Header 1

Our future, our universe, and other weighty topics

Thursday, December 5, 2013

The Only Way to Verify Robot Self-Consciousness Would Be to Become a Robot

Mankind is getting better and better at making robots. Before many decades we should be able to create robots that can converse with us with what seems like a large amount of intelligence. But will such robots be self-conscious? That, as they used to say decades ago, is the $64,000 question.

In this discussion by self-consciousness I mean an inner self, an inner life, an inner stream of consciousness similar to the inner life led by a human. I do not simply mean the ability of a robot to recall or process some information concerning the robot or its previous interactions. A robot that can describe itself and list all of its previous decisions or visual inputs is not at all self-conscious unless it has some kind of inner self and inner life like we have.

Some people think that once a robot has the ability to pass for a human, it must have self-consciousness. The underlying reasoning may go something like this:
  1. We are self-conscious
  2. If robots could talk as well as we can, their minds would be at least equal to ours
  3. Therefore such robots would also be self-conscious
But such reasoning is flawed. Self-consciousness is the crown jewel of the human mind. The fact that a robot might have other information processing capabilities equal to ours does not necessarily mean that it would have self-consciousness. It is important to distinguish between two different things: intelligence and self-consciousness. A robot can have a high degree of intelligence without any self-consciousness at all.

If you doubt this, consider an ordinary computer program. The average complex computer program has a degree of intelligence, a certain amount of smarts. By intelligence I simply mean the ability to process inputs in some rational way, producing some reasonable output. But there is not any computer program that has self-consciousness in the way that humans have. From a self-consciousness standpoint, every computer program ever written is as insentient as a stone. Even programs that run on supercomputers and perform millions of calculations per second have not even as much self-consciousness as you have in your sleep.

You know yourself to be self-conscious, but how can you be sure that the people you see every day are also self-conscious (rather than just phantoms or illusions with no real inner selves)? Perhaps you cannot be 100% sure that they are self-conscious, but you have a good rational basis for inferring that they are. The people you see with your eyes are other beings who look pretty much like you, who had an origin similar to yours, and who are made of the same materials that you are made of. So it is quite logical for you to infer that they are self-conscious just like you are. But such reasoning completely fails in the case of robots. Robots may one day look like you, but they will always be beings that originated in a completely different way than you originated, and also are made of materials vastly different from the materials in your flesh (plastic and steel being very different from protoplasm and bone). So it would seem that we might never have an adequate basis for inferring that robots are self-conscious.

Let us imagine an advanced robot which looks like a human, which moves around like a human, and can talk very well with a human being. The robot is capable of sparkling conversations. We can imagine three different possibilities:

  1. The robot has no self-consciousness at all. Destroying the robot is no more of a murder than destroying a car. Any impression that the robot may give that it is self-conscious is a mere illusion, the product of clever programming.
  2. The robot is fully self-conscious. It is a person with an inner life and inner self just like you and I. Destroying the robot would be as much of a murder as killing a person.
  3. The robot just has a dim trace of self-consciousness. Imagine your self-consciousness during one of those very rare times when you lay asleep in bed, but realized (just before waking up) that you were sleeping. At that point you had a minimal dreamy fragment of your normal waking self-consciousness. It could be that the robot has that type of minimal self-consciousness.

The problem is that we would have no way of ever knowing which one of these possibilities is the truth. We can't find the answer by simply asking the robot. Asked about his self-consciousness, robots will give whatever answer they have been programmed to give, or they will say they don't know, or they will merely guess.

To ask a robot whether it is self-conscious is to ask it basically: do you have an inner life like my inner life? But the robot will never have lived a human life, so it won't be able to answer, unless it guesses at something it doesn't really know about.

But there is one way that you might be able to verify that an advanced robot of the future really is self-conscious. You might verify such a thing if you actually became a robot.

This possibility has been discussed in the context of mind uploading. One day there might be some machine that scans your brain, and somehow extracts all your memories and transfers them to a computer inside a robot.

Now imagine you paid for such an operation to be performed. You go to a medical center and are put under anesthesia. You then wake up with the body of a robot. You move your robot arms, and through your robot eyes you see your former human body (possibly dead) lying on an operating table. You also have all your memories, and your self-consciousness.

Then you would truly know that robots can be self-conscious just like humans. But maddeningly, you would never be able to prove this to a human who had not undergone such an operation. Any human who had not undergone such an operation might still maintain that you were a robot without real self-consciousness, who had merely been uploaded with the memories of a human. You would know that this was untrue, but you would be unable to prove it to others.

But such an uploading may be impossible to ever achieve. So humans may never know whether the sophisticated human-like robots they create are truly self-conscious entities with inner lives, or merely souped-up adding machines who fool us into thinking they have selves.