coupler profile
coupler's Avatar

coupler profile

4
coupler Avatar
coupler responded to Belimawr comment in
4 Years ago
Julian Gollop: “AI should be a lot better now”Julian Gollop: “AI should be a lot better now”
Belimawr Avatar

but what you describe isn't learning in the general sense, it is a logic based program working out what it is looking at, it doesn't learn it just records information based on the constraints of the program.


if it was truly learning the computer/AI would be able to work out what it's goal is and where it needs to go allowing it to change the way it looks at things, instead of following the exact same programming.

coupler Avatar

machine learning does not use logic, it uses math. ML techniques *do* allow a computer to figure out what it needs to do and where it needs to go without being strictly programmed to do so.


recording information is not machine learning. machine learning is analyzing information and producing a result.


with machine learning, an algorithm is potentially bad at doing something over and over again, but is able to analyze it and figure out what makes the outcome more desirable. after some time and a lot more data, it gets better at this task. how is that not learning in the truest sense of the word?


I think you are arguing that computers are not sentient and therefore don't "understand" what it means to learn which is true I suppose.


if you are interested in this topic, I recommend learning about it (no pun intended). it seems like you don't really understand the basic concepts at all.

sign in to comment
Tim Edwards Avatar
Tim Edwards responded to coupler's comment in
4 Years ago
Julian Gollop: “AI should be a lot better now”Julian Gollop: “AI should be a lot better now”
coupler Avatar

What you describe is the antithesis of machine learning. Computers absolutely can learn. It is not just "logic charts taking different turns for different actions". That's more like "How to get an F in a Machine Learning class". Sure, you are right that a bunch of if statements can not make a computer learn, but you are incorrect that this is not possible today.


Machine learning and neural networks have been around since the 1950s. Right now machine learning is used to help analyze cancer research data (recently computers have helped cure certain diseases using its ability to learn faster than a humans), and solve other major problems like what you should watch next on Netflix.


One example of machine learning in games as AI I can think of was recently written about in the New York Times (look up "The Steely, Headless King of Texas Hold ’Em" - it's a *really* good article - I highly recommend reading it). It's a heads-up poker bot that uses machine learning (the machines in the casinos are "locked in" as it is illegal for them to change on the spot) and has even learned how to bluff.


The "problem" with machine learning is it takes a LOT of data to analyze. For a game AI to get good with machine learning, it has to play itself hundreds of thousands of times or more and also against humans. Every game that it plays is more data to analyze and figure out what to do.


FWIW I am pretty sure the new Planetary Annihilation uses neural networks for its AI.

Reply
Tim Edwards Avatar

"FWIW I am pretty sure the new Planetary Annihilation uses neural networks for its AI."


O rly. That's super interesting.

sign in to comment