Welcome to HackerNoon’s Writing Prompts! If you’re an actual winner of an award, or simply feel like a winner today, the link for the template is HERE.
Hey Hackers! I’m Sharad Sundararajan, an AI enthusiast and EdTech entrepreneur (co-founder and leading software at Merlyn Mind).
Merlyn is the teacher’s digital assistant, a multimodal solution built to give time back to teachers by automating their workflows and untethering them from the front of the classroom.
I would like to first thank the HackerNoon community for the nomination and a huge thanks to all readers for voting and recognizing me as the winner of HackerNoon Contributor of the Year - MATHEMATICS.
HackerNoon Reporter: What does it mean for you to win this title?
More visibility for the topic of AI in logic and mathematics, specifically Automated Theorem Proving (ATP), an area that has a long and rich history but still doesn’t get as much mainstream attention as (say) computer vision, speech recognition, NLP or self-driving cars might.
A lot of progress in ATP can be attributed to real-world applications like formal verification in both hardware and software (Eg. spacecraft control, RSA encryption…), but there remains a long way to go.
In a recent Nature article describing The Ramanujan Machine as an AI that can generate new mathematical formulae, one of the mathematicians George Andrews aptly captures the state of ATP:
“…although computers might be able to come up with mathematical statements, and even prove that they are true, without human intervention, it is unclear whether they will be able to **distinguish profound, interesting statements from merely technically correct ones. **
“Until I can detect a well-developed ‘sense of mathematical taste’ in AI, I expect its role to be that of an important auxiliary tool, not that of independent discoverer.”
Now there are efforts gaining popularity like the GPT-f where OpenAI is exploring transformer-based language models for ATP, but the jury is still out on whether directionally it’ll generate meaningful inference steps. ATP still appears to be an esoteric niche of AI and am hoping this piques the interest of a wider audience.
Personally, as an amateur ATP enthusiast, it is encouraging to get a nudge from the tech community to keep at it.
How do you or your company intend to embrace the responsibility of this title in 2022?
We at Merlyn Mind have stayed very close to key advancements at the intersection of multimodal interactions (voice, touch, remote control…) and AI (speech, NLP, knowledge representation…), but with an unwavering focus on making life easier for teachers in the classroom.
How AI is applied to real-world applications is core to our mission and I see this as another boost to our pursuit of AI in education, mathematics, and math-education.
What are some of the goals you’re looking forward to accomplishing in 2022 (whether it be through company initiatives or your personal journey)?
Teachers are overloaded with having to make several decisions a day on various topics including but not limited to their pedagogical approach, classroom management, technology or assignments.
Anything we can do to offload some decision-making, reduce their stress and avoid burnout will go a long way. So one of the goals for 2022 is to get Merlyn (the digital assistant) to more teachers to help them do what they do best, teach and light that fire in all students to build a better future.
On the personal front, I would like to complete the third and final part of the series on Centaur Theorem Provers this year.
As you look at the variety of concerning trends in the world around you, what solutions can you think of? Name three trends and three solutions. Be as brief or as detailed as you like.
Three concerning trends:
- Gap in teacher training worldwide.
- (D/M)isinformation.
- Digital divide.
Gap in Teacher Training
The United Nation’s SDG-4 (Sustainable Development Goal-4) states “Ensure inclusive and equitable quality education and promote lifelong learning opportunities for all”.
And teachers are key to achieving the SDG-4 targets. Yet not all of them are equally well prepared for the job.
As indicated in this 2021 World Teachers Day fact sheet from UNESCO (Figure-1), the gap in teacher qualification in some geographies (Sub-Saharan Africa in particular) is gravely worrying.
Some stats from the above source that are very telling:
-
“Globally, 83% of primary and the same proportion of secondary teachers held the minimum required qualifications. In primary, this proportion ranges from 98% in South-eastern Asia to 67% in Sub-Saharan Africa, while in secondary, it ranges from 97% in Central Asia to 61% in Sub-Saharan Africa.”
-
“In Sub-Saharan Africa, the proportion of teachers with minimum required qualifications has been decreasing since 2000, from 84% for primary and 79% for secondary due to growing number of private/community education providers, recruitment of contract teachers and constrained budgets.”
A related trend is teachers fleeing the profession due to burnout and low pay.
Teach for America for instance is seeing a 15 year low in teacher enrollment. (Figure-2 Source is Chalkbeat, a non-profit covering education news in the US)
There may not be a silver bullet to solve all the problems but a few key directions that we do need to invest in are:
-
Raise awareness and improve both domestic funding and international aid for teacher training. See International Task Force on Teachers for Education 2030 for related efforts.
