Ship of Fools: AlphaZero = Skynet

Thread: AlphaZero = Skynet Board: Purgatory / Ship of Fools.

To visit this thread, use this URL:
http://forum.ship-of-fools.com/cgi-bin/ultimatebb.cgi?ubb=get_topic;f=2;t=020390

Posted by TurquoiseTastic (# 8978) on :

So... AlphaZero has now become top at chess as well as at Go. Many keen to apply it to fields outside board games.

SO are we just a few years away from a Terminator super-AI scenario? Even if not, it seems as though there might be some unintended consequences here...

Posted by Martin60 (# 368) on :

None that will be connectable to this signal which will disappear without trace in the noise...zzzzzzz

Posted by Ian Climacus (# 944) on :

I tend to think we are a little way off.

Playing Go is impressive, but not compared to the complex decisions, and movements, we make on a daily basis. I may be wrong but I think robots have a way to go. But the advances are amazing. I only hope ethics and the law keep up!

Posted by Jengie jon (# 273) on :

Here is the problem. While board games are often susceptible to logical reasoning to win, in real life a lot of decisions are made based on a mixture of emotions and logic. This is because we need to act before we know enough information to make a totally rational decision. Secondly all board games have a single desired outcome, winning. In real life we have several competing ones such as sustaining relationships, saving money and having a good time.

Jengie

Posted by Martin60 (# 368) on :

Near perfect Ian, Jengie. Real world problems are far too complex, truly chaotic to be meaningfully influenced by a Go master AI. Ever. Will this make the travelling salesman problem easier? The permutations of logistics? Trading? Imperceptibly. I worked for a software house that created logistics software. It saved one household name a million a year by reducing its fleet by one truck. I was caught by the fourth wave of redundancies in my 7 years there. Nothing works. That's how it works. I lasted 9 years in my last job. Trading software. How could a Go master AI help with any social or psychological issue?

Posted by TurquoiseTastic (# 8978) on :

quote:
Originally posted by Martin60:
How could a Go master AI help with any social or psychological issue?

Well indeed. But it could easily create some issues. I would imagine that if you set it a military goal, such as "kill as many people as possible", it might help one achieve that with great effectiveness.

Posted by Martin60 (# 368) on :

Not gonna happen. I can imagine autonomous drones taking out vehicles and other human IR (infrared) at night in a free fire zone, instead of a guy in RAF Waddington. But that isn't going to solve any problems in counter-insurgency without a proper do-no-harm hearts and minds policy which AI can do nothing about. The best sci-fi on this of all time is The Killing Thing by Kate Wilhelm, '67, late Golden Age.

Posted by Marvin the Martian (# 4360) on :

quote:
Originally posted by Martin60:
But that isn't going to solve any problems in counter-insurgency without a proper do-no-harm hearts and minds policy which AI can do nothing about.

The AI might well decide that the best way to defeat terrorist insurgents is to destroy their entire village/town/city/country. AIs don't have hearts, emotions or empathy, and as such will be a lot less concerned about collateral damage than we are.

A significant difference between chess and real war is that in chess sacrificing a piece or two - even your Queen - in order to win is a good thing. Would a war-planning AI know that some real-world "pieces" are too valuable to ever be sacrificed, even if that sacrifice would lead to victory?

For example, the chief thing preventing America from bombing North Korea back to the Precambrian is the knowledge that NK would be able to wipe Seoul from the map in retaliation. Would the hypothetical AI have the same attitude, or would it just see Seoul as an acceptable sacrifice in the pursuit of overall victory?

Posted by Martin60 (# 368) on :

Which is why AIs will never be in command of modern population based conflicts, only manoeuvre warfare.

Posted by rolyn (# 16840) on :

quote:
Originally posted by Jengie jon:
Here is the problem. While board games are often susceptible to logical reasoning to win, in real life a lot of decisions are made based on a mixture of emotions and logic. This is because we need to act before we know enough information to make a totally rational decision. Secondly all board games have a single desired outcome, winning. In real life we have several competing ones such as sustaining relationships, saving money and having a good time.

Strangely enough I have recently taken up computer chess.
Interesting to start with, and mentally challenging which is the idea, but after a few games you get the feeling of being up against a pre-programmed machine.

AI in warfare will lead us ultimately Gene Roddenberry's futuristic war whereby the computer says 'bang bang, you’re dead'....

Posted by no prophet's flag is set so... (# 15560) on :

We discussed this previously with autonomous cars. Altruistic programming may cause a self-driving car to go into a crash with the script to preserve all lives, weighting them equally. Selfish programming may cause the car to kill or injure others to preserve its occupants.

As for drone warfare, someone will decide that the programmers and maintainers of the computers are belligerents, and they will be shot in their homes or while shopping, in front of their children, as justice demands.

Posted by lilBuddha (# 14333) on :

quote:
Originally posted by Jengie jon:
in real life a lot of decisions are made based on a mixture of emotions and logic. This is because we need to act before we know enough information to make a totally rational decision.

Needed. Our brains work in ways for what we used to need and that really mucks up our responses in the more complex world we inhabit now. Emotion and instinctive reaction are often more detrimental in modern times.

quote:

Secondly all board games have a single desired outcome, winning.

Yes, games are limited representations of human thought. However, machine mastery is still very impressive and pause for thought in the progress made and directions possible.

UBB.classic^TM 6.5.0