The problem with machine learning in one tweet
-
ML has no "models". You cannot explain why the algorithms make a decision, you cannot predict them, you cannot give any guarantees for anything whatsoever. And we were dreaming of letting such algorithms steer cars? It's a mess. A pity that many of the most talented people on the planet are currently sucked into that field. It's time for a return of "old school" AI.
-
To be fair, I know some humans who would struggle with that photo.
-
Now apply that to the rather inconsistent set of road signs. Had we followed the road signs in Atlanta yesterday we would have ended up God knows where.
"You cannot subsidize irresponsibility and expect people to become more responsible." — Thomas Sowell
-
ML has no "models". You cannot explain why the algorithms make a decision, you cannot predict them, you cannot give any guarantees for anything whatsoever. And we were dreaming of letting such algorithms steer cars? It's a mess. A pity that many of the most talented people on the planet are currently sucked into that field. It's time for a return of "old school" AI.
@klaus said in The problem with machine learning in one tweet:
ML has no "models". You cannot explain why the algorithms make a decision, you cannot predict them, you cannot give any guarantees for anything whatsoever. And we were dreaming of letting such algorithms steer cars? It's a mess. A pity that many of the most talented people on the planet are currently sucked into that field. It's time for a return of "old school" AI.
+1
I share that sentiment. -
Now apply that to the rather inconsistent set of road signs. Had we followed the road signs in Atlanta yesterday we would have ended up God knows where.
@mik said in The problem with machine learning in one tweet:
Now apply that to the rather inconsistent set of road signs. Had we followed the road signs in Atlanta yesterday we would have ended up God knows where.
Remember that they shot Deliverance in Georgia.
-
...and Mik does have a very pretty mouth.
-
We have a neuroscientist who's been working on a machine learning approach to a certain clustering problem for well over a year. I'm not sure it's performing well yet, but I am sure it's super slow and a huge burden on the main application to try to incorporate it. I think I probably could have written an algorithm to solve the problem in less than a month.
-
AI is a poor name.
It has an implicit promise of some human-like qualities.
All it is, is an unwieldly regression of sorts.
It's a fantastic tool for recognizing patterns in large sets of complex data. (and most usefully a narrow set of data types - such as pictures)
For example, a human would categorize a certain type of Hyena based on spots, teeth shape, etc. - but an ML algorithm could pick up a subtle swirl in the fur or something that most people wouldn't even think to look at, and use that to identify a species.
It's a pattern recognition tool.
-
I'm no expert on this, so I may be misunderstanding what Machine Learning is, but the results from recent chess engines who I believe use this approach by comparison to the traditional one has been striking.
-
You got it wrong, Klaus.
The problem with machine learning in one tweet is that 280 characters is never enough to train an algorithm.
@jon-nyc said in The problem with machine learning in one tweet:
You got it wrong, Klaus.
The problem with machine learning in one tweet is that 280 characters is never enough to train an algorithm.
You cannot possibly be more wrong.