-
Effective Professional Development (PD) programs. There are always multiple goals and priorities competing for teachers’ time, energy and attention, so more effort needs to go into carefully integrating PD into teacher workflows and enabling hands-on learning with a feedback loop where teachers can monitor their PD.
-
AI for PD. Directionally this is very much in line with Merlyn Mind’s own motto of “AI for the people building a better future”. Merlyn allows teachers to move around the classroom without compromising their access to technology (both hardware and software).
Teachers can control both the display and their own laptop in the front of the room using multimodal input (far-field voice on the AI hub, remote control with near-field voice input, an air mouse, a directional pad etc.). But one of the most critical functions of Merlyn is to enable micro-automations of teacher workflows to save them time and effort.
So teachers can just as easily jump directly to educational applications on their laptops and share material to the entire class as they can switch displays with a click or a voice command.
But there are a few efforts focused on AI for PD such as Edthena’s AI Coach that tries to personalize content to teacher’s goals. As part of their coaching cycles teachers get to try out changes in the classrooms and monitor their own PD, an important feedback loop.
Digital assistants are the future and we need to embrace it and invest a lot more into AI for PD to accelerate bridging the gap in teacher training.
(Dis/Mis)information
Another concerning trend is fake news, i.e. the intentional or unintentional spread of false information.
This is particularly troubling when sporadic efforts turn into systematically organized ones leading to disinformation campaigns that can be disruptive to entire countries at its worst.
Figure-3 shows an increase in the yearly frequency with which newspapers have uttered the phrase “fake news” and as this PNAS journal article titled “Science audiences, misinformation, and fake news” states “have arguably made more familiar—and therefore more believable—its false connotations.”
Nature had this article in 2020 titled ‘The COVID-19 social media infodemic” in which a comparative analysis of user’s activity on five different social media platforms during the COVID-19 emergency was done.
The authors charted out the ratios (regression coefficients) of questionable posts and engagements versus reliable ones. In one of the sources (Gab), the volume of questionable posts was only around 70% of the volume of reliable ones, but the volume of engagements for questionable posts were close to 3 times bigger than the volume of reliables ones!
The authors suggest that Gab is an environment most susceptible to misinformation dissemination. Their analysis also suggests that “information spreading is driven by the interaction paradigm imposed by the specific social media or/and by the specific interaction patterns of groups of users engaged with the topic.”
They conclude that the main drivers of information spreading are related to specific peculiarities of each platform and depends upon the group dynamics of individuals engaged with the topic.
Deepfakes
In 2017, a Reddit user by the name ‘deepfakes’ used deep learning to swap faces of celebrities with inappropriate videos and posted them online. The search for the word “deepfake” worldwide on Google trends yields the following trend:
Publications on deepfakes have shot up.
Here are a few timelines, one from a publication on synthetic media from the the Dept. of Homeland Security and the other from the Data & Society magazine that should make us revisit our tools to discern the difference between real and fake.
It is getting progressively harder to tell the difference.
If you’re interested in a deep dive into how deepfakes are created, here’s an excellent comprehensive survey from Georgia Tech. The authors show how with a combination of just a few basic neural network structures (Figure-9) deepfakes can be detected and created (Figure-10).
Also from that survey the authors present an information trust chart (Figure-11) that provides a helpful framework to think about true versus fake news.
At the risk of stating the obvious, for starters we simply need to use our common sense and be more vigilant, use more logical instruments like counterarguments and parallel argumentation to expose absurdities.
But given the sophistication of the disinformation attacks, we do need a multipronged approach that include (but are not limited to):
-
A call to action for tech companies to fight this. A good example is the Microsoft Video Authenticator that returns a confidence score of how much the media (photo, video) is artificially manipulated.
-
AI/algorithmic fact-checkers with humans in the loop to get a distributed consensus.
Digital Divide
The digital divide hasn’t been as pronounced as during the pandemic when k-12 education was upended with the poor internet connectivity and computer availability.
While availability is top of mind when we discuss digital divide, there are other dimensions that characterize the divide, such as affordability (cost, % of income paid for access), quality of service (upload/download speeds), gender (see stats from ITU and the Gender Digital Divide Index), security, interconnectivity or digital literacy.
This image from the 2022 Global Digital Overview Report shows the disparity in internet adoption across the world.
Here’s another view (Figure-13) showing the trend where developing countries aren’t growing in adoption as fast the developed ones.
Figure-14 further reveals the disparity between urban and rural populations even within developed nations.
There are no doubt several fantastic efforts out there already to bridge the digital divide.
It’ll be good to see a more holistic approach to expand access by including efforts like private LTE to cover the last mile and community networks like Metamesh (one of the first non-profit wifi efforts) especially for the underserved communities.
Looking at the device ownership data in 2022 (Figure-15), we should look at advancing mobile learning experiences. We also need to work around supply chain delays to maintain continuity in device access.
An honorary (and potentially very expensive) mention is the very troubling trend of ransomware damage costs increasing.
Here’s a snapshot from Cybersecurity Ventures predicting ransomware damage costs to hit $265 Billion by 2031.
The global cybercrime costs are expected to hit a staggering $6 Trillion by 2025!
After participating in this journey, is there anything that you would like the HackerNoon team to know?
You guys are doing a terrific job encouraging tech enthusiasts to participate and engage with the community.
What is something special about you or your company that you’ve rarely had the chance to properly share but really wanted to? (Give this one a proper think, I’ll know if you didn’t!)
The most special thing I can say about Merlyn Mind is the people. In house, we have an amazing and talented team driven by the mission.
Just as special are our end users (educators, IT admins/directors, partners) whom we believe to be our extended team and collaborators in this journey!
On the personal front, I have a penchant for spicy food, something many people might know. What may be less known is the certificate I got for my “dubious judgement” which (color me crazy) weirdly enough I do proudly advertise!
As a fan of Adam Richman’s Man Vs Food, I decided to try one of his feats - the Phall Curry challenge and successfully completed the challenge and got on the 'P’hall of Fame!
The chefs wear gas masks when they cook as they apparently use several different types of peppers! So yes, dubious is right.
On the left (Figure-18) is an unrelated (to the Phaal) picture from a restaurant in China that served Schezwan food. The crazy part is that everything floating on top is a chilli and it was delicious.
What song or music album are you currently enjoying?
These days I have Pandora on with different stations and offload the choice to the machine.
What are some concepts you wish people knew more about?
I have no idea what concepts people already know but I can mention a few that have served me well. Note that none of these are original but they’ve influenced my thinking.
- The twin ideas: Computational kindness and Computational stoicism.
- Multiple Representations as multiple ways of thinking
The Twin Ideas
I first encountered these ideas in Brian Christian and Tom Griffiths’s book “Algorithms to live by”. Here are the definitions:
“Computational kindness is a design principle that uses computer science to identify areas where seemingly simple tasks create lots of complex work for the people who must undertake them, and takes steps to reduce those overheads.”
“Computational stoicism is the use of computer science to identify the limits of certainty in the hard decisions we all must make in the course of our lives, and to help us be as certain as is practical in these areas — while letting go of the self-doubt that comes from wondering whether we could find more certainty by working harder.”
Machines try to be computationally kind by absorbing much of the complexity. This applies equally well to humans, where we can make it easier (by simplifying) for others to consume our thoughts and words. When we fail to do that, we are being computationally unkind.
Multiple Representations
One of the many definitions of the word representation that has stuck with me for close to two decades now is Ontological Commitment.
At the moment we choose to represent or model the world/system, we commit to one understanding of it, one functional or structural possibility.
But to understand the world in its entirety there is a need to probe it with as many tools as possible, use as many languages and levels of abstraction as available to us and generate as many theories as can help explain the different phenomena in the world.
Let’s take this Physics example of the dynamics problem in Figure-19.
The same idea of an elevator slowing down on its way up can be represented in a qualitative way, a semi-qualitative way with vectors to conceptually describe the problem and a quantitative way with symbols and equations to codify abstractions, rules, theories all to to solve problems in the real world. Each provides a different level of expressivity.
A related but wonderful idea from Bret Victor from over a decade back is Explorable Explanations where changing one representation impacts others. For a learner, this is a very powerful tool. I’d encourage readers to check it out here. A snippet from his writings that explains the idea beautifully:
"It's tempting to be impressed by the novelty of an interactive widget such as this, but the interactivity itself is not really the point. The primary point of this example -- the reason I call it an "explorable explanation" -- is the subtlety with which the explorable is integrated with the explanation" - Bret Victor
I would like to leave you with a simple yet powerful quote from George Polya (mathematician) on the same topic:
"It is better to solve one problem five different ways, than to solve five different problems one way”
Thank you for everything that you do! – HackerNoon Team.
For more information:
Read about the Noonies
Read the FAQs about
Thank you to all of our sponsors! We truly appreciate your contribution to the tech world and thus the future!
The Noonies Awards are sponsored by: