Will Robots Become Conscious?

  • By Greg Kestin
  • Posted 11.02.17
  • NOVA

What makes a brain or a machine conscious? Will robots become more conscious than we are? This theory, which can actually calculate consciousness, called IIT, is beginning to provide some answers. Find out in this episode of What the Physics?!

Close
Running Time: 06:13

Transcript

How to Build the Universe

Aug. 17, 2017

Greg Kestin: Consciousness.

We seem obsessed with it.

Evelyn Caster: We can upload his consciousness.

Greg Kestin: And consciousness in machines...

It seems imminent.

Some people worry that they could actually be a threat.

Maybe not today.

So what can science, real, hard, quantitative theories of physics and information, tell us about consciousness?

Could machines surpass us, becoming more conscious then we are?

And even develop motives of their own?

There are a number of theories on the physics of consciousness, and stay tuned for those in future episodes. But there is one theory that’s being developed which can already calculate the amount of consciousness in machines, and before long, in us.

Isn’t this a beautiful image? OK, maybe it’s more beautiful in this familiar form. In any moment, your brain puts together all the sounds, shapes, colors... all the signals it receives, into one unified conscious experience. That integration is due to connections in your brain, intricate connections, that form loops, and those loops allow you to be self aware.

Integration is the key principle to understanding the integrated information theory of consciousness. This theory, called IIT, for short, says that the more interconnected or integrated a brain or machine is, the more conscious it is. We usually think of consciousness as either being on or off, Like, “is she Conscious?” Or “He’s out cold!” But in IIT, consciousness is a spectrum, a continuum. Like, this iguana, it’s conscious, not as conscious as you, but more conscious then, say, a thermostat, or this guy. 

So, IIT provides a consciousness meter, a mathematical equation that, when applied to a brain or a machine, gives you a number for how much consciousness it has. That number’s denoted by the Greek, phi, because we physicists love our Greek letters.

There are now hospitals actually using these ideas, coupled with EEG and transcranial magnetic stimulation to determine how conscious comatose patients are. In a highly conscious interconnected mind signals can travel across the brain, while in a less conscious mind, signals are isolated to a relatively small pocket of the brain.

Alan Turing’s imitation game considered a machine that could convince you it’s a person. 

Alan Turing: Could machines ever think as human beings do? Most people say not.

Greg Kestin: So what’s the difference between the experience of a thinking machine and us? A functionalist would say that if something could be programmed to act like a conscious person, then it’s conscious. For example, by installing the right program. 

IIT refutes that.

Siri are you conscious?

Siri: Are you?

Greg Kestin: Why is she avoiding the question?

In IIT, for something to be as conscious as a person, it would need circuitry that’s massively interconnected, because humans have about 100 billion neurons, and a single neuron could connect to thousands of others.

Humans have a value of phi that’s so large it hasn’t even been calculated.

Here’s the circuitry of a simple computer. A single simple loop, like the clock here has a phi of 1. But the computer as a whole has these separate sections, with no loops connecting them all, so it has a phi of zero. That’s right, the clock has a little bit of consciousness, but the computer as a whole has no consciousness. So that means that your computer, or the phone Siri’s running on, is pretty much not conscious. Even neural networks, which have many interconnected computer processors, aren’t as conscious as we are, but we’ll get back to that.

But what about Delores from Westworld or Sonny from I, Robot? If you build a new type of machine, on that’s really interconnected like our brain, that’s called neuromorphic architecture, and that could be as conscious as we are. If we do build robots with neuromorphic architecture, beware because at first they might seem harmless, but if they evolve the way that simulations imply, then from generation to generation, they’ll become more and more conscious, surpassing us, and possibly developing their own goals. On the other hand, if robots are built like our current computers, or like Siri, then they’re just behavioral zombies. They act like people, but they’re not people – it’s all action and no internal experience.

So how do you know your friend isn’t a philosophical zombie like Siri, because we all have that friend?

Ari Daniel: Hello Greg, I have six updates for you. Would you like those now, in an hour, or should I remind you tomorrow? Remind me tomorrow.

Greg Kestin: Here’s a pretty good test. It’s a more sophisticated, less annoying version of Captcha. Show your friend a picture of something like this. Where something’s a little bit off. If they have an integrated mind, then they’ll be able to put the pieces together and know the wine shouldn’t be floating like that. But if it’s a computer program, it’s going to be a little bit to specific, and it can’t put all the pieces together.

Ari Daniel: Are you going to drink that?

Greg Kestin: Of course someone could build a neural network to specifically look for these kinds of problematic images, so the only way you really know is by looking in your friend’s brain.

So, why do we really need a test for consciousness? Well, one reason is that there’s a law related to the stock market that you can’t buy or sell shares to intentionally manipulate the market. But the will to create new intentions is considered a feature of consciousness. Neural networks that are currently being used for trading can learn on their own, and will likely learn this trick of trading to manipulate the market, so maybe it’s time to measure their phi, because if they have consciousness, we might have to make some more room in our prisons.

Have you ever wanted to get the universe in a box? Check out our other videos. Let me know in the comments how conscious you’re feeling today. Of course, I’m talking to the other machines.

Credits

PRODUCTION CREDITS:

Host, Writer, Producer
Greg Kestin
Animation & Compositing
Marquee Productions
Greg Kestin
Scientific Consultant
Masafumi Oizumi
Larissa Albantakis
Christof Koch
Contributing Writer
Lissy Herman
Filming, Writing, & Editing Contributions from
Samia Bouzid
David Goodliffe
Image Contributions from
Drew Ganon
Editorial Input
Julia Cort
Lauren Aguirre
Anna Rotschild
Special Thanks
Ari Daniel
Allison Eck
Kristine Allington
Lauren Miller
Eric Brass
Gil Kaplan
Fernando Becerra
Funding provided by
Foundational Questions Institute

MUSIC:

provided by APM

Sound Effects

Freesound.org

Additional footage:

Pond5
Rise of the Robots
Find Credits at http://www.pbs.org/wgbh/nova/tech/rise-of-the-robots.html

Related Links