3 ms·
> But the computer can be coded to do causal calculus. Which is an elegant way for it to reason about the observations it sees. I'm not convinced that number-c
by woodruffw 10y ago
> But the computer can be coded to do causal calculus. Which is an elegant way for it to reason about the observations it sees.
I'm not convinced that number-crunching could be fairly cast into the same pattern of rationality as reasoning. For one, reasoning seems to imply the potential for a multitude of (not necessarily correct or even veridical) outcomes - it is possible to reason as a human being about what to eat for dinner without predicating that on an "acceptable" meal.
In contrast, there is only one (and necessarily veridical) outcome for the sort of "reasoning" that a computer does. You can make that outcome very complex by virtue of a very complex program, but any deviation from it would be something wrong with the machine or program, not just another outcome.
> About the child and the brussel sprout, it's because they don't like the taste.
Possibly. It could also be that they've never had them but have heard their friends complaining about them, or that they just don't like the way they look.
What's interesting is that any of these would be a reasonable response from a child (or an adult). Contrast that to the explanations we would accept for a machine failing to produce the correct outcome - would it be reasonable to say (beyond in a colloquial sense) that my desktop "didn't like" the input I gave it?
- lordnacho 10y ago>>In contrast, there is only one (and necessarily veridical) outcome for the sort of "reasoning" that a computer does. Why? Say a human is driving a car, and a kid steps out on the road, you can swerve, or you can stop. Same goes with a computer. Except with a computer, you can have all the stats about the likely outcomes to hand immediately and choose whichever is marginally better. It's only the right choice to the extent that the data says one choice is superior to the other. You can easily imagine that if the choices are close to each other, the machine would just pick one of them. Or even do whatever improves the data gathering, eg if one of the choices had been picked a lot less that the other. In general there's scope for optimization when you have data and calculation power, but there's also scope for satisficing, ie doing any of several things that will give an acceptable outcome. >>would it be reasonable to say (beyond in a colloquial sense) that my desktop "didn't like" the input I gave it? Yes. The only difference between the kid and the computer is the kid evolved naturally, and one of the things he got from evolution is self-interest. Of course a machine can have self-interest as well, but we don't use the language of preferences to describe it.