AI And The Great Bifurcation of 2024

What happens when physics constrains AI?

This newsletter is 100% human, written by me. No ghostwriters or AIs.

Happy Sunday and welcome to Investing in AI. I’m Rob May, CEO at Nova. I also run the AI Innovators Podcast, and the most recent episode is about using synthetic users for product feedback. Definitely worth listening. I’m also a very active angel investor (most recently in Consensus) so please reach out if you have an interesting startup.

Today I want to write about what I believe will be the great bifurcation of late 2024 - the splitting of the economy into things that can advance with AI, and things that can’t.

The world has been awash in predictions about AI since GPT-4 was announced with results that have surprised and impressed many people. Microsoft published a controversial research paper claiming GPT-4 has early indications of AGI. Over 1,000 AI experts signed a letter asking that the major AI research labs pause, so we can have time for AI safety and security to catch up to the power of things we are building.

But I want to take a step back and think through the implications of GPT-4 (and the next few versions of similar models). Yes, things like creating digital content, writing code, building apps, and making predictions is going to get much much easier with these models as they approach human level generalized intelligence. However, that doesn’t necessarily mean they can act in the world. And when they do, through robots or some other means, they are still subject to the constraints of the physical world.

I believe we are making a dangerous mistake by extrapolating the speed of progress AI can have in a digital world to the speed of progress AI can have on the world in general. Think about this - even if you have the world’s smartest AI, 10x better than a human, there are still things that can’t get that much more productive.

In manufacturing, you can only turn a screw so fast. A robot can’t build a car in seconds. It takes time for motors to get moving, for electricity to flow, for tension to be applied, for welds to take hold. Chemical processes take certain amounts of time to work. Concrete takes time to dry. AI may be able to help improve some of these processes but, it can’t instantaneously dig an oil well or frame a house. It runs up against limits defined by the laws of physics.

I think we in the tech industry tend to be overly optimistic about the breadth of the impact tech can have. Reading books like, How The World Really Works by Vaclav Smil can help to reset that - pointing out how much of the workflows of the world still rely on cement, fertilizer, and oil. These are massive industries that are governed at some level by physics and chemistry and things AI can’t change 10x.

What I think this leads to, over time, is a bifurcated economy. We saw a bit of this over the last 20 years where things computer and internet related saw serious deflation while real world goods saw inflation instead. More of the economic gains also accrued to industries with more intangible assets like tech and finance.

I expect that trend to accelerate starting in late 2024, which is why I think we will look back on 2024 as the beginning of the great bifurcation. This will be the time when new AI models (GPT-5?) really start to have massive impacts on everything non-physical, exacerbating the more minor bifurcation between intangible and tangible assets we’ve seen in recent decades.

What impacts does this have on society? What does it mean for investment in startups? How does it impact business models? There are many interesting questions that will need to be answered. But rather than think of AI as a tidal wave that washes over the whole world, improving everything, we should think of it as a rapidly flowing group of tributaries that deeply nourish some places while only having a modest impact on others. It’s important those of us grapple with the possible sociological implications of this coming bifurcation, and design a system that works for the broader benefit of society.

If you’ve thought about this, I’d love to hear your feedback as always. Thanks for reading.