杏吧原创

Who are you calling a zombie? (Image: Vincent Fournier/Gallerystock)

IN A robotics lab on the eastern bank of the Hudson River, New York, three small humanoid robots have a conundrum to solve.

They are told that two of them have been given a 鈥渄umbing pill鈥� that stops them talking. In reality the push of a button has silenced them, but none of them knows which one is still able to speak. That鈥檚 what they have to work out.

Unable to solve the problem, the robots all attempt to say 鈥淚 don鈥檛 know鈥�. But only one of them makes any noise. Hearing its own robotic voice, it understands that it cannot have been silenced. 鈥淪orry, I know now! I was able to prove that I was not given a dumbing pill,鈥� it says. It then writes a formal mathematical proof and saves it to its memory to prove it has understood.

This is the first time a robot has passed a classic test called the wise-men puzzle. It sounds like a simple test and it is, hardly scaling the foothills of consciousness. But showing that robots 鈥� in this case, off-the-shelf Nao models 鈥� can tackle logical puzzles requiring an element of self-awareness is an important step towards building machines that understand their place in the world.

Selmer Bringsjord of Rensselaer Polytechnic Institute in New York, who ran the test, says that by passing many tests of this kind 鈥� however narrow 鈥� robots will build up a repertoire of abilities that start to become useful. Instead of agonising over whether machines can ever be conscious like humans, he aims to demonstrate specific, limited examples of consciousness.

鈥淭hey try to find some interesting philosophical problem, then engineer a robot that can solve that problem,鈥� says John Sullins, a philosopher of technology at Sonoma State University. 鈥淭hey鈥檙e barking up the right tree.鈥�

The work, which will be presented at the RO-MAN conference in Kobe, Japan, next month, highlights the murky waters of artificial consciousness. The wise-men test requires some very human traits.

The robots must be able to listen to and understand the question 鈥渨hich pill did you receive?鈥�, as asked by a human. They must then hear their own voice saying 鈥淚 don鈥檛 know鈥� and understand that it was they that said it, connecting that with the idea that they did not receive a silencing pill.

Bringsjord鈥檚 robots may appear conscious in this specific case, assessing their own state and coming to a conclusion. But the broader, deeper intelligence that we humans have is completely missing. The Nao robots can pass the wise-man test but wouldn鈥檛 have a hope of recognising their own feet.

Bringsjord says one reason robots can鈥檛 have broader consciousness is that they just can鈥檛 crunch enough data. Even though cameras can capture more data about a scene than the human eye, roboticists are at a loss as to how to stitch all that information together to build a cohesive picture of the world.

The test also shines light on what it means for humans to be conscious. What robots can never have, which humans have, argues Bringsjord, is phenomenological consciousness: 鈥渢he first-hand experience of conscious thought鈥�, as Justin Hart of the University of British Columbia in Vancouver, Canada, puts it. It represents the subtle difference between actually experiencing a sunrise and merely having visual cortex neurons firing in a way that represents a sunrise.

Without it, robots are mere 鈥減hilosophical zombies鈥�, capable of emulating consciousness but never truly possessing it. 鈥淭he idea requires that there is something beyond the physical mechanisms of thought that experiences the sunrise, which philosophical zombies would lack,鈥� Hart says.

This article appeared in print under the headline 鈥淚 know it鈥檚 me talking鈥