Machines like me
Hi there,
So there's this guy named Hod Lipson who literally believes he's invented self-aware robots. But he's no crank; he runs a laboratory at Columbia University. I interviewed him for Quanta.
Technically speaking, he created a one-armed bot that can "self-simulate": it builds up an internal image of itself by moving its body around in the real world (akin to how infants "motor babble" by waving their limbs around to learn how their own bodies work).
But how much can this simplified version of "self-awareness" really tell us about... well, actual self-awareness? Here's a telling detail that didn't make it into the published version of our Q&A.
To gauge the accuracy of the robot's self-simulation, Lipson did an "open loop" test, which is the equivalent of asking a robot to perform a task with its eyes closed (e.g., picking up a ball and placing it into a cup while getting no feedback from the actual world from its sensors). All the bot had to guide its action was its own internal model of itself.
Turns out the robot's self-simulation was, on average, about 4 centimeters off compared to the real world in this open-loop test. (In other words, when it tried to put a ball into a cup "with its eyes closed," it would position its gripper-hand about 4 cm from the cup's actual position.)
That sounded pretty lousy to me, and I said so.
Lipson's reply?
"If you're sitting at a desk right now, look at an object in front of you and then close your eyes and try to reach it with your eyes closed—how far off are you going to be?"
I focused my attention on the top edge of my laptop. I closed my eyes. I reached out to where I imagined the edge would meet my fingers. Nothing. I opened my eyes.
I was about 4 centimeters off.
I just finished a novel about self-aware robots that was un-put-downable. It's called Machines Like Me, and it's by Ian McEwan, the guy who wrote Atonement. It's a perfect example of what Robin Rendle (who writes one of my favorite newsletters) calls "cardigan sci-fi":
Yes – there are technological wonders...But there’s also a lot of cardigans and socks and there’s an open pack of cereal on the table. It’s the absence of technological fetishism...The science or the tools or the spaceships do not sit at the heart of Cardigan sci-fi — it’s all about the people that wear the cardigans instead.
Machines Like Me is like a novel-length episode of Black Mirror—it has the same kind of unpredictable, relationship-driven plot set against a not-overdone background of alternate-world-building (Alan Turing never committed suicide! 🙌), and the final sentences left me feeling vaguely stunned. Maybe the best way to describe it: It may knock your internal self-model off by an extra centimeter or two. I highly recommend it.
[Still working on my sign-off],
J