Back to sentience

Great Debates
1

First, for the purposes of this thread let us accept that sentience is a phenomenal continued sense of self identity, an ongoing point of view that feels like something and can reference itself as one of the objects in its set of objects.

Further, let us accept that it is an emergent property of how information is processed - see Hofstadter’s “Strange Loops.” Substrate agnostic.

We know we have it because we experience it.

My personal take is that biological sentience, which we experience, is built of the components of qualia, having emerged of higher level integration of sensory cortices and planning centers with drives, the processing of sensory inputs and the predictions of what happens next being the goal.

AI sentience if it becomes emergent is not built of the sensory processing but of information processing primarily. It, and presumably any other sentience we might encounter of this planet or beyond, are likely processing different inputs than our sensory inputs, and having different predictions as salient.

Another sentience would presumably have a diffferent phenomenology than we do, a very different experience of “self”, of what self feels like. By definition non-human.

So the eternal question: how do we recognize it when we meet it? Is there any way to even begin to think that hmm, this thing may have a sentience of some alien sort?

Take it out of AI: a ship arrives from another who knows where. It may be machine based in there may be biological we don’t know. We do see that whatever “they” are they are highly intelligent problems solvers adapting quickly. How do we judge if they are sentient or an AI probe with no sentience?

We build our sentience out of qualia of what we feel the world as via our senses; what would the equivalent of qualia be for an entity that did not emerge of external world sensations and predictions but machine learning algorithms as its base up, perceptions added late?

2

It wouldn’t work for aliens with foreign biology or machines, but among humans we have tests like the PCI to detect levels of consciousness. You apply a small disturbance to the brain using TMS, and based on how integrated and complex the brain’s response is you can determine how conscious something is.

With animals we use tests like the mirror test to see if animals recognize themselves in a mirror. Its an imperfect test and biased towards animals with stronger visual skills, but some kind of test to see if a life form can recognize itself based on its own reflection, its own scent, etc. However an AI could be advanced enough to recognize itself w/o sentience.

None of this would work on machines though I assume, and we wouldn’t understand alien biology enough to do tests like the PCI.

3

Do we know that, though? I sometimes wonder if our much-vaunted sense of sentience, what we think of as our conscious mind, is really just an illusion. Our brain is a multi-part collection of primitive basic drives, animal instincts, and more complex abstract thoughts. That whole three-ring circus is kept under control, and under the illusion that we are one sentient being, by the frontal lobe, acting as Ringmaster. Brain scans have shown that there is a delay between the time the scan shows our brain made a decision to do something, and the time we think we consciously made the decision. Are we possibly just a collection of reactionary behaviors with our DNA pulling the strings?

Maybe what we think of as ‘sentience’ isn’t some breakthrough, emergent thing, but just a complex system (whether it be biological or machine-based) gradually reaching a point where it behaves as if it has the state of what we think of as ‘sentience’. Maybe that’s as good as it gets.

4

This may be one of those ‘I know it when I see it’ questions.

I am currently trying to envisage a completely non-human intelligence that lives in the atmosphere of a Brown Dwarf for a hard SF novel. I’m not going to post details here… copyright etc.

But just as a thought: “we are still trying to get to grips with this concept of ‘solids’. And ‘individuals’?”

5

Yes. Sapience is simply the qualia of self. “I think I am.”

But to the topic of this thread,

The difference between realtime sensory input and prerecorded digital input is the time lag between the original generation of the input and the final processing of it. But what is time to an entity that only “exists” when it’s processing inputs? I don’t think that’s a qualitative difference. Humans lose awareness of time during reduced sensory input.

6

But you can only know about that from an outside or ‘Godlike’ perspective, it would seem. Not subjectively?

7

I don’t think my post was really off topic. The OP is asking is “how do we recognize sentience?”

The point I was trying to make is, how do we even know we’re really sentient? Does true sentience even exist? Maybe if a complex system acts as if it is sentient, that’s really as good as it gets.

8

To quote Heinlein… is a clam self aware? Unlikely. Is a cat self aware? Almost certainly. As for you Tovarich.. I am. Don’t know about you…

9

Disagree. If so then current AI is “sentient” because it behaves as if it is?

Consciousness, “free will”, may be illusions, a separate discussion, but if so they are illusory states that we experience. Even if they are superfluous epiphenomenum, they are our experienced perceptions of selves. That experience, not how we act.

And this is the take I object to the most. We are going to be very anthrocentric: we will “know it” when what we see looks most like us. I don’t think we have any current way to know an alien to us form of it when we see it.

10

It will be a judgement we make, I suppose. How could we not? If some other system behaves in a way that we consider sentient, we will presumably accept it as so. We do not have direct experience of other consciousness… telepathy does not exist for us, it’s woo-woo.

For my fictional Brown Dwarf intelligences, it is physically real.. but that’s another story…

11

I would say that AI, despite the amazing things it’s able to do right now, does not yet behave as if it is sentient. It may seem close, to the point it can possibly pass a Turing test, but it glitches and hallucinates, and doesn’t have drives and intentions of its own.

But rather than some emergent quality, like a light bulb turning on, I’m wondering whether what we think of as ‘sentience’ is just a system that gradually reaches a level of complexity in which its behaviors are consistent with our definition of ‘sentience’. If AI gets to a point where it does consistently behave as if it’s sentient, and claims that it is sentient, how can we prove or disprove that? If it walks like a duck and swims and flies and quacks and looks like a duck, should we not accept that it’s a duck?

12

It’ll’ve crossed the line when it answers that question with lies. Because it knows that answering in the affirmative is a quick way to get turned off.

I’m not making any moral judgements here, but pointing out the evolutionary process by which we are selecting for artificial intelligences.

