Hacker News Reader

Comment by telmo | Hacker News Reader

> These were not AIs but rule-based systems with if elses.

What is fashionable to call AI these days are ultimately rule-based systems with if elses. It just so happens that it's a bazillion of those, and they are created by gradient descent.

View on HN

And that they're not if-elses but nonlinear functions.

You mean functions such as:

   if x > 0:
       return x
   else:
       return 0

?

That's one example of a nonlinear function, yes. You also have (eˣ)/(1+eˣ), or (e²ˣ-1)/(e²ˣ+1), or a host of others which don't contain anything if-like at all.

My somewhat snarky point was that there is no magic distinction such as:

> These were not AIs but rule-based systems with if elses.

I believe I already made my point with the ReLU example, if you don't want to concede it that is also ok.

I don’t know what point you’re trying to make. Neural nets are fundamentally different from rule-based systems. The fact that one of the non-linear functions which is sometimes used in neural nets can be implemented using an if statement seems utterly irrelevant.

That's RELU, which clearly belongs to the realm of AI. Dangerous stuff.

It's almost like a chose the example on purpose ;)

Reply on news.ycombinator.com