How AI-powered Robots Will Reshape the Physical Economy
musings on trends in robotics, and on deflationary forces
In recent years, surprisingly powerful tech has emerged from AI research labs. For instance, chatGPT, which made headlines last November, has already revolutionized online interactions. Household robots may soon follow suit. Picture a world where intelligent robots fetch you your favorite snack, and clean up after dinner. You might be surprised to learn that impressive prototypes of this tech already exist today!
In this blog, we will explore the economic implications of “general purpose” robots, examine the technical progress made towards their development, and speculate on their societal impact.
The Deflationary Economics of AI + Robots
AI-powered, general purpose robots could marry the deflationary characteristics of software to the physical economy. Let me explain.
Software has long stood apart in its unique ability to drive down costs while exponentially increasing utility. Moore's Law, the prediction that computing power would double every two years, has largely held true, while open-source development has fueled an ever-growing body of free, collective knowledge.
In the world of software, distribution is global and instant. The marginal cost of serving another customer is often close to zero. Winner-takes-all dynamics increasingly rule the day. While in the past every single village may have needed it’s own blacksmith, the world really only needs one Google.
As we move towards a world filled with ubiquitous, capable robots, over-the-air updates could make the distribution of a new capability (for example, cooking chicken tikka masala), global and instant.
On the industrial side of things, consider an army of robots, mining raw materials and working in factories. As articulated by Elon Musk, the limiting factor in producing goods and services may well be reduced to the cost of electricity. The cost of solar panels might become a sort of Moore’s law governing physical production.
The State of Robotic Progress
It’s natural to ask how close this reality might be. Where is Elon’s rumored army of Tesla Bots?
A turning point in my personal perspective on this type of general robot was in November of last year, at a machine learning conference in New Orleans (“NeurIPS”). In sessions with researchers from DeepMind and Apple, it became clear to me that really smart people, across the biggest tech firms, have been working on general, helpful robots for years. Among these researchers, there is little remaining doubt about technical feasibility.
The progress of AI "large language models" like chatGPT serves as a useful reference point. While chatGPT might seem like an overnight success to an uninformed observer, the underlying technology has gone through a public and incremental progression in capability, particularly over the last five years. It was only when GPT reached a certain tipping point that it became so recognizably, broadly useful.
Robots are experiencing a similar evolution. Google's most recent AI can complete 67% of kitchen tasks (such as “fetch me a healthy snack”, or “pick up the napkin out of the toaster”), up from 47% just a year ago. Their newest AI also does a much better job of “generalizing” to new kitchens.
The Robot Cost Curve
Now, let's consider the near-term cost of this type of robot. Boston Dynamics' "robot dogs" cost $75k each. A small, used robot arm from manufacturing supplier Kuka costs $13k. It’s likely that Google’s household robot (pictured below right) could initially fall on the lower end of that price spectrum. It follows that the more advanced humanoid robots, under development at companies like Tesla, Boston Dynamics, and Figure, will probably cost meaningfully more.
Brett Adcock, CEO of the startup Figure, has stated that he plans to pilot 50 of his humanoid robots in a real warehouse setting in 2024. But, in terms of realizing the full potential of these advanced, general purpose robots, he foresees that “this will take 20 or 30 years for us to really build out.”
Increasing the volume of these robots, and bringing down their costs, will be a long term endeavor.
Societal Implications and the Future
Looking ahead towards the remainder of this century, the widespread adoption of AI-powered, general-purpose robots seems poised to have significant consequences for employment, elderly care, wealth distribution, and even human purpose. It’s worth pondering this future with both excitement and trepidation.
Science fiction movies and books provide thought-provoking and rich perspectives on what a robot-filled future might look like. Two of my personal favorites are the animated children’s movie The Mitchells vs the Machines, and the quirky novel Sourdough.
This convergence of the physical and digital worlds extends well beyond robotics. Software touches most aspects of our built environment these days, and the connections are only growing deeper. I often find myself pondering how characteristics of software, such as personalization, might increasingly embed themselves into everything from the clothes that we wear, to the board games we play, to the systems of transportation that move us.
In 1929, Albert Einstein reflected on his development of the theory of relativity, stating that "imagination is more important than knowledge." The emergence of general purpose robots certainly challenges us to envision a future that stretches the limits of our imagination.