The Uncanny Valley Isn't a Law of Nature. It's a Design Decision

You know the feeling, even if you don’t know its name. A face that’s almost right, but something’s off (the eyes a fraction too still, the smile a beat too late) and a cold little wave of unease runs up your spine. Not fear exactly. Something stranger: revulsion at the almost.

That feeling has a name, and a birthplace. In 1970, a Japanese roboticist named Masahiro Mori described what he called the “uncanny valley”: the idea that as a robot becomes more humanlike, our affection for it rises, until it gets too close to human without quite arriving, at which point warmth collapses into dread. Plot it on a graph and there’s a sudden, sickening dip right before full humanness. A valley.

For half a century we’ve treated Mori’s valley as a kind of natural law, an inescapable feature of the human mind that dooms any near-human machine to creep us out. And now, with humanoid robots arriving in force, that assumption matters more than ever. It’s also wrong.

The robots are actually coming this time

First, the context. After decades of humanoid robots being a research curiosity (impressive lab demos that never left the lab) 2025 and 2026 have turned into something like a stampede.

Boston Dynamics rebuilt its famous Atlas as an all-electric machine and pointed it at real work. Tesla keeps pushing Optimus, though that one deserves a footnote in an article complaining about hype: the company has never published a production count, it missed its own 2025 target of 5,000 units by a wide margin, and as of the middle of 2026 it was still installing production lines. Agility Robotics’ Digit is the one with a commercial record you can check, working under paid contracts with GXO, Toyota, Mercado Libre, and Schaeffler. Figure sits in between: a completed factory pilot and a demonstration, rather than a standing job. So the two-legged, two-armed, roughly-person-shaped robot is no longer only a demo. It’s becoming a product category, and some of it is already in workplaces alongside people who never signed up to share a shift with a machine that walks like them.

Which drags the uncanny valley out of the seminar room and into the break room. If these things unsettle the humans around them, they fail, no matter how capable they are.

What Mori actually said, and what we forgot

The part everyone forgets is that Mori’s valley was never a wall. It was a map, and like any map, it shows you both the ditch and the roads around it.

Read his idea carefully and it’s really a warning about a specific mistake: chasing realism. The unease spikes precisely when a machine reaches for full human likeness and misses. A robot that’s obviously, comfortably mechanical (think of the friendly, cartoonish machines people happily coo over) sits safely on the near slope of the valley, before the drop. It’s the ones straining to pass as human, and failing in small wrong ways, that fall in.

That reframes everything. The valley isn’t a fact you’re stuck with. It’s a consequence of a design choice: how human are you going to try to look? Aim for photorealistic skin and lifelike eyes, and you’re betting you can clear the far wall perfectly, and any imperfection dumps you into the ditch. Aim for stylised, honest, obviously-a-robot, and you never approach the edge at all.

Mori made one more point that the popular retelling drops, and it lands squarely on the machines arriving now: movement amplifies the whole curve. Both the peaks and the valleys get steeper once the thing starts to move, because motion is a second set of signals to get wrong. A slightly odd face at rest is uncomfortable. The same face walking towards you is far worse. For a robot that spends eight hours a day moving around people, that isn’t a footnote in Mori’s essay. It’s the part that decides everything.

How solid is the valley, actually

Since the argument here is that Mori’s curve should not be treated as law, it would be dishonest to then treat it as fact. So, plainly: the experimental record is much thinner than the fame of the idea. Researchers who have gone looking for that clean valley shape, with warmth rising, plunging into a ditch, then climbing out the other side, have come back with weak and inconsistent results. It’s a memorable drawing rather than a measured constant.

What does survive testing is the narrower claim sitting underneath it. Mismatch unsettles people. A realistic face with eyes that move wrongly, a lifelike hand with mechanical timing, a warm voice attached to a fixed stare: when the signals disagree about what sort of thing you are looking at, people react badly, and that finding is solid. Which happens to strengthen the case, because mismatch is precisely what a designer controls.

The industry finally took the hint

Watch the current wave of humanoids and you’ll notice something: most of them are not trying to look like people. They have visored, screen-like faces instead of synthetic flesh. Smooth, panelled bodies that read as “sophisticated appliance,” not “artificial human.” Movement that’s competent and legible rather than eerily lifelike.

That’s not a failure of ambition. It’s the lesson learned. The companies serious about putting robots next to humans, day in and day out, have largely decided not to fight their way across the uncanny valley, because there’s no prize on the other side worth the risk of the fall. A warehouse robot doesn’t need a convincing human face. It needs to be capable, predictable, and pleasant to be near. Friendly-machine beats fake-human every time.

The places still chasing perfect human likeness tend to be the ones where the illusion itself is the product (film effects, certain kinds of companion robots) and those are exactly the places that keep tumbling into the valley and unnerving people, because they’re taking the one path Mori warned about.

The real takeaway

There’s a lovely bit of intellectual freedom in this. We tend to think of our gut reactions as immovable: the uncanny shudder as something done to us by biology. But the shudder is a response to a choice someone else made about how a machine should look. Change the choice, and the shudder never comes.

As the humanoids arrive, the ones that end up in our workplaces and eventually our homes won’t be the ones that most convincingly fake being human. They’ll be the ones designed, deliberately, to sidestep a fifty-year-old valley that was never a law in the first place, just a warning we’re finally learning to heed.