3 ms·

To be a little cynical, I remain unconvinced that humans are not just pattern matching too. Go players thought there was something uniquely human about masteri

To be a little cynical, I remain unconvinced that humans are not just pattern matching too. Go players thought there was something uniquely human about mastering Go... until an AI proved them wrong. We are just beginning to see that machines are better at diagnosing some diseases earlier and with higher precision than human experts... Can you even think of a behavior that isn't pattern matching, at its core? I'd argue it doesn't exist. Humans are awful at random number generation. If you really think about it, there is really no evidence that humans are doing anything other than crunching numbers.

Barrin92 7y ago
>Can you even think of a behavior that isn't pattern matching, at its core? I'd argue it doesn't exist. Yes, quite trivial ones, for example winograd schema. (1) John took the water bottle out of the backpack so that it would be lighter. (2) John took the water bottle out of the backpack so that it would be handy What does it refer to in each sentence? ML systems absolutely suck at this because the structure of the sentence is identital. The answer is derived from understanding the meaning of the sentence. ML systems in NLP right now esentially correlate strings without having any idea of the underlying world. To solve a problem like the one above you need to go from syntax -> model of the world -> back to syntax. That would constitute actual understanding. Similar problems are of course questions of the kind "X was alive in 2000. X was alive in 2010", was X alive in 2005? Pattern matching doesn't help you here, you need a deep model about the world to actually answer the question. (unless you're guessing). The reason why pattern matching can't solve every problem is because real world problem spaces are open-ended. There are infinite sentences or infinite problems, some entirely new without any solution in historical data.

Barrin92 7y ago

>Can you even think of a behavior that isn't pattern matching, at its core? I'd argue it doesn't exist. Yes, quite trivial ones, for example winograd schema. (1) John took the water bottle out of the backpack so that it would be lighter. (2) John took the water bottle out of the backpack so that it would be handy What does it refer to in each sentence? ML systems absolutely suck at this because the structure of the sentence is identital. The answer is derived from understanding the meaning of the sentence. ML systems in NLP right now esentially correlate strings without having any idea of the underlying world. To solve a problem like the one above you need to go from syntax -> model of the world -> back to syntax. That would constitute actual understanding. Similar problems are of course questions of the kind "X was alive in 2000. X was alive in 2010", was X alive in 2005? Pattern matching doesn't help you here, you need a deep model about the world to actually answer the question. (unless you're guessing). The reason why pattern matching can't solve every problem is because real world problem spaces are open-ended. There are infinite sentences or infinite problems, some entirely new without any solution in historical data.