The Autonomous Future: Humanoids vs. Self-Driving Cars - Part 1: Setting the stage

This is part 1 of a 3 part series of post on the topic. You can find part 2 and part 3 here.

It has been a while since I last posted something substantive here. I've used the past years to immerse myself more in the field of AI, both in the philosophical as well as the technical sense.

Even though there still exists a wide gulf between the technological realities of what AI systems are and can do versus what they are imagined to be by the general public or sell-side stock analysts, it is pretty obvious that massive capital flows flooding into the field, probably in part inspired by stochastic parrots such as ChatGPT, will drive progress at a much faster rate than decades past. At the same time, the field of robotics has also seen steady advances, in part now enabled by more powerful AI systems. What once used to be choreographed, dumb mechanical actuators are increasingly becoming autonomous machines. While there exist strong incentives to make factory automation—the aforementioned choreographed mechanical actuators—easier to configure, self-optimizing, and also safer for their human co-workers, the public's attention, much as with large language models, driven by our inherent need to anthropomorphize inanimate objects, is now firmly focused on humanoid robots. Separately, the idea of self-driving vehicles is also still pursued and publicized by the likes of Tesla's CEO, Elon Musk, who often likes to draw a direct connection to humanoid robots.

While in the case of Musk this seems to be primarily driven by a desire to generate marketing hype for his company, and neither Tesla's robots nor its self-driving technology are at the leading edge of their respective competitive landscapes (no matter what some sell-side analyst with a business degree and vastly over-inflated sense of their own expertise might want you to believe), there exists in deed an interesting connection between the two technologies.

One might consider cars just two-tonne robots on wheels with a quite limited mission - to get their occupants safely from point A to point B, whereas humanoid robots would need to be generalists, capable of making sense of complex challenges in an environment that, unlike the relatively standardized nature of modern roadway infrastructures, was never designed for them. To make this more explicit, a self-driving vehicle has a ready-made infrastructure with clearly defined usage/traffic rules to assist it in its core mission of transporting passengers between destinations. A humanoid robot, on the other hand, may encounter a vastly more complex environment for the simple task of getting from point A to point B. Yes, there will be roads, but walk on them and you might get run over. There are also stairs, elevators (some of which may not work), trains, buses, and bikes. These different modes and means of transport interact in ways that a humble car would never have to contend with. And then there is another thing: navigating whatever world will require means of inference about that world, and in case that inference work is done by neural networks, will quickly require quite substantive compute power. A vehicle provides a much broader envelope within which to power and cool the compute hardware to run said inference, compared to the much more limited resources available to a humanoid.

And yet, I am convinced that useful autonomous humanoids might actually occur before fully autonomous vehicles become commonplace on our streets. The latter of which holds another set of quite dystopian possibilities, by the way, which lie outside the scope of this article, but I strongly suggest you watch this video by "More than bikes": "How Self-Driving Cars will Destroy Cities (and what to do about it)"

That's all for this first part of the blog post, in the next post we'll delve deeper into the specific advantages that humanoids may possess over self-driving vehicles. This will reveal why, despite the additional hurdles, humanoid robots are more likely to achieve usefulness sooner than robo-taxis.