Porter dares rival labs to race a 5-year-old at CoRL

Brad Porter wants to put a robot next to a five-year-old child at CoRL next week and see which one picks up more stuff. He made the offer on X, replying to Jitendra Malik, with one condition. "If we can find at least one other physical AI company willing to do this at CoRL next week, we'll bring our rig," Porter wrote. "Who's in?"

Brad Porter

@BradPorter_

X

If we can find at least one other physical AI company willing to do this at CoRL next week, we'll bring our rig.

So far, nobody in the sources has said yes. This is a public dare, not a scheduled event.

What the Malik test actually is

The format comes from Malik, who proposed what he called the "5 year old paired comparison test". Set up tables at a conference. Put a robot at one, a five-year-old child at another. A neutral person picks 100 manipulation tasks using only household objects, the kind found in a typical American home, and both sides run them in parallel. Pick up a coin. Pick up a book. Untwist a bottle top. Insert a plug into a matching socket. Rotate one face of a Rubik's cube.

Malik's reason for proposing it was blunt. He wrote that he sees an announcement or demo on X every week implying manipulation is solved, and the only reason he does not believe it is that somebody else already solved manipulation the week before.

Why Porter likes it

Porter's argument is that the test resists gaming. Replying in a thread that pulled in @DrJimFan, @chris_j_paxton, @ericjang11 and others, he compared it to the Turing test for text-based intelligence and said the Malik test is the real benchmark for physical intelligence because "Any other benchmark will be optimized."

Brad Porter

@BradPorter_

X

Any other benchmark will be optimized.

That is the live complaint in manipulation right now. Benchmarks with fixed task lists and fixed objects get tuned until the number goes up, and the number stops meaning anything about open world performance. A neutral person choosing tasks on the day, against a child who has never trained on them, removes that lever.

Porter has been making noise about manipulation timelines for a while. He recently said general dexterous manipulation is two to three quarters away, which makes a side-by-side with a five-year-old an unusually exposed thing to volunteer for.

Shown versus claimed, nothing has been shown here. No company has publicly taken Porter up on it, no tables are booked, and the task list does not exist yet. If a second lab says yes before CoRL, it becomes the most honest manipulation demo of the year. If not, it stays a tweet.

Jitendra MALIK

@JitendraMalikCV

X

Until your robot can do all the "open world manipulation" that a 5 year old kid can, some humility is in order.