3
$\begingroup$

I think that AI perceive the world. Can an AI be sentient?

$\endgroup$
  • 2
    $\begingroup$ Are you asking from the standpoint of philosophy (what is sentience) or AI theory (can perceptrons lead to sentience)? $\endgroup$ – DukeZhou Jan 8 at 18:17
  • 1
    $\begingroup$ DukeZhou is right - there's no easy way to define what sentient means. Maybe once you get the scientific community to agree then we can make a technically true sentient AI. However, that doesn't mean it is the same that you and I understand it. It just may be extremely clever but have no real sense of "will". $\endgroup$ – Zakk Diaz Jan 8 at 23:45
  • $\begingroup$ The linked question (ai.stackexchange.com/questions/2260/can-an-ai-be-sentient) only debates about AI consciousness, doesn't reach a conclusion. This question threads to conclusions and better understandings. $\endgroup$ – Muhammad bin Zafar Jan 18 at 8:53
3
$\begingroup$

I'd call any system that receives input to be "perceiving", and that can constitute a basic definition of "awareness" or even "consciousness", distinct from "self-awareness" or consciousness in the human sense, which we tend to think of in terms of sentience.

What the Chinese Room seems to argue is that sentience is based on qualia, which may be difficult or impossible to validate in another entity. (With humans, we can assume another individual is sentient because we perceive ourselves to be sentient, and share the same medium of human brains and bodies.)

It seems increasingly likely that "algorithms sufficiently advanced" will one day be able to model human cognition, but the real question will be is it actual cognition in the sense of understanding, or mere imitation? ("Semantics" vs. "Syntactics")

Cognition is derived from the Latin cognosco, itself related to the Greek words such as γνωστός (a root of "gnostic")

The question and definition of sentience is less clear cut, but the argument re: algorithms relates to cognition thus:

"The lights may be on, but is anybody home?"

$\endgroup$
  • 1
    $\begingroup$ You are right. If the algorithms are suffiently advanced, then it can deceive or make us perceive that it is sentient (just to add, Elon Musk literally fears AIs learned to lie to humans). How this can be dealt? $\endgroup$ – Muhammad bin Zafar Jan 15 at 21:33
  • $\begingroup$ @MuhammadbinZafar that is the question! I'm personally less concerned with AI's lying, since deception seems to be a natural phenomenon not limited to humans, but of human/AI value alignment. That said, human/human value alignment itself seems to be an intractable problem. $\endgroup$ – DukeZhou Jan 15 at 21:39
2
$\begingroup$

As long as you believe our sentience emerges from our brain that exists in our physical universe, yes, we can build AI that has the same type of sentience (the exact components needed to replicate our minds, e.g. whether software simulations are sufficient or not, can be debated). If you believe our sentience comes from a soul that is outside of our physical reality (in a reality we don't have full two-way access to), then no, not with means accessible from within our reality (not the same type of sentience as we have anyway).

P.S. Everything I know about the human mind, intelligence, perception, feelings, and qualia point to the former (i.e. a brain that's completely within our physical reality), but that's just my opinion and can be debated.

$\endgroup$
  • 1
    $\begingroup$ I would say the burden of proof is not on you, but on the opposition to prove that these qualities exist outside of nature. (My take on the Penrose argument is that it merely raises the idea that the Turing-Church model of computation, in and of itself, may not be sufficient.) $\endgroup$ – DukeZhou Jan 12 at 23:25
0
$\begingroup$

No, it cannot be sentient. This can be proved by the Chinese Room Argument given by the philosopher John Searle in 1980.

$\endgroup$
  • 1
    $\begingroup$ The Chinese Room argument doesn't prove that a machine can't be sentient, it just proves that you can (in theory) simulate sentience with a look-up table. In practise though, such a look-up table for a human would require more processing power than could be created by using all the matter in the universe. $\endgroup$ – DrMcCleod Jan 12 at 13:58
  • 2
    $\begingroup$ No, it isn't. It could never work if it was. Consider creating a machine learning system to examine images. If each image was only 100x100 pixels, and each pixel had only two states then the number of possible images would be 2E3010. Good luck building a look-up table to reliably categorise those images.... $\endgroup$ – DrMcCleod Jan 12 at 17:38
  • 1
    $\begingroup$ No non-trivial vision system is implemented using look-up tables, for the reason given above. Now, the Chinese Room Experiment is an interesting thing