Philosophy Thursday: The Artificial Intelligence Problem

Since my Philosophy class for today was cancelled, I'll take a few minutes (about an hour) to share some Philosophy with you.

Today's topic is AI or Artificial Intelligence.

AI is the basic idea that computers, with sufficient capability and programming, can have minds. Basically, that they might be able to think.

I want to start (where to start, where to start) with some of the basic assumptions we have about computers, and how they'd be if they were capable of human-like thinking.

• 1. They would have personalities like that of a human.
• 2. They'd be capable of conversation.
• 3. They'd operate in a way that would seem, to the outsider, indistinguishable from the way humans, and their brains, operate.
• 4. They'd be capable of understanding.
• 5. They'd be capable of learning.
• 6. They'd have emotions.

There are plenty more. Before I go any farther, I just want to clarify that I'm referring to Strong AI versus Weak - Strong is the kind that allows machines to become sapient. As in self-aware. The Cartesian "I think, therefore I am."

Now, here's the beginning.

Is it possible to design a computer being that was an example of Strong AI? One might argue that with the proper programming, yes. A program that was self-writing. A program that allowed for constant revision and monitoring by the program itself. However, in this case, it is not the computer that would be AI, it would be the program. A sentient code. It is an abstract - something floating in space, attached to nothing. Of course, the code itself isn't complete, not until it's running. And it needs a machine to run it. A machine that will house it, run it, and give the program various opportunities for output - and, more importantly, inputs. Blinking lights, sounds, movements facilitated by the ability to observe. Without outputs or inputs, the program is blind, deaf, and mute. It is essentially aware of nothing - and as so, its ability to edit itself is moot. Why should it, if there's no new information to edit into it? But I digress.

It has been argued that a computer becomes sentient when it is conversationally indistinguishable from a human. There is a test, called the Turing Test, which basically states that if you can't tell if the converser is a human or not, then the converser has a mind. This test is fundamentally flawed, however, as it's easy to imagine that a computer mind would be far different from a human mind, and would know that it was a computer. If you asked it "What are you?" and it responded "A human," you'd know that while it had passed the Turing Test, it was a program on a computer. How? Because self-categorizing is one of the most important criteria for labeling a being "sentient." If it was a sentient computer, it would tell you what it really was. It would know.

So here we are. How do we create Strong AI? Must we develop it completely on our own? I don't think so. I think, that with the right approach, Strong AI could be achieved fairly quickly. But the approach many are taking is the opposite of the one I think would have the most success.

Look at the adult human mind. Does it start off this way? Are the mental states involved completely reducable to equations and digitations? If you wanted to create a man, would you try to create his brain so that it could do all of the things an adult human mind can do? Or would you create a more general brain, one with possibilities for growth and learning? An infant mind.

Create a machine with (approximately) infinite capacity. Program it to observe, to listen, to feel, to experience. Program it to assimilate those experiences. Program it to respond to those experiences, and react, to ask questions about those experiences. Program it to LEARN.

That's all you have to do.

Then, teach it. Teach it like it was a human. Teach it human reasoning, and morals, and ethics. Teach it different ways of thinking.

You can't program a complete mind. You have to program one incompletely first, and then help it to complete itself.

That's my take on it.

Now, onto the challenging of the assumptions I started with.

1-Personalities. Would computers have personalities? I don't think so. I don't think the computer would act like a person. It would act like a computer that had learned all the same things as a person, only without the experience of actually being one. It would assimilate its experiences differently than a human would, and would react to them differently. A human has certain needs, most if not all of which the computer does not. A computer does not need water, oxygen, food, blood, the ability to excrete, sex, or to be cared for. Why should it act like it does? If it does, then it is either pretending to be human for the human's sake or it is not possessing of a mind. Personalities? No. Computalities? Sure.

2-Capable of conversation. Why not? Assuming you taught the thing your language and didn't just program it. Why, Adam, couldn't I have just programmed it? Because, by simply programming it, you remove all of the meaningful associations created during the learning process that allow the computer to actually UNDERSTAND what the words it is using MEAN and how they are used within context.

3-Indistinguishable from Humans. Um. I don't think so. A computer mind would have a different set of capabilities from a human mind. A computer mind could calculate billions of numbers at the same time, could read a book in an instant, could pull up every past memory with a detail that would astound anyone currently considering themselves "alive." A computer mind would not have emotions, at least not ones that we'd understand as human. A computer mind would not be able to experience the world as a human, as it knows it is not a human, and we know that it is not. So, on this point, the answer is no.

4-Capable of Understanding. Yes, but from a computer mind's point of view. You cannot expect human comprehension from a non-human entity.

5-Capable of Learning. They'd better be! Without learning, there's no Strong AI!

6- Capable of Emotions. Um. Again, I'm going to have to go with no. Of course, I'm relating this to human emotions. There is no situation I can think of where a computer is going to need something human emotions are used for. There are no hormones released in a computer, and there are no chemicals that would create certain emotions like love, joy, or sadness. What does a computer need that emotions like anger, happiness or sadness are going to give it? Now, it is entirely possible that the computer would have emotions. But they'd be entirely of the computer mind, and we as humans without the experience of growing up with computer minds would be pretty much out of the loop (to make a slight programming pun) of comprehension.

***

Basically, I'm trying to get this point across - Strong AI, if it existed, would not be the same as HI (Human Intelligence.) It could never be. Just as I will never be able to relate the experience of being a dog or a cat from a dog or cat's point of view, the computer mind would not be able to relate the experience of being a human from a human's point of view. It would have different values, morals, ethics - based on the experience of being a computer. It would most likely not care for us, except for when it needed something.

And that's why, if Strong AI were developed tomorrow, I would be very close to being very afraid.