MurraysWorld  >  Chit Chat  >  Automation, AI & Basic Income
Pages: 1 2 3 [4] Reply

Automation, AI & Basic Income

Quote

No.....It was me who said that, but one of the things he did say, was that it may be the end of mankind!
Yeah, it's a very realistic scenario. Currently AI experts are leaning more on the positive side but not by much (52% to 31%)
[ Last edit by Mark March 10, 2016, 04:05 PM ] IP Logged
Caz
Quote

Well, I think it's down to us to make sure it can't happen, though it doesn't take much imagination to see how difficult that would be, as they would be constantly evolving, while we stood virtually still.
IP Logged
Quote

AlphaGo AI is different to chess ai, it works on experience to mimic what we call intuition - this is a different dimension to brute force pattern searching that is used in chess.

As for learning the playing style of an opponent such as what mistakes he commonly makes - AlphaGo can actually adapt to that and learn different strategies based on how an oppponent plays. Thats why they say this AI is flexible and can be applied to different fields, not just Go.

The way it learns to play is the same as the way we learn to play, by being taught the rules and different plays then learning through experience and trial and error - this is the way we learn most things.
IP Logged
Quote

Google has also got this project in the pipeline:
http://www.techxplore.com/news/2016-03-google-unveils-robot-neural-network.html
[ Last edit by Mark March 10, 2016, 04:47 PM ] IP Logged
Caz
Quote

That's interesting Laundry, but a bit scary too....when you think that robots could be doing all the jobs that used to be done by people. Maybe the 'free money' scheme if it ever comes in would compensate, at least in part.
IP Logged
Quote

Not really sure what point you're trying to make. Are you saying you disagree with the entire definition of artificial intelligence until it makes an achievement that meets your definition of intelligence?  For me creating the game itself would be computation with the addition of creativity. And once we figure out exactly how creativity works in the brain, no doubt some won't consider conquering that as intelligence.

I guess I'm saying that the ability to learn Go isn't a sign of some overall intelligence.

There's a philosophical debate behind this (funnily enough, I did a couple of years of that as well at Uni, and one lecturer actually quit what was at the time the leading work in AI as she felt the whole concept was flawed to study just what intelligence is in the first place) - if intelligence is simply learning rules, consequences and what actions make for a better choice, computers have been doing that for years in various fields.

If that is the case as well, it has far, far reaching consequences for the notion of free will or decision making in general.

Or does it need to go beyond that? Is just recognising the best choice the sign of intelligence, which we've made computers do for years now, or is it about coming up with the choice that couldn't be seen except by them?

And yes, the latter would mean there's not many "intelligent" human beings either. That's a huge debate in itself.

My main point really is that intelligence is such a wooly, undefined concept in the first place. What they're using to measure it may not be completely valid.

All of this conversation has sparked an idea for a short story in my head mind you, I love this stuff! Very Happy
IP Logged
Quote

Intelligence is not black and white, there isn't just a line you jump over and say hey thats intelligent now. Lots of very primitive organisms and computer programs display intellence even if its limitted from aemeba to the ai in pong.

All intelligence is purely progressive, our own intelligence is progressive - it's arisen through incremental changes via evolution.
IP Logged
Quote

Intelligence isn't the whole story of course - there's adaptability as well. If your operating environment is changing more quickly than you can evolve to cope, then your number's up, doesn't matter how "smart" you are. 
IP Logged
Quote

Adaptability is a part of intelligence albeit certainly an aspect of intelligence that is severely lacking with the current level of AI.
[ Last edit by laundry March 13, 2016, 02:18 PM ] IP Logged
Quote

Here's another thought.  More and more the internet is about machines talking to machines - whole infrastructures depend on this now, from energy management to transport networks. Obviously it's meant to improve efficiency and save money. But can we be absolutely sure that this won't trend away from meeting the needs of the master to promoting the priorities of the servant?  You could argue that this is already happening. It is sometimes extremely difficult to get redress for complaints when things go wrong nowadays.
IP Logged
Quote

Interesting to see that the GM won the 4th game - that was required to test how the AI will progress at such a high level, where improvements are harder to find or measure.
IP Logged
Quote

AlphaGo was foiled by an unconventional play. AlphaGo is programmed to play in a way that beats standard play type - so when a curve-ball was thrown in it gave an advantage to the GM. The thing to note here is that the GM would never make that play against a human opponent, he only made it because he knew he was playing an AI.

A similar situation arises in poker, a poker AI can be programmed to play very well and beat standard play but a smart poker player who knows he is playing an AI will be able to exploit it by playing unconventionally.

This is something that simply can't be overcome, not by AI or humans. You can play optimally against an 'average field' of plays so you have the best chance of winning against a random opponent but in a game where it's not possible to see every variable and outcome so you need to predict what an opponent is going to do then it is impossible to play in a way that will always beat them.

What is required to play well in these situation is a higher level of intuition based upon the opponent themselves - so you play the player as well as the game. The AI needs to recognize that it is getting beaten by these unconventional plays then play with some variance adjusting it's strategy to counter-act these plays - and the amount the AI does this should be based around the statistical chance that the opponent will be making an unconventional play against it.

However AI is nowhere near at a level of being able to achieve this.

* Firstly in a game like Go it is very difficult to recognize that you might be in a situation where your opponent is able to make an unconventional play and it's very difficult to train it to deal with situations like this because they don't arise often enough to have much data on them. What the AlphaGo team will likely try and do after this loss is to train AlphaGo itself to look for and make unconventional plays so it can train itself to recognize some potential danger spots.

* The second part though in trying to predict when your opponent is going to make one of these plays is above the level of the game itself. It's like trying to read a person's mind. This is something that an AI is simply not equipped to do. It certainly needs to add a degree of variance into it's own style of play (going against what it considers optimum play) so a human will not know exactly what moves it's going to be making - but it's not possible to know the optimum level of variance to program into the AI without delving into human emotions. Such a tasks requires guessing how your opponent is feeling based on his history, his personality, his body language, what his advisers might be telling him to do, etc. This is something that an AI will never be able to do that well because it requires living in the real world, knowing what information to best access and knowing human emotions.

So looking at things overall the best possible way to play a game like Go or poker is to play as a Human - AI combo. It's when people team up with AI that you can create the most formidable force. The AI will tell the human the best moves it has come up with and it's reasoning for making the moves then a human can do their best to read their opponent and the situation to make the most informed decision they can. This is how I suspect AI will ultimately be implemented into our society - not so the AI is making all the decisions but so it advises people on the best solutions and then people make the ultimate choice based upon this knowledge.
IP Logged
Quote

Nice to see Labour’s deputy leader respecting the gravity of the upcoming automation era:

http://labourlist.org/2016/04/tom-watson-a-slavish-devotion-to-money-for-moneys-sake-will-harm-society/
IP Logged
Quote

That was an extremely good speech. Not sure what to make of Watson. There's such a massive disconnect between his (sensibleish) position and the lunatic McDonnell/Corbyn "coal mining is god's chosen work, back to the 70's we go" school of thought.

Watson for next Labour leader? Possible. Desirable even...
IP Logged
Pages: 1 2 3 [4] Reply