If your beloved zero-knowledge proofs are so awesome – why don’t we have multiple multi-billion companies serving ZKPs for real-world use cases?

It’s not always about solving the problems of yesterday. Not all products are built for clear existing Product Market Fit, especially when it comes to DeepTech. Some technologies take time. Some products are based on intuition and guesses about the future.

Think about AI. One of the largest AI companies, Palantir, was started in 2002 and founded in 2003. In 2002, there were 631M internet users in the world and 135M smartphone users in the world. Internet penetration reaching almost 70% (as is today) was a very doubtful take (with large share of mobile only users because of the huge drop in smartphone prices). Back then, in 2002, AI assumed analyzing tons of collected data, and meanwhile there was no clear idea of what type of data that will be and how it will be collected (if ever).

The whole reasoning of starting an AI company in 2002 was based on a guess that internet penetration one day in the future will be so high that we’ll be able to collect enough data and furthermore we will learn to analyze it in a sophisticated way such that it will allow us to make meaningful data-driven decisions.

It doesn’t mean that every single DeepTech company should rely on doubtful guesses about the future. But it does mean that sometimes it is the case.

Here is my guess about the future. Blockchain is NOT the next big thing. Neither are Ethereum nor zero-knowledge. Even if we want to believe in it – we should soberly understand that this is an illusion.

Meanwhile, the next big thing is AI. And this is already not a guess, it is our reality. Most digital environment we interact with – banking, dating apps, retail platforms, – all of them are powered by AI 24/7. Huge share of physical spaces – facilitating traffic on the roads, coordinating manufacturing facilities, making decisions on battlefields – all of that is powered by AI 24/7. Furthermore, our relations with people, even the closest ones, and our self-identity, personality, and individuality – all of that is deeply affected by what we’ve recently consumed from our  smartphone and laptops and all of that is is powered by AI 24/7. We don’t know how many LLMs touch our mind every day, but I can bet it’s at least dozens and maybe already hundreds. And what is the most fascinating and scary aspect of it – we are not aware of it.

We live in the reality penetrated by hundreds of AI models manipulating humans 24/7 to make decisions they are expected to make. But what are these decisions? The true answer – we already don’t know.

We were happy to outsource to AI routine tasks such as writing code or translating a letter. But how will we feel when AI outperforms us in creativity? Writing books that are sold better, making content that gets more engagement, writing music that was played more times? Will it happen because AI can analyze our preference data and meet our expectations better or because it will manipulate our preferences for long enough so that it can shape our preferences – it probably will be the mix of both. But I don’t have any doubts that it will happen on a timeline less than 20 years (probably way less). Humans, however we want to believe in our individuality and personality – as a fact we are trained neural networks. And these are AI models who are training us 24/7.

We already see the stories when kids say ‘Alexa’ earlier than ‘mama’. Meanwhile, AI models, especially those modern ones and those that will pop up in the near future – they are gigantic creatures designed, trained, and implemented by hundreds and thousands of people and while there are some caution measures aimed at holding the model accountable – from some point no one knows how they work.

GPT4 has a trillion parameters and this number is just going up and it will continue going up. And there is no doubt that one day (rather sooner than later) AI will make decisions autonomously without reporting to humans. Don’t get it wrong, I am not saying that it will take over the world, I don’t think it will happen in a straightforward way. AI will be granted a right to make its own decisions for the sake of efficiency. With thousands of LLMs engaged in making all core decisions in the world – they just won’t have time to report to human and allow the human to make a decisions. Because in many cases, timing matters. If one party makes a decision fast another party will always aim at making a decision faster.

This is the reality we have to handle. This is the reality we will put in front of the next generation. When it happens (i.e. when there is a straightforward product market fit) – we already won’t have time to look for some special tech stack to handle the issue. The worst case scenario is in at most 20 years we are living in a world fully powered by LLMs where humans have absolutely no idea how these LLMs work, what they are doing, and why they are doing it. LLMs will autonomously make decisions in nuclear, buttlefields, economics, geopolitics, media, and all other domains.

We can pretend we are shaping the models but at the end – they will shape themselves. And it doesn’t even matter if AGI ever happens or not.

The question that makes sense to ask today is will the machine remain subordinate to its creator? Will AI remain subordinate to humans? The true answer is we don’t know but we can try to do our best for it.

We can try to come up with a tool for collaboration between humans and AI. And my best guess is that zero-knowledge proofs are a technology that can make it possible. Because while in most human interactions (either personal or at a country level) trustless relations are rarely needed for real. In human-to-machine relations, the machine will remain subordinate to the human only if whatever is done and reported by the machine can be verified by the human.

tbc…

~

Have a thought you want to share – drop me an email lisaakselrod@gmail.com.