solve the problem of common sense in AI.


We would be focusing on Project Alexandria a bit later in this blog. For now, let's take a look at the different techniques researchers have adopted to provide common sense to AI technologies.


“If we want AI to approach human abilities and have the broadest possible impact in research, medicine and business, we need to fundamentally advance AI’s common sense abilities.” — Allen


Common Sense: Previous and Current Attempts to Meet This Need in AI

Symbolic reasoning

This technique is now commonly referred to as the “ Good Old Fashioned Artificial Intelligence”.


Symboling reasoning basically refers to mathematical logic, providing explicit embedding of human traits and knowledge into the machines. This initiative came into existence in the 19670s-1980s; however, was not very successful in providing common sense to machines, because there are millions and millions of rules that need to be programmed explicitly to the machines; which is not possible when dealing with real-world scenarios, and creates fuzziness, which in turn makes the process complicated.


Semantic networks

Semantic networks basically refer to representing data and the relationship among data in the form of graphs, nodes, and links. This network was able to solve the fuzziness problem faced by symbolic reasoning, but the other problem faced by this technique was that the relationship among data is dependent on the creator and would vary from creator to creator and hence would mean different meanings to different people — put simply: semantic networks are not intelligent.

Deep learning

Deep learning is a subset of Machine Learning. It mimics the human brain in its working. It does so by adding the concept of “neurons” in its implementation for decision making.


No doubt neural networks are a breakthrough in the artificial intelligence domain, and had boosted the process of learning by a huge factor, but still it is far from achieving our motive — that is to add common sense in machines.


Although some of the Neural Network models have come much closer to human reasoning such as AlphaGo which uses the famous Monte-Carlo tree search and GPT (Generative Pre-Trained Transformer).



COMET

Comet’s main motive, as its name suggests (Commonsense Transformers), is to implement common sense in Artificial Intelligence technologies. COMET is a combination of both Symbolic reasoning and Deep Neural Networks.


The main focus of COMET is to make probabilistic responses, rather than deducing results from the knowledge base. Presently, Yejin Choi of the Allen Institute is working on COMET, and she believes that the neural networks could make much better progress in achieving common sense among machines where the symbolic approach has not produced the desired results.


Project Alexandria:

Project Alexandria (named after the ancient library in Alexandria Egypt) could be considered as a lighthouse for guiding the future of Artificial intelligence technologies. Allen had invested a whopping amount of 125$ in the Allen Institute of Artificial Intelligence to facilitate this research.


Common sense, which we have inherited/consider very simple or maybe even not notice, is not so easy to implement in machines. Project Alexandria is a complex solution to the common-sense problem. Alexandria would help in achieving the common sense capabilities in AI by combining Machine reading, Natural language Processing, Computer Vision, and crowdsourcing techniques and thus building a foundational common sense knowledge base for future AI systems to build open.


The inputs/data feed is expected to come from the following sources —

Common sense is the precondition for general intelligence; until we get there we will be stuck with narrow AI that is rarely robust and never as flexible as human reasoners,” said Gary Marcus, Founder of Geometric Intelligence (acquired by Uber) and Professor of Psychology and Neural Science at NYU


There are a lot of prospective and applications of employment of machines showing common sense capabilities.


Last year in May 2019, Microsoft released a research paper comprising of their work under the project titled “Alexandria: Unsupervised High-Precision Knowledge Base Construction using a Probabilistic Program” for which they also received the Best Research Paper Award. The research paper focuses on using the probabilistic approach for converting the knowledge base facts into unstructured data. You can go through the full research paper here.


The researchers at the Allen Institute of Artificial Intelligence have now been working very hard, and devoting an admirable amount of time and resources, in the hope that soon they will come up with a solution to incorporate common sense in artificial intelligence technologies. Currently, COMET has given some hope and their work in this domain has already come a long way and if successful could prove to be a major breakthrough in the Artificial intelligence domain.


Just like at the beginning of this decade no one would have ever imagined that the Artificial Intelligence and Deep Learning domain could create such a boom in the technological field and they would have such a strong impact on our lives. It’s my gut feeling that, by the end of this decade, or maybe even earlier, we will be seeing some truly mind-blowing advancements in this field.

.