Happy Sunday! This is Technically Sentient, a newsletter about AI and Intelligent Systems of all kinds. I’m Rob May, a Partner at PJC, where I invest in AI and Robotics. If you are raising an early stage round, please reach out.
This newsletter approaches AI from a slightly technical perspective, but more applied and business focused. I’m interested less in new techniques for squeezing better performance out of neural networks, and more what that technique would mean for applications, business models, and the economics of AI. Today in the commentary section I’m talking about financial capital vs production capital and why we haven’t seen an AI financial crash like we have in other areas. But first, let’s get started with reviewing some great links from the last two weeks and a few research papers.
— Links —
If you are only going to read one thing, read Nathan Benaich’s State of AI report. It’s excellent. State of AI.
The Future of Propoganda Will Be Computer Generated. The Atlantic.
Google Launches AI Platform Prediction. Venturebeat.
Facebook Rethinks AI Benchmarking with Dynabench. Facebook.
Can AI Model Economics Choices? Brookings Institute.
Object Detection With Synthetic Data. Synthesis AI Blog.
Is TOPS The Best Way to Measure AI Chips? Venturebeat.
Women AI Entrepreneurs Take Issue With Gender Bias. WSJ.
— Research —
Deep Learning For Time Series Classification. Link.
Adaptive Multi-Grain Graph Neural Networks. Link.
Emotion in Future Intelligent Machines. Link.
— Commentary —
I’m a huge fan of Carlota Perez’s work as laid out in her book “Technological Revolutions and Financial Capital: The Dynamics of Bubbles and Golden Ages.” The book lays out how financial capital is invested in new technologies during the “installation period”, often in speculative ways, but the overinvestment helps lay the ground work for the “deployment period” that comes after. In between there is normally a financial crash. The cycle is charted below.
The question is, why hasn’t there been an AI financial crash? Perez’s framework would imply that with the burst of AI investment in the 2015 - 2019 era, we should be in the middle of a big financial crash in AI companies. Yet, the failure rate so far seems to be in line with startups of all types, and we haven’t seen any pets.com style high profile AI companies crash.
My belief is in two parts. First - it’s coming in another 4-5 years for software/hardware. The issue with AI is that is came about using infrastructure that was already in place. The idea and mathematics of neural networks is several decades old, and GPUs to program there were in place for computer graphics. And of course everyone mostly used existing data sets, or pushed to get newer relevant data sets (which probably doesn’t qualify as financial capital overinvestment at those levels).
So I think the crash will come after the AI infrastructure changes that are coming finally get deployed. It starts with AI chips - new chips designed for new workloads. There are dozens of companies working on these and only a handful of them will win, but financial capital has chased the potentially stellar returns that 3-4 of these AI chip companies will provide.
Once the AI chips get to market, most of which work differently than the x86 chips we are used to, the tools and applications will change as well. While the chip companies have tried to use existing tool chains as much as possible, I expect to see a minor explosion in new tools 2-3 years after the first real production AI chips are in market, and a huge explosion in new applications and ideas 3-4 years after. That’s where the crash will come from.
AI chips and the relevant tooling are the infrastructure that is coming to really change AI from simple use cases like predictive analytics to more advanced and complex use cases like logic and reasoning. Applications will align with chips, and when a chip crashes it will take down big pieces of that ecosystem. But it’s coming.
The second thing I’d argue is that, if you include robotics as a form of AI (which is debatable, depending on the company and use case), I’d argue we are in the middle of a mini-bust. Several major robotics companies have failed in recent years (Jibo, Rethink both come to mind), and a few more are on the ropes. These have required a lot of investment, but have laid the infrastructure for a robotic boom by standardizing many of the parts for robotic systems.
The moral of the story then is, if you invest in AI, what we’ve had is almost like a pre-boom. The real boom and bust is still ahead of us. Be careful. And if you have thoughts on this issue, I’d love to hear them.
Thanks for reading.
@robmay