Talking about AI is always so weird in these articles. For one thing, AI isn't even a well defined term. Strictly speaking, it's about creating rational agents, but everyone knows what news...
Talking about AI is always so weird in these articles. For one thing, AI isn't even a well defined term. Strictly speaking, it's about creating rational agents, but everyone knows what news articles and startups mean when they say "artifical intelligence": dEeP LeArNinG. Apparently a binary classifier is now "aRtiFicIaL InteLigEnce". Does my calculator have AI when it does linear regression?
So when they talk about "advancements in AI", it's advancements in a specific part, deep neural networks, in a specific field, machine learning, when only a small segment of that is actually producing agents. It's super weird that the article contrasts Monte-Carlo Tree Search with "new AI". You do realize that AlphaGo, and AlphaZero, both use MCTS? The neural networks replace the traditionally handcrafted functions which approximate board value. When you think about it, the actual "agent" part of AlphaGo is MCTS.
Stuff like counterfactual regret, which performed well in games like heads up texas hold'em in the last year or so, is of course not mentioned, because idk.
edit: unrelated but I also love when somehow evolutionary optimization comes into it. No, neural networks have nothing to do with evolutionary methods. Evolutionary optimization would just completely fail on something with as many parameters as neural networks. It is sometimes necessary when you are truly working with a black box function, but when loss functions are differentiable, like in NNs, it's naturally inferior to something that can make use of more info.
To grasp this fields, one has to dedicate hundreds of hours. I doubt this journalists know anything besides the buzzwords that are commonly tossed around. It must be really hard trying to explain...
To grasp this fields, one has to dedicate hundreds of hours. I doubt this journalists know anything besides the buzzwords that are commonly tossed around. It must be really hard trying to explain something that you don't understand to people who are going to read your explanation in a passive way.
I'm starting to think that it might be even beneficial having most people believe that "ai" is black magic. At least this way, curious people that invest in the field can be rewarded with high salaries. Otherwise the money would be put into something else...
Talking about AI is always so weird in these articles. For one thing, AI isn't even a well defined term. Strictly speaking, it's about creating rational agents, but everyone knows what news articles and startups mean when they say "artifical intelligence": dEeP LeArNinG. Apparently a binary classifier is now "aRtiFicIaL InteLigEnce". Does my calculator have AI when it does linear regression?
So when they talk about "advancements in AI", it's advancements in a specific part, deep neural networks, in a specific field, machine learning, when only a small segment of that is actually producing agents. It's super weird that the article contrasts Monte-Carlo Tree Search with "new AI". You do realize that AlphaGo, and AlphaZero, both use MCTS? The neural networks replace the traditionally handcrafted functions which approximate board value. When you think about it, the actual "agent" part of AlphaGo is MCTS.
Stuff like counterfactual regret, which performed well in games like heads up texas hold'em in the last year or so, is of course not mentioned, because idk.
edit: unrelated but I also love when somehow evolutionary optimization comes into it. No, neural networks have nothing to do with evolutionary methods. Evolutionary optimization would just completely fail on something with as many parameters as neural networks. It is sometimes necessary when you are truly working with a black box function, but when loss functions are differentiable, like in NNs, it's naturally inferior to something that can make use of more info.
To grasp this fields, one has to dedicate hundreds of hours. I doubt this journalists know anything besides the buzzwords that are commonly tossed around. It must be really hard trying to explain something that you don't understand to people who are going to read your explanation in a passive way.
I'm starting to think that it might be even beneficial having most people believe that "ai" is black magic. At least this way, curious people that invest in the field can be rewarded with high salaries. Otherwise the money would be put into something else...