Like realtors' photos of beautiful houses on sunny days, a shiny YouTube video of robots working flawlessly should be taken with a grain of salt. The robot working at its best, in ideal circumstances, is really not representative of a typical day.
Of course, robot-makers want to show off their work in its best light, just as you'd want to arrive for a job interviewed washed, rested and focused. It's not deceptive to show your machine at its best, or to demonstrate that your concept actually can work. But we viewers just need to keep in mind that robot reality is different.
You should also take the opposite sort of video with an equal grain of salt. I mean the kind of pratfall and stumble video in which the robot fails completely, like this famous interview with Hanson Robotics Sophia (at the 12 second mark)
or this compilation of burrito bot fails (some of which inspire humans to help, others of which seem to make people feel what I call maschinenschadenfreude -- pleasure in robot failure.
These cherry-picked examples of flop are the mirror image of the videos that imply that all the problems of robots dealing with the real world are already solved.
Much more true to life is this video, from a recent CNBC broadcast, in which the network's Singapore correspondent, Sri Jegarajah, meets two robots — a WalkerX and then a Walker S from UBTech, the humanoid maker based in Shenzhen.
It's kind of awkward. But it's not a fiasco. The appeal of this video, to me, anyway, is that it comes out at neither extreme. The robots sort of screw up. But they don't keel over in a heap. It's not smooth or fast, but after a few mistimed actions, the first robot, the WalkerX, manages to say some welcoming words, wave, and shake Jegarajah's hand. The correspondent has to help a bit, which he gamely does -- as people tend to with humanoid robots. That is, he plays along with the act, using broad, slow gestures and a loud voice, as you might with a young child, or on a stage if you'd been dragged onstage into someone's improv showcase.
The second robot, the Walker S, is slow and fumbling. But it manages to give Jegarajah the right fruit off a counter. Except for one impatient look into the camera, our correspondent stays game. ("And he is going for the apple. And unfortunately he missed it. But let's try again, shall we?")
While Jegarajah and UBTech's Jimmy Zhang wait for Walker to pull itself together, Zhang points out that the first wave of the company's robots will be performing repetitive actions in factories (where, he need not add, they won't need to hand out fruit or shake hands). He then goes on to make the claim that future models will be able to read emotions and help around the house. Clearly, though, today's intro pitch doesn't need the robots to do that now. Also clearly, more work is needed (and is being done -- just as Figure and 1X are working with OpenAI to make robots more adept at human language, UBTech is in a similar partnership with Baidu and its powerful ERNIE large language model.)
In other words, this is a video that reflects robot reality: We see tech that doesn't always work, timing of human interactions a little off, lots of patience and pretend-play skills needed on the part of the human. But by no means a total flop. It would be great to see more robot videos as real as this one.
What's Your Problem, West Virginia?
This company recently surveyed nearly 2,000 Americans about how they act with AI. And, surprisingly, they found regional differences. According to their self-descriptions, people who live in West Virginia, Michigan and Connecticut were ruder to AI agents than were other states' residents. (New Mexico, Florida and Maryland residents were nicest to the machines.)
Why on Earth would this be? Beats me. (Obviously, people's reports about their own behavior aren't nearly as reliable as observations by researchers, but it's still interesting that even people's account of themselves with AI should differ from place to place.)
One other stat caught my eye here. For all that research shows people love their Roombas and care about Siri's feelings and so on, people in this survey say being nice to AI is meaningless. That is, eight out of ten respondents said they were polite to AI out of habit. So if you ask GPT-4o to "please" explain inflation to you, it doesn't mean you've blurred lines between human and machine. You were just raised right -- to be unthinkingly polite.
Literary Note
"In the past, when it came to novelty, we had hardly ever seen anything but solutions to problems or answers to questions that were very old, if not age-old. But novelty for us now consists in the unprecedented nature of the questions themselves, and not the solutions, in the way these questions are asked and not the answers. Whence the general impression of powerlessness and incoherence that rules our minds."
So said the French poet Paul Valéry -- in 1935! Which suggests that however novel our tech problems are, the feelings created by rapid change aren't new.