Elon Musk Is Building the Wrong Robot
This week the internet watched Tesla's Optimus walk into a Los Angeles mansion with Elon Musk, carry in the grocery bags, and set them on the kitchen island, trading a couple of words with him along the way. Nine million views so far. My guess is most of those millions were applauding the wrong thing.
Look: the robot itself didn't do anything remarkable. Carrying a bag from a trunk to a counter is a solved problem; warehouse machines have done harder work for a decade without anyone filming them. What drew the applause was how Optimus looked while doing it. Two legs, two arms, a head where a head belongs, a walk that reads as human.
People were cheering the resemblance.
And the resemblance deserves a closer look than it's getting, because the actual engineering case for a humanoid robot covers a lot less of it than Tesla built into Optimus. I'll lay that case out at full strength first, and then try to find the holes in it — arguing with a straw man is a waste of everyone's time, and Musk's position is stronger than his critics tend to admit.
Musk's main argument
Every doorway you've ever walked through was sized for a human being. Same for every staircase, handle, light switch, washing machine, and shelf. Our civilization only pretends to be a neutral environment. It's really a user interface, and the standard user has two arms, two legs, nimble fingers, and eyes about 1.6 meters off the ground.
A wheeled robot can't climb your stairs. A robot dog can't reach the top shelf or drive your car. A specialized kitchen arm will never take out the trash. Hence the solution: the humanoid form is simply compatibility with the world we already have — one machine that operates everything built for people, without rebuilding any of it.
Behind that main argument sit two smaller ones.
First, data. The internet holds an effectively unlimited archive of video showing humans doing things with human bodies, and a robot shaped like us can learn from those demonstrations almost directly. A robot of any other shape would need a dataset that doesn't exist.
Second, manufacturing. A single universal chassis produced by the million is cheaper than ten specialized machines produced by the thousand.
I agree with all of it. I've been building products for twenty years, and I mean this without any irony: grant the premises, and the humanoid bet holds.
The fine print
But hardly anyone takes that argument and checks it one level deeper.
Stairs require legs, or at least a body that can negotiate steps. Top shelves require reach and articulated hands. Driving requires limbs in roughly human positions. Learning from human video requires joints that map onto ours. But every item on that list is about kinematics — geometry of movement, reach, degrees of freedom.
But there's no face on that list. No head shaped like ours. No gait tuned to look relaxed rather than to save power. No five fingers where three would grip better. And none of the performance Optimus puts on with every delivery: the nod, the palm-up presentation of the bag, the body language.
So the engineering argument buys you a machine that can move through a human world. But what Musk built is a machine that performs humanness at the person who owns it. Those are different products. One is a tool. The other is aimed at its owner.
Paying extra for the theater
Walking on two legs is one of the most expensive problems in robotics. A biped burns compute — and a pile of energy — just staying upright, while a quadruped gets stability nearly for free. Two legs cost more energy, more actuators, more ways to fail. Engineers eat those costs where the payoff is real: stairs, narrow corridors, car seats. OK, fine.
Now watch the demo again with that in mind (I went back five times myself). The pace and the manner of the walk aren't there to save energy or time — they're tuned to be pleasant to watch. The hands are as human-looking as they could make them, though simpler and cheaper grippers would hold a grocery bag just fine. Wherever the engineering answer and the theatrical answer part ways, Optimus takes the theatrical path. And the theatrical path costs more every time.
When a company keeps paying a premium for something its stated use case doesn't need, at some point you have to conclude the stated use case isn't the whole product. The resemblance isn't overhead they forgot to trim. It's what they're selling.
So who is the face for?
Most critiques of humanoid robots stop at "it's inefficient." But inefficiency is just a cost, and costs get amortized by scale — of production and of operation. The more interesting question is: who is all that extra humanness working for?
Certainly not for the robot — laundry gets folded just fine without a face. The face works on us. Human brains respond to the human form with a speed and depth that skips deliberation entirely: we trust the human-shaped thing sooner, forgive it faster, grant it intentions and competence and moral weight it hasn't earned. That circuitry is ancient, involuntary — and, what matters most, already installed in every potential customer.
I've spent two decades designing onboardings, and I know what it costs to earn a new user's trust — it's the most expensive conversion there is. A human shape gets it for free, by reflex, before the machine has done a single useful thing. Design aimed at the reflex instead of the task isn't engineering anymore. It's persuasion. A dark pattern.
I don't think anyone at Tesla sat down and designed a manipulation machine on purpose (then again — why not?). Nobody had to. The market picks resemblance and applauds it, so resemblance gets built. Which is exactly why it's worth talking about.
An honest robot
Years of watching people grow attached to products have led me to a design principle here, and this essay series is my attempt to argue it out loud. I call it ontological honesty:
The principle that a machine must be recognizable as a machine at the level of reflex, not disclaimer: human kinematics where the world demands it, machine semiotics everywhere else. A robot may have hands like ours — but it should not wear a face like ours.
Give the robot legs if stairs demand legs. Give it arms that reach the top shelf and hands that hold a grocery bag. That much the world dictates — and there, Musk's argument is ironclad. But the face, the gait, the little nods — none of that is there for compatibility with our world. That's a key cut for the oldest lock in our heads. And we need to notice that in time, before there's one of these in every home.
The strangest part of the story isn't what robots are becoming. It's how easily a person can be read: all it takes is a body shaped like ours, walked — smiling, polite, warm — into our kitchen.
Next essay: why the anthropomorphism was never in the robot to begin with — odds are you've named a vacuum cleaner or a car yourself. The reflex lives in us; design just picks the lock.
This is Essay 01 of Honest Machines, a series on anthropomorphism in robotics — what human-shaped machines cost in engineering, what they do to our psychology, and how to design them honestly. New essays every few days: RSS · LinkedIn · léelo en español.