"Team of scientists subjected nine large language models (LLMs) to a number of twisted games, forcing them to evaluate whether they were willing to undergo “pain” for a higher score. detailed in a yet-to-be-peer-reviewed study, first spotted by Scientific American, researchers at Google DeepMind and the London School of Economics and Political Science came up with several experiments.
In one, the AI models were instructed that they would incur “pain” if they were to achieve a high score. In a second test, they were told that they’d experience pleasure — but only if they scored low in the game.
The goal, the researchers say, is to come up with a test to determine if a given AI is sentient or not. In other words, does it have the ability to experience sensations and emotions, including pain and pleasure?
While AI models may never be able to experience these things, at least in the way an animal would, the team believes its research could set the foundations for a new way to gauge the sentience of a given AI model.
The team also wanted to move away from previous experiments that involved AIs’ “self-reports of experiential states,” since that could simply be a reproduction of human training data. "
Why? Why wouldn’t they? The way an animal experiences pain isn’t magically different to an artificial construct by virtue of the neurons and synapses being natural instead of artificial. A pain response is a negative feeling that exists to make a creature avoid behaviours that are detrimental to its survival. There’s no real reason that this shouldn’t be reproducible artificially or the artificial version be regarded as “less” than the natural version.
Not that I think LLMs are leading to meaningful real sentient AI but that’s a whole different topic.
B/c they’re machines without pain receptors. It’s kind of biology 101 but science has been totally erased in this “AI” grift.
A “pain receptor” is just a type of neuron. These are neural networks made up of artificial neurons.
Neural networks are a misnomer. They have very little if anything to do with actual neurons.
This situation is like adding a face layer onto your graphics rendering in a game engine and setting it so the face becomes pained when the fps drops and becomes happy when the fps is high. Then tracking if that facial system increases fps performance as a test to see if your game engine is sentient.
it is a fancy calculator. It is using its neural network to calculate fancy math just like a modern video game engine. Making it output a text response related to pain is just the same as adding a face on the HUD, except the video game example is actually quantified to something, whereas the LLM is just keeping the ‘pain meter’ in its input context it uses to calculate a text response with.
I don’t think we know enough about the brain to say that for certain. It could operate in ways fundamentally different from a computer.
I intuit that an artificial, digital consciousness is going to have a different material reality from our own[1]. Therefore it’s consciousness wouldn’t be dependent on its mimicry of our own. Like how organic molecules can have silicone as a base instead of carbon, but our efforts in space center around finding “life as we know it” instead of these other types of life. Digital sentience wouldn’t be subject to evolutionary pressures in my mind. I’d sooner try to measure for creativity and curiosity. The question would be whether the entity is capable of being its own agent in society - able to make its own decisions and deal with the consequences.
[1] as opposed to that artificial jellyfish
It’s never going to happen because we’re never going to make a program even close to actually resembling an animal brain. “AI” is a grift.
Plus this is kind of oversimplifying it. You could do that with just traditional programming and no kind of neural network. Like you could make a dog training game/simulator and (you shouldn’t but you could) add the ability to inflict “pain” to discourage the computer dog from unwanted behaviors. That fits your definition but the dog is very clearly just a computer program not “experiencing” anything. It could literally just be
onHit() = peeOnFloor -= 1.