AI will make deep tech entrepreneurship even more exclusive

I've been tracking deep tech funding rounds over the last few months. Increasingly, I am seeing many software companies focus their story on their impact on the physical world, earning the "deep tech" tag along the way. Lots of the AI technology I am seeing solves some optimization problem (e.g., electric grid dispatch), runs predictive simulations to inform decision-making (e.g., wildfire dynamics), or identifies new areas for innovation (e.g., materials models). This may be part of the broader "physical AI" trend, but I'm starting to think it marks a fundamental shift in what it means to be a deep tech entrepreneur.

AI is transforming science and engineering

As funding priorities for fundamental research in the USA increasingly emphasize AI in science and engineering, it's getting harder to distinguish hard tech from software innovation. More and more graduates from science and engineering degree programs are working almost exclusively on AI-related research.

This will transform the role of a human scientist. The Economist has highlighted this shift in the October 17th issue:

"...human-guided machine science will require scholars to reinvent their disciplines. Scientists will have to learn to love—and credit with status and career advancement—guiding models and studying, verifying and teaching their results. Governments will need to allocate research grants on the same basis. For their part, AI labs should seek to be good scientific citizens, submitting their work for peer review and publishing their methods and code."

Scientists are losing their hands-on experience

And this trend is only beginning. As the cost of using AI falls, cloud computing becomes ubiquitous, and AI can set up model runs and other analytics, we are entering an era where computational research will make up a larger share of research. To push this trend further, we will begin designing automation solutions, with an AI research suite in mind. This will further isolate students from physical reality.

Science and engineering graduates will have less and less hands-on experience.

This is troubling. We live in a physical world, which often requires physical solutions. In many situations, the solution-space constraints are multi-variable and uniquely human. The best entrepreneurs start with a deep understanding of the customer and their physical reality.

Can a scientist or engineer without hands-on physical experience create physical solutions?

I don't know.

Physical experimentation will become a luxury good, available to an elite few

AI will, in many ways, be a powerful democratizing force in science and engineering. When anyone with access to compute and data can do science, more people can participate. Unfortunately, this could also commodify these activities.

Part of the reason the US has dominated science and engineering is that it had the wealth to build the most advanced labs with cutting-edge instrumentation. I worked with researchers from India and Egypt during my graduate studies; in their home countries, their research was mostly limited by access to equipment. My friend in Egypt had to do all of his inorganic chemistry in ambient conditions. They didn't have the resources to buy and maintain glove boxes to conduct air- and water-free research. He was amazed by what we could access in the US.

Labs and institutions with the resources to build the next generation of labs, run the experiments, and collect data for AI systems will be in a privileged position. We are already seeing this trend playing out with frontier labs racing to stay just a few months ahead of developers who can create comparable performance for a fraction of the capital costs.

But the data to train these models? That's becoming harder to source. Google recently paid $10M for emails and spreadsheets from the defunct Spirit Airlines for training.

I see no reason why we won't see this trend play out in the physical and biological sciences as well. An elite few will do most of the physical work.

Deep tech entrepreneurship becomes even more exclusive

As physical science and engineering become more exclusive, deep tech entrepreneurship will become even more exclusive. Deep tech already has a have-and-have-not situation. Research hubs like Boston and the Bay Area consistently produce more deep-tech startups and entrepreneurs than elsewhere.

I believe this disparity will only grow, and I worry about it. As US governments at the federal, state, and local level face an aging population, fiscal pressures will make it harder to invest in research infrastructure. Politicians will be tempted by AI-centric research, which will require less capital expenditure. It's an easy trade-off to make when facing a budget shortfall.

Unfortunately, this will stunt the development of scientists and engineers in their region, giving the major hubs an advantage in tech development and subsequent entrepreneurship.

It's the trap of AI-centric research.

Uncertainty is the only thing that's guaranteed

I'll admit that I could be way off base. This is one possible scenario. We will likely see several years, maybe decades, of uncertainty and volatility in science and engineering.

But there is ample reason to believe that this is where we are headed. AI can easily drive greater inequality; science and engineering studies will not be immune.