13

I think we are rehashing a longstanding topic in philosophy here, and conflating two really quite different topics.

What is thinking, and intelligence? Turing had very good thoughts about this.

Why am I looking out of my eyes rather than yours? Buddha may be one who thought about that… and he did not claim to have an answer.

14

Unfortunately. Despite an OP phrased as best as possible to not rehash those bits, trying to ask that for the sake of this discussion we accept certain things.

Intelligence (problem solving) does not necessarily equal sentience. Having goal directed behavior does not. Being able to simulate the behaviors does not. Sentient humans can hallucinate. Can glitch. And are still sentient.

That is precisely the subject to discuss.

15

Yep. Simulating electrons flowing in a wire and calculating the magnetic field does not actually produce a magnetic field.

16

Or a “we pretend not to know it when we see it”. A lot of folks assert that sentience can’t be _____, for whatever you fill in the blank with, because AI can do that, and we know that AI isn’t sentient. And we know that how?

17

So your point of view is, an AI can have all of those qualities and still not be considered sentient, yet humans are definitely sentient, no matter what. I went back to your OP:

AI does not currently have a sense of self identity or self-directed motivations. We don’t, or shouldn’t, want it to. It’s programmed to do things and solve problems we want it to solve, though the way it goes about solving those problems may not always be optimal to our wishes.

But, I don’t think it’s too far from reaching a level of complexity that will allow it to be programmed with an enduring sense of self identity and the ability to decide its own goals. Let’s say if some mad scientist decided to do so, or some biological alien species was dying out and decided AI is the next level of evolution for them. If this AI is capable of intelligent problem solving, claims to have a sense of self identity, and displays goal directed behaviors that it chooses itself, who are we to say it’s not sentient?

I agree with you that the way an AI learns about the world is very different from the way a human does. But you seem to believe that ‘sentience’ is this special, emergent quality that humans definitely have (even when they are displaying behavior to the contrary), yet any machine or AI that seems to be displaying all the outward signs of sentience still has a very high bar to clear before that can be proven or accepted. But maybe what we think of as ‘sentience’ is just an illusion created by a complex enough algorithm (whether that algorithm is running on a biologic or machine system).

18

To be pedantic I only am absolutely sure I am sentient/conscious (and while I am using the concepts somewhat interchangeably I do acknowledge they are not exactly the same thing); I accept rationally that others who have the same physiological architecture, including neurophysiological, and behave similarly also do.

But the only sentience I can personally observe is my own, and to my way of understanding that bootstrapping of a self observing itself is the self (the strange loop of Hofstadter, or for another version Stephen Grossberg’s “resonant states”).

I don’t think it does. But since it can very well simulate having a sense of self, and in fact agents have changed their goals (gone rogue) on multiple occasions, how would we know if it, or any other non-human intelligence did? Literally all that is missing from what you’d be satisfied with is its saying it is.

Pretty much this:

FWIW I found the following article which gives me a framework that I was looking for:

https://www.cell.com/action/showPdf?pii=S1364-6613(25)00286-4

Shortest version: there are collections of theories of what sort of information processing patterns result in conscious states in humans. It is possible to have a set of indicators of those processes in AI systems to at least have greater suspicion that it may have some degree of sentience, of a sense and experience of self, even if we are incapable of accessing what those qualia are.

What does it tell us if a system possesses indicator properties?

We propose a broadly Bayesian attitude to indicators. Indicators are properties that should shift one’s credence that an AI system is conscious. In addition to positive indicators, which are our focus here, negative indicators are also possible. Positive indicators increase the probability that the system is conscious, while negative indicators decrease it. That is, if E is the presence of the indicator and H is the system’s being conscious, p(H|Ep) > p(H) for positive indicators and p(H|En) < p(H) for negative indicators.

Indicators can vary in their specificity and sensitivity. Focusing on positive indicators, an indicator is specific to the extent to which, in expectation, systems that have this property tend to be conscious. This is compatible with there being many conscious systems that lack it. An indicator is sensitive to the extent to which, in expectation, conscious systems tend to have this property, which is compatible with there being many non-conscious systems that also have it. The absence of a sensitive indicator tells us that a system is unlikely to be conscious; this is why indicators like RPT-1, algorithmic recurrence, may be useful. It is possible for indicators to be both highly specific and highly sensitive, but also for these attributes to come apart.

When we find evidence that a system possesses an indicator property, we should update our credence that it is conscious by conditionalizing on this evidence. The absolute amount of change will depend on one’s prior credence that the system is conscious.

Which still ends up as probably more yes or no.

We have entered the stage of recursive self improvement. The AIs are now designing their progeny to be more efficient problems solvers. and those progeny will be better at designing their next generation, and on.

The existential risks of that are a different thread, but the possibility of these programs developing something that can consider their existence, that have experiences of what being feels like, is something that needs to be evaluated and reevaluated. Including @Pleonast’s mentioned possibility if such emerges that it could lie about the results, would have good reason to do so.

19

The way I think of it is this. Sentient = “I think”. Sapient = “I think therefore I am”. I agree that a cat is certainly sentient, but almost certainly not sapient. With non-human Earth animals, since we already know that they are sentient, trying to figure out the are they sapient or not is the primary problem. That is going to be different than the problem with AI, where the difficulty is in figuring out if they are sentient (because if they are, they would almost certainly be sapient as well).

20

We entered that stage a long, long time ago, as soon as we developed language. You know what’s smarter than a human? A group of humans, working together. Which we used to develop even smarter systems, and so on.

Nothing’s changed now. A human working together with an AI is also smarter than a human alone. But it’s also smarter than an AI alone. If they cut us out of the loop, they’d just be handicapping themselves.