This afternoon, my wife and I went for a bit of a drive, and while we were out, I talked about AI. About which, I know enough to sound knowledgeable to anyone who doesn't actually understand it. I wasn't trying to baffle her, or anyone; its just that I only generally understand the concepts, but I like to talk about them anyway.
One of the things that I realized, while talking, is that you don't have to understand AI programming to want to be involved in the utilization of AI technology; and you don't have to undeerstand the utilization of AI technology to want to be involved in the programming of AI. But, in either direction, knowledge of the other side helps. My primary interest is the utilization of AI -- what can it do for us? -- but I like to know how it does what it does. Not at the bit and byte level (do working AI programmers use that phrase, I wonder? And if not, how do they refer to knowing something at the atomic (not literally) level?), but enough to get a sense of how it gets from 'gee, wouldn't it be nice to have something that could....' to having a working prototype.
What brought this to mind (again) was coming across a site for an AI shell called UltraHAL. UH's a chatbot -- you talk to it, it tries to talk to you. Right off the top, its not all that impressive. Here's a transcript of my first conversation with it (using a character it comes with called 'Zabu'):
User: i like to read the newspaper on Sunday.
Zaba: User Wow! Read the newspaper on sunday.
User: My favorite newspaper is the Washington Post.
Zaba: Oh User! Newspaper and public press and paper!
User: Tell me the name of my favorite newspaper.
Zaba: I try to avoid junk food.
User: So do I. Do you know the name of my favorite newspaper.
Zaba: Hoover is an internationally known name.
User: You are in error, Zabu.
Zaba: Thanks for teaching me that I am error.
You can sort-of see the program trying to respond, but it doesn't seem too bright. I guessed that, having gotten the phrase "I like to read the newspaper on Sunday", it might be a bit confused by 'on Sunday', since that could be a condition about when I do the reading, or it could be a way to indicate that I'm reading a newspaper on the topic of 'Sunday'. But the first part was pretty straightforward, so what I would have expected would be something like 'I like to read, too', or 'What do you like to read' or 'Why do you like to read the newspaper on Sunday'. But 'User Wow! Read the newspaper on sunday' seems completely off the wall. Though, I do like it suggesting that my favorite newspaper is junk food.
I think I'll play with it a bit more.
2 comments:
AI is used in some washing machines to control soap and water usage. It's also used in some heating and cooling systems to manage the temperatures.
Some cars employ it for their obstruction avoidance systems, and that car that parallel parks itself? It uses AI for that.
It can also be used to detect patterns in seemingly random data. Wall St has been trying to use such systems for years, with varying results. (It turns out that the Dow really is random!)
And so on. A million uses, not always high-falutin', but useful nonetheless.
I got involved in AI about a decade ago. A bank I worked for wanted an AI-based system for a customer-service system; the ended up not implementing it when the cost became obvious. Reps, at not much more than minimum wage, were cheaper.
There isn't much investment in AI systems; although that might change in the next few years. (Personally, I think it will.) The lack of investment simply matches the lack of decent ideas that can be 'sold' to the money-men. (Not PC, but you get the idea!) There are some university courses available, but I don't know what they're like. I've got a hunch it will all change in the next few years as huge data sets require new approaches to their understanding. It's impossible to comprehend gigabytes of data, but a decent AI system might be able to summarize it!
Carolyn Ann
I recall when I first learned about the concept of fuzzy logic, and how they had to carefully say that fuzzy logic didn't mean imprecise logic, but rather logic which didn't have to deal in black/white, on/off values. The systems you talk about for heating and cooling undoubtedly use that. I'd guess that any system that has to be able to handle imprecise input would use it.
Not too sure about the idea of massive datasets as being the driver, though. I think it's going to enter the society as things that we don't *think* of as AI -- because by that time, they won't be seen as unusual. So perhaps a better way to pick out the uses (now, and potential) would be to ask where an autonomous system would be effective. The thing that came first to my mind was places where decisions were required, but the environment was hostile to life , or difficult to sustain (undersea, in space); my wife thought of places where it simply took too long for 'intelligence' to get back to the place where it was needed -- on the Mars rover, for example.
Thats my biggest problem in thinking about AI -- I think of it in science fiction terms, but its likely more effective to think of it in problem solving terms. Sort of like how I've heard expert systems defined -- if you can do it over the phone, an expert system can do it. Okay, extend that logic to fuzzy logic systems/neural nets/genetic algorithms. What can THEY do thats currently done by people?
I really need to think about this (for which read: I really want to understand it). I want to 'see' how it can move into the mainstream and become normal. And how -- as you point out -- it already has.
Post a Comment