DeepMind AI learns simple physics like a baby

Re:DeepMind AI learns simple physics like a baby by exodust on Tuesday July 12, 2022 @01:34AM (#32066043)

I wonder how much the child's sensory reward system leads to memories and associations. How can artificial neural networks ever "feel" good about anything, and use that to learn?

A child learns that petting a cat is better than hitting the cat. Petting feels better, and the cat will stick around for longer. But how does AI learn the same thing without any pleasurable sense the petting provides?

Also will we see robots throwing tantrums? Children learn this tactic early, and it works to achieve desired outcomes. Later they progress to crafty manipulation tactics, such as carefully executed tone when asking for treats etc.

[ Parent | on HN ]

Re:DeepMind AI learns simple physics like a baby by hndamien on Tuesday July 12, 2022 @02:03AM (#32066195)

We have a cost function that could loosely be correlated to a pain/sadness function. Perhaps it makes sense to also include a pleasure function to give them some kind of autonomous direction. Perhaps it is this part of AI that will ultimately lead to trouble. Is pleasure the inverse of pain?

[ Parent | on HN ]

Re:DeepMind AI learns simple physics like a baby by exodust on Tuesday July 12, 2022 @06:29AM (#32067890)

I think the inverse of pain is just comfortable or satisfied. Like a cat sitting on the mat. If the cat finds an electric heated blanket nearby and moves from mat to warm blanket, we are now in pleasure zone. Cat now has more than it needs.

AI could use a pleasure function, but I'm not sure how it could justify seeking or getting more than it needs. The "pleasure" would need to be linked to a measurable benefit for its hardware or systems, and energy would need to be used to achieve that. I guess it could happen, but then we'll need greed functions to cap the pursuit of pleasure!

[ Parent | on HN ]

Re:DeepMind AI learns simple physics like a baby by atwood22 on Tuesday July 12, 2022 @02:00AM (#32066186)

Neural networks have loss functions. In a sense, feeling good and bad can be looked at as outcomes of a loss function.

[ Parent | on HN ]

Re:DeepMind AI learns simple physics like a baby by littlestymaar on Tuesday July 12, 2022 @02:31AM (#32066355)

> Also will we see robots throwing tantrums? Children learn this tactic early, and it works to achieve desired outcomes.

Sigh. Tantrum isn't a “learnt tactic” (it's the child's inability to control his frustration / other emotion) and it doesn't work to achieve desired outcome which is why they stop when the child is mature enough to have better emotional control (unless you're trying to raise a tyrannical child, in which case giving your child what he wants when he throws a tantrum is a good way to succeed…)

[ Parent | on HN ]

Re:DeepMind AI learns simple physics like a baby by exodust on Tuesday July 12, 2022 @05:54AM (#32067680)

> Sigh.

Is sighing a learnt tactic? How about a written sigh, what's that about? Inability to use normal words?

I'm quite sure some "tantrums" are last resort efforts to reverse whatever decision the parent made that upset the child.

[ Parent | on HN ]

Re:DeepMind AI learns simple physics like a baby by mdp2021 on Tuesday July 12, 2022 @10:20AM (#32070362)

I think yours was a good question in a way (it may make sense in some contexts, and with undefined translation): it may be part of the "paperclip maximizer" idea of the strategies a non-commonsensical agent may adopt to achieve its goals.

In automated chip layout, for example, already one can be proposed odd NOP circuitry that reveals to be just relevant to timings - all tricks are possible.

"Sighs" could presumably be the reversal manouvers in decision trees walking.

[ Parent | on HN ]