Artificial Unintelligence

What a start to 2023....

I have been writing this post for three months since the hype openAI-GPT4 created . I've written, updated, expanded and rewritten - and now it has become 4 posts! Every time I felt there was something urgent and important to be shared, it would take a week to write up. And I would feel this world has dramatically changed... ever had that feeling?

Yet I don't think the media and pop academia has remotely understood the really important aspects. Now, I don't have some special divining power nor PHDs that the AI field seems to require as an entry criteria. But I do have something that most of them don't. I will explain that in my next article, and focus here on the main thrust of my arguments and predictions...

RIP Coase’s law

So, there's this complicated thing called Coase's theorem, right? Something fundamental to the corporate and legal world - It says that if there are no pesky transaction costs, people can just sort things out by talking and making deals. In other words businesses organise because? they reduce transaction costs. And so, over centuries our organisations grew larger and larger to achieve value for the consumers (and greater value to its owners).?

The increased value has been questionable for some time now as organisations became less robust and more bloated with size. In the past few years we have seen a Cambrian explosion in exponential technologies indicating an?inversion in Coase’s idea - and portending smaller organisations - as E.F. Schumacher envisioned.

Now, if we extend this vector further it would to lead to decentralised economies. At some point. Direct deal-making will be increasingly feasible with smart contracts and the ability for an individual to produce diverse services with low transaction costs made possible recently thanks to the developments in AI.

As AI continues to evolve, the cost of production and the supply of goods and services should rapidly approach zero. This paradigm shift is driven by AI's ability to automate more and more complex knowledge and mechanical processes - increasing efficiencies, and reducing? overhead costs. Simultaneously, the emergence of Web3 technologies, albeit slow, such as blockchain, is revolutionising market dynamics by enabling trust-less interactions and digital property rights. Through decentralised platforms and smart contracts, Web3 ensures transparency, security, and efficiency in economic exchanges.?

I fully expect in the coming decade (perhaps 3-5 years if no brakes are applied), AI-powered automation and Web3's trust-less networks will disrupt the traditional economic landscape, presenting both challenges and opportunities for businesses and individuals alike.

This is a three-post essay. I will provide a short thesis of the trilogy here today. In the coming fortnight, I will detail out these three parts of the story of the new era about to unfold -?

> Post 1 - Prelude: a little History of AI and I -?

Summary (Full article will be Linked in the coming fortnight) -?

1. My background: The last few months saw a lot of exaggerated hype - many claimed that AI/ML has shaken up the world in just 5-10 years. While the impact it has had and is yet to have is definitely unprecedented, this has not been an overnight phenomenon. The first (successful) implementation of AI was by Rosenblatt et al, in sorting punch-cards for Bell labs in 1950s! Heck, even I wrote my first back-propagation C++ program in 1996! For a 3-layer perceptron, as part of graduation thesis in IIT Madras, to achieve process control in chemical reactions. And it took me 2 days to train the weights on the spanking Pentium I PC!! AI has come a long long way since then...
2. AI history: To fully understand why this is a watershed moment it really helps to get a sense of history and the fundamental advances in AI - from the "symbolic" vs "connectionist" models, the 3 decade scepticism to connectionist neural networks, to the critical breakthrough of deep learning and generative networks. Genius Makers does a great job in chronicling this story
3. My lens of AI: cross-functional. Giving me a vantage point that is quite unique - . I came across only recently an expert who share many of these - a) An understanding of AI architectures + mathematical foundation b) Brain anatomy and cognitive neuroscience c) Understanding of intelligence and consciousness - as a philsophy of science + experientially in many forms of deep meditation. I was euphoric to come across this lecture by Shamil Chandaria - hope to reach out this kindred soul, for until now could never discuss this field holistically with anyone. This lecture combines what I think is a fairly complete set of mental models that would ultimately help us evolve as a species

> Post 2 - Wonderland: a summary of trends and disruptors

Summary (Full article will be Linked in the coming fortnight) -

• Machine intelligence, finally!: After the 1997 episode of Deep Blue beating Kasparov it seemed like the machines had finally come of age. Yet even 15 years after that, AI was a subject for PHDs and dreamers. It wasn't until 2015 when AphaGo showed the potential of machine learning that AI entered mass awareness and impact
• AGI /Singularity : Ex Machina seemed dystopian and distant. Now it looks prescient. Or is it really? Microsoft claimed GPT4 was beginning to show glimpses of AGI. That may be a little more hype than truth. But Jeff Hinton's moves and the insider, Sam Altman's evidence have suggested that GPT5 is definitely an order of magnitude more intelligent. And now Jeff Hinton's new direction on forward-forward algorithm only make AGI even more likely. But then again, is AGI just further in the continuum of human/neuronal intelligence? I believe only a super-intelligence can know that for sure.

• Near-term predictions: Kai-fu Lee's AI super-powers would already seem dated in 4 years, but does a pretty good job of predicting an arm's race. The past 6 months has seen this drama play out - many sane voices have spoken out about dire risks facing an unchecked advancement of AI tech. Contrapuntally the same voices have also acknowledged that the power of this tech will inevitably mean it will be the differentiator between the future super-power in a changing geo-political landscape. A cold war definitely may be on the cards.

> Post 3 - Looking glass:? The opportunities and horrors of the next decade

Summary (Full article will be Linked in the coming fortnight) -?

• Singularity: The AI tech unlike many before, will not be an asymptote or terminal hype-cycle. It is an exponential curve which we cannot see the (tr)end of as beyond the fast approaching singularity of AGI - critical to achieving a happy ending to the AI story would be our ability to solve the alignment problem of AI, so the human race does not get wiped out in pursuit of paper clips!
• Evolution of co-operation: many smart and influential people have understandably rung the alarm bell when the AI curve dramatically inflected in the past years. Basically there are two types of concerns - socio-economic and evolutionary/alignment. The (smarter ones) have been concerned about the near-term issues, while the dystopian luddites worry about a AI-pocalypse a pretty interesting narrative of which is presented in the Animatrix short film series, a backstory to the Matrix trilogy. While I believe the short-medium term concerns are definitely likely, the long term doomsday worries do not take into account the fundamental idea of altruism as a consequence of evolutionary co-operation - as beings get more intelligent, ipso facto they would optimisers/satisficers of a greater number of constrains. Axelrod's framework of game-theoretic optima naturally leading to co-operation is only likely to get better as AI becomes smarter - assuming it does... but then, we still have to worry about "natural stupidity" and the evil weponisers who will create "stupid enough" AI that can hurt us.

• Web3: AI was an extremely high-investment tech in the past 2 decades. Yet most super-orgs with deep pockets (The MATMAN - Meta, Apple, Tesla, Microsoft, Amazon, Nvidia...) have fumbled (relatively) - Microsoft and Meta made multiple faux pas in the past five years in an attempt to play catch up. Only Alphabet is arguably ahead. Its close competitors are OpenAI and Deepmind - both newcomers and both are a fraction of the size. In the future AI could be even more democratised. Blockchain capabilities will be crucial to ensure the innovation reaches the masses making way for a whole new class of 10x wealth to be made
• Abundance: this is a very large topic and could be a book in itself. For, the inevitable abundance of resources/wealth that could result if the exponential technologies keep improving (AI/AGI, Genetics/Biotech, Solar/Nuclear, Moore's law of hardware, Web3) will impact the world like nothing has done in the past - social impact from ending of time-scarcity, economic impact of zero-cost production, political impact of decentralised nation states and economic zones, cultural impact of creativity at speed of light, physiological impact of healthy and productive life spans...

Conclusion

These are unprecedented times. We would do well to remember Amara's law now. I believe a golden era is indeed coming in the next decade. AI and more generally Digital Intelligence is likely to be the inflection point for carbon-based evolution, and we break into a Type II level in Kardashev's scale. But as with most rainbows we should expect some stormy times to precede it. Here is to hoping some smarter people or intelligence would chaperone us through it...