Ramesh Raskar was born in Nashik, India and he finished his engineering education in Electronics and Telecommunication from College of Engineering, Pune in 1991. He was an avid photographer in his college days and this interest led him to do a PhD in projector based 3d graphics from UNC, Chapel Hill. Raskar then joined Mitsubishi Electric Research Laboratories where he worked in computer vision. After 6 years at Mitsubishi labs, Ramesh joined academia.  He became one of the youngest faculty members to get tenure at MIT Media Lab. Because of his work on camera software, a lot of professional photographers are being replaced by amateurs, without anyone noticing.

At MIT, Ramesh developed a computational display technology that allows patients cataracts to perceive a focused image on a screen without wearing refraction-corrective spectacles. The technology uses customized filtering algorithms that pre-distort  presented content. His lab produced a number of extreme high speed pictures using a femto-camera that took images at around one-trillion frames per second. They have also developed a camera to see around corners using bursts of laser light. Watch Ramesh’s TED talk on ultra high speed photography here: https://www.ted.com/talks/ramesh_raskar_imaging_at_a_trillion_frames_per_second

Ramesh was in India in connection with the Kumbh. Being from Nasik, he has a connect with the festival – and he has been a regular observer / advisor since 2015. With digitalisation, the Nasik Kumbh went flawlessly with no stampedes and no masala for future Hindi film makers about brothers lost in the Kumbh mela. He believes that gatherings like the Kumbh are a major sandbox for all kinds of entrepreneurship. The 2025 Kumbh in Prayagraj is expected to host 40 cr visitors and create 10 b dollars in revenues. Ramesh, a true blooded generalist, has seeded an NGO to encourage Kumbh entrepreneurs, https://kumbhathon.com/ His team has pledged to make the 2025 Kumbh as AI first.

Here is an article from Forbes by John Werner, one of his students, which summarised most of his talk at COEP. I have added some of my own masala to the original essay.

When we talk about “god-like AI” what we really mean probably has to do with the convergence of trends that will supercharge these models and agents beyond what we can keep up with. In other words, AI may become so capable that as its contemporaries, we may end up being a little like cavepeople in the early world, less able to control and outthink other sentient forces in our environment.

If all of this sounds esoteric, we can get a bit more of a real-world flavor from Ramesh Raskar who runs classes here at MIT on the future of AI (he’s also head of the MIT Media Lab’s Camera Culture research group.) He presents a sort of AI mythology, if you will, that you can think of as maybe similar to the ancient Greek or Roman or Norse formulations, in its treatment of forces beyond our full understanding.

Raskar talks about being “the voice of God” in a hotline setup at Burning Man in Nevada, back in the landline days. Strangers would talk to him about their fears and secrets. This is exactly what AI is doing in our world, quoting Raskar: “What we really need is a trusted, honest, impartial broker – the definition of God – we need a newfangled AI, a god-like AI, to do that orchestration.”

I found his examples interesting. You have Google Maps as the “God of traffic” and ChatGPT as the “God of Public Information,” but also, Raskar suggests, we might want even more connected “gods” involved. “What we need,” he says, “is that deep-down information that’s hidden well below the surface.”  What else could we have? Maybe Dall-E, the “God of art” and Amazon the “God of instant delivery”. Or, as Raskar notes, entities that can use a decentralized Internet to, say, sell products or find cures for diseases.

I thought about his analysis of two big trends that are unlocking potential for god-like AI: one is ubiquitous compute, which I think many would agree we have in the age after Moore’s Law, and the other is the proliferation of personal agents, AI entities that will do our bidding, or at least do things on our behalf. To quote Raskar: “When you take these two trends and intersect them, suddenly AI can start thinking like a scientist.” “We can wait for them to become smarter and smarter, like what we’re doing now, or we can take these simple agents and let them start talking to each other.”

In Raskar’s scheme of things, different AI agents will talk to each other. These would be decentralised – which means that no single entity will be in control, and so to some extent the fear of a Matrix film inspired power can be mitigated. Using a bad analogy, we can compare this to a drone swarm that is used to overwhelm defences. This is also good for data privacy and trying to set a stage where damage by bad actors is minimized. These specialised GPTs can be on the cloud to start with – but they may become smaller – allowing migrations to PCs. The components of this GOD can be compensated by a block chain model where service providers whose data is used to build the answers to disciple queries, can be compensated depending on the contribution made by the respective God to the query’s solution.

Raj Simhan, Ramesh’s colleague at MIT, demoéd a God prototype – the God of research grant proposals. He took an idea from the audience – non-invasive thyroid problem detection in young women. This was fed into the agent – and at the end of 10 minutes, it had done a detailed primary research, come up with 3 alternate hypotheses that it would explore and a possible solution that would involve wearable tech, algos for preserving data privacy. It also assured us that no one else is doing something in this area. The last one turned out to be untrue, as the person who had asked the question had her daughter already working on this – with a grant made by DST. 

Here’s the acronym that Raskar gives to explain the workings of the “deux ex machina” – Global Orchestration of Decentralized AI. In terms of privacy, as he notes, we’ll have to embrace a certain flavor of pseudo-anonymous delivery in order to facilitate these techniques.

Another idea I thought was relevant is the study of incentives on data markets: how do you get people to participate? Raskar breaks down a system called “CrowdX” which relies on orchestration, discovery, visualization and decision support.

In the end, we had a Q and A. I managed to squeeze in a question. In one of RR’s talks he tells about how tech changed style. When sandpits started getting replaced by foam beds for high jumps, innovations like the Fosbury flop happened. I asked him about whether AI was providing a Fosbury Flop moment for education – how it feels to teach a class which doesn’t need to be taught anymore. 

MIT, and most top US schools, allow AI to be used by students. RR used an analogy of today’s AI being equivalent to yesterday’s calculator. I pushed him further to share his take on non AI skills. He is on the verge of publishing a book on that soon – but let me get you a sneak preview: critical thinking, problem solving and understanding relations – are races that humans may still win against the AI bots!

Appendix – Raskar Gyan

Gyan on Inventions

Ramesh is a modern day Edison – and seems to be working with his team to push in a fair number of innovations. Here is the Master sharing his bullet points on invention:

• Cleverness alone is not enough to become a good inventor
• Difference between problem-solving and invention – working in isolation can just solve a problem, while to invent you need give and take.
• Invention is all about people. If you don’t work with the right people you don’t get inspired to work in the right way.
• The true power of an inventor is less about expertise on one subject, but rather the ability to ask questions no one else is asking and follow the trail of answers as they are revealed.
• Inventor’s job is to think in an anti-disciplinary manner – look beyond disciplines
• To make a grand difference, ensure the problem you’re trying to solve is the right problem. 
• The “spot probe” methodology is something every inventor needs to master. It is a continual cycle: Ask a lot of questions. Spot a lot of problems. Articulate those problems. Then probe their potential solutions.

Gyan on PPTs

Creating intentional errors in your data can help do a plagiarism check with AI. Worth trying out for academicians.