Frozen in Paris

When I read that Paris Hilton may have herself frozen when she dies, I immediatly thought of two things:

1) Why wait?

and

2) Idiocracy

I wonder what it would be like for scientists in the far flung future to thaw out Ms. Hilton.

AI - Part 3

The concept of passing the Turing Test is easy. In practice, well, it’s a bit more difficult. As a mind exercise try thinking of one thing you would ask to determine if you were talking to a computer or a person. Now try and think of a way that you would do the programming for it.

For instance, you may ask “What is an egg?” Well, that’s relatively easy. Just have a definition for egg handy. But should it be hard-coded, or should it be something searchable then parse it into a more “human” type of response? You’ll find that one question leads to other thoughts. Like, “What are chickens famous for?” Well, laying eggs. How would you work that one? Or, “What’s got a shell on the outside and a yolk on the inside?” Think of a question then think how you would expect a computer imitating a human should answer and you’ll see it get more and more complicated.

Of course, you have to expect the unexpected. This is where outside information is necessary. To cope with things that you may not have thought of at the time but are so common that any human should be able to answer it. If someone were to ask you what rain was you should be able to answer it. But while you’re working on your code, tossing down Twinkies and Jolt, you just may forget to add that in there.

Learning systems are great, and necessary, but you can’t rely on them. If someone asks your program what rain is and it doesn’t know it can ask the person what it is. But what if they lie? Then it’s a dead tip-off to the next person who asks the same question. It can ask for another explanation and that person may tell the truth. How does your program figure out who the filthy liar is?

AI - Part 2

 

Artificial Intelligence

Part II

Just for the fun of it, I went to a couple of places on the web that have a web-based chat bot. I tried six of them, and only one could answer this question: What is 27 + 4? Four gave me nonsense responses, one said it didn’t have a calculator, and one actually gave the correct answer. When I told the one that said it didn’t have a calculator to just use the computer, it decided it was the president of the USA. Hardly convincing stuff.

As another aside, I’ve spent years living with people in households with computers and working in places that were full to the brim with computers. I am frequently asked questions about math and what time it is. I’m the world’s worst mathematician and it astounds me that people can’t take the time to pull up the calculator in any operating system and get a correct answer.

Even if a computer could fool someone into thinking they were a real person it doesn’t mean that it’s intelligent. All it’s doing is matching up its input to a likely response. To borrow John Searle’s argument for a moment, imagine you are stuck in a box with a set of cards that have symbols printed on them and an instruction sheet. Another card drops through a slot in the box you’re in. You compare the symbols on the card with the instruction sheet, which tells you which of your cards you should push out of your box. That done, you sit and wait for the next card to drop in.

Here is what happened. The box you’re in is situated in a shopping mall in Japan. Someone walked up to the information box, wrote down “Where is the bathroom?” in Japanese and then put it in the box. You picked up the card, matched the symbols on the sheet and found out which card to return. The card you sent back out instructed the person to look near Takashimaya’s.

You don’t read Japanese. You have no idea what the question was (or if it was a question) and no idea what the response was. This is, essentially, what a computer is doing. It’s rather unfortunate that many people spend a lot of time functioning like this as well. You know it, we all do it.

Kim: Hi! How are you?
Sally: Fine!

But computers have a larger handicap. If I were to ask you what time it was (even if I’m sitting in front of a clock) you could find out. If I asked you what the weather was like, you could look out the window. If I asked you what was on television you could check that, too. We have the ability to do things, to interact with the world. The computer is not so lucky unless it’s been programmed to do so. Being able to check things like the time, or the weather would be excellent if you wanted your program to pass the Turing Test. Certainly it would be more convincing than a canned response.

Me: What’s the weather like where you are?
AI: It’s foggy out.
Me: It’s been foggy for the last four days, then.
AI: I have a little red pencil box.

Versus

Me: What’s the weather like where you are?
AI (checking weather.com): It’s cloudy, but at least it isn’t raining.
Me: Oooh, clever.

Which one seems more like a real person?

If I asked a woman what a dress was, she could tell me. She would know what a dress is. She would have seen one, probably worn one, felt one. If I asked a computer what a dress was I would, at best, get a definition. It wouldn’t know what a dress was. It wouldn’t have seen one, felt one, and probably not have worn one. Can there be a true intelligence without outside stimuli? I don’t think so. I think this is true for humans as well.