Robot Deepfakes

Product demos, and especially robot product demos, are one of my favorite chaotic challenges. At a typical baseline, giving a live robot demo can feel like juggling flaming swords that are somehow also angry possums. Demos then get harder as a function of technical things like wifi, bluetooth, and batteries, and as a function of human things, like audience size and callous optimism.?

The job is to then make it look easy.

Robot demos are such a spicy third rail that many companies - big companies you’ve heard of and might even like - now commonly fake their demos through clever video editing, and then fumble half apologies when they get called out. These demo deepfakes are almost trivially easy to do, which is interesting given how absolutely earth-shatteringly hard it is to make a robot do anything reliably. Brian Heater at TechCrunch earlier this year penned an entire article about how to spot fake robot demos. It’s a great & informative read, and (I believe) evidence of an interesting ‘valley of acceleration’ for humanoid robots. It’s a valley because this movie magic is making it easier to seem like robots are working, and a descent into any valley provides natural acceleration. What’s coming is the climb out the other side, and friends, there’s work ahead on that climb.

A few days ago, Andra Keay , Managing Director of Silicon Valley Robotics , posted an article in her substack (go subscribe!) in which she takes this idea a bit further. After drawing lines between the ‘hype machine’ and reality, she makes a really good point about how with autonomous mobile robots (AMRs), such has Amazon’s Kiva, most of the development was funded internally and done quietly. Then just a short decade-or-so later </s>, they were everywhere. There was no Valley of Acceleration. They just stalagmited their way into existence, so robust they’re almost boring.?

Keay then implored the robotics community to be honest about any humanoid robot demo videos they post, perhaps because they’re being judged out in the open, and funded (or not) on a vastly different timescale. She makes a clear and valid case for transparency in these videos, though she mostly makes a moral case for what's going to be seen as a business decision. As much as I want her advice to be taken, I can’t see a pathway to that being what happens.?

My expectation, based primarily on my own experience designing and producing successful* robot demos in real-life environments (see aforementioned flaming sword possums et al), is that of the many companies building humanoid robotics, some will keep getting better at fooling us with fakes, and end up fooling themselves into non-existence. The companies that have the discipline to stay honest will be the ones who win, because they’ll be the ones that get to an unaided real-life application first.

Here's the important bit: That probably means a lot of secrecy, and companies that appear to be “missing the boat” will be the only ones actually building a boat. If that ends up being true, climbing out of the valley of acceleration will be slower than Hollywood and Sora would have us believe.

A real life unaided application is really what should be the litmus test here. Who has the confidence in their work to put their humanoid robot in the hands of customer, and walk away??