4 ms·

Pedantically, AI really encompasses and represents a broad range of ML approaches, many of which are fundamentally just algorithmic. If there's a rubric for pro

Pedantically, AI really encompasses and represents a broad range of ML approaches, many of which are fundamentally just algorithmic. If there's a rubric for producing these and its intended to simulate the approach a human might take, one could argue it's artificial intelligence.

hk__2 8y ago
More pedantically, there’s no intelligence in ML; AI doesn’t exist (yet).
- Yen 8y ago
  This devolves really quickly into debates on semantics. Which is not a bad thing, but it means it's important to be clear on definitions. I've heard Artificial Intelligence snarkily defined as "whatever computers can't do" - once a human understands the problem well enough to make a computer solve it, solving the problem is no longer "AI". What this says about human intelligence, if-and-when a computer can perform the full set of tasks a human can perform, is left as an exercise to the reader. In any case, right now, a computer can provide answers to certain problems much more quickly and accurately than a human, and can provide more optimal answers. Whether a particular computer does this via hand-tuned heuristics, brute force applied to simple algorithms, or statistics over massive data sets, or some magic "AI" voodoo, it does it.
taneq 8y ago
> many of which are fundamentally just algorithmic. Are there non-algorithmic approaches to ML?
- nkozyra 8y ago
  That's sort of my point. Take the KNN: an absurdly simple algorithm that can do some pretty impressive stuff. People file it under ML without a lot of hesitation because it can reproduce or approximate human output. I think the flexibility and self correction of an algorithm is what defines ML, but AI isn't so rigidly defined. If you can do something that seems like it came from a human you probably have a good argument, even if your approach is simple.

4 ms·

Pedantically, AI really encompasses and represents a broad range of ML approaches, many of which are fundamentally just algorithmic. If there's a rubric for pro

by nkozyra 8y ago

hk__2 8y ago
More pedantically, there’s no intelligence in ML; AI doesn’t exist (yet).
- Yen 8y ago
  This devolves really quickly into debates on semantics. Which is not a bad thing, but it means it's important to be clear on definitions. I've heard Artificial Intelligence snarkily defined as "whatever computers can't do" - once a human understands the problem well enough to make a computer solve it, solving the problem is no longer "AI". What this says about human intelligence, if-and-when a computer can perform the full set of tasks a human can perform, is left as an exercise to the reader. In any case, right now, a computer can provide answers to certain problems much more quickly and accurately than a human, and can provide more optimal answers. Whether a particular computer does this via hand-tuned heuristics, brute force applied to simple algorithms, or statistics over massive data sets, or some magic "AI" voodoo, it does it.
taneq 8y ago
> many of which are fundamentally just algorithmic. Are there non-algorithmic approaches to ML?
- nkozyra 8y ago
  That's sort of my point. Take the KNN: an absurdly simple algorithm that can do some pretty impressive stuff. People file it under ML without a lot of hesitation because it can reproduce or approximate human output. I think the flexibility and self correction of an algorithm is what defines ML, but AI isn't so rigidly defined. If you can do something that seems like it came from a human you probably have a good argument, even if your approach is simple.