• Security incident: ISF was recently accessed by intruders. Please change your password, and change it anywhere else you used it. Read more

Merged Artificial Intelligence

steenkh said:
If they are programmed to maximise scoring, some agents are not doing what they are programmed to do.
Click to expand...
Or they're not doing it in a way you expected them to, or in a way that was anticipated.

I get that AI is more complex, definitely. It's much more adaptive and it adds to it's base data set as it goes. But I suspect there's still that old truism at the heart: If a computer doesn't do what you told it to do, you didn't tell it what you thought you did.

It's like that elementary school project where you have to write out step-by-step instructions for how to make a peanut butter and jelly sandwich. Every kid thinks they've written down perfect steps... but the teacher who follows them exactly ends up peanut-buttering their hand because the kid forgot to tell them to pick up a piece of bread. And how to open a package of bread.
 
Right. And I won't be excited about AI as a useful tool for tasks involving anything we'd call judgement until we've figured out how to build in a reliable "don't act like Amelia Bedelia or Monkey's Paw" bit into it. We really should not be using a tool that you have to instruct as though it's going to do its level best to figure out how to do something you didn't want it to, because it's a 'try everything including everything you didn't think of telling it not to' engine with no guardrails on it. If we genuinely cannot figure out cuss-cuss guardrails what use is the thing? It's a button you press where you either get what you want or you get something that complies with the letter of what you asked for whilst also unexpectedly cutting your grandma into 8 equally sized cubes, because that was the best way it found to solve your problem.
 
Lithrael said:
Right. And I won't be excited about AI as a useful tool for tasks involving anything we'd call judgement until we've figured out how to build in a reliable "don't act like Amelia Bedelia or Monkey's Paw" bit into it. We really should not be using a tool that you have to instruct as though it's going to do its level best to figure out how to do something you didn't want it to, because it's a 'try everything including everything you didn't think of telling it not to' engine with no guardrails on it. If we genuinely cannot figure out cuss-cuss guardrails what use is the thing? It's a button you press where you either get what you want or you get something that complies with the letter of what you asked for whilst also unexpectedly cutting your grandma into 8 equally sized cubes, because that was the best way it found to solve your problem.
Click to expand...
Asking for a friend... Which AI are you talking about....?
 
Emily's Cat said:
Sounds like any other algorithm that is designed to learn an optimal approach to something.

You know, like what videos YouTube recommends, only better.
Click to expand...
You are suggesting Queen to Bishop 3, Bishop takes Queen, Knight takes Bishop, Mate... is like a Captain and Tennille music video?
 
An interesting story I came across today. There have been examples of people entering AI generated or modified images in photography competitions (often being detected by people on the internet after the event). In a reversal, Miles Astray won a category for AI generated images with a regular photograph.

www.milesastray.com

Miles Astray | news - reclaiming the brAIn

how my real photo F L A M I N G O N E won an AI award
www.milesastray.com www.milesastray.com

(He revealed what he had done, and was disqualified, which he had expected, but was pleasantly surprised at the supportive words from the judges.)
 
Darat said:
Problem for them is where are these GPUs? We know they are not in datacentres at the moment since the "AI" datacentres haven't come on line, so that means they are in a warehouse. Have they even left nVidia's warehouses?
Click to expand...
Someone's been reading my posts again:


Comments about Oracle's recent issuing of a "force majeure" with one hyperscaler, an update on lack of datacentres coming online, where are the GPUs and the general craziness behind the AI bloom.
 
theprestige said:
Because they're doing it.
Click to expand...
Now you are just falling back on dogmatics: if we don’t know why an AI agent does what it does, then we have programmed it to do so.
theprestige said:
◊◊◊◊, the futurists have been telling us about AI having more than one goal at a time for almost a hundred years. How is this still a foreign concept?
Click to expand...
What other goal are you talking about?
 
You must log in or register to reply here.

Share this page

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom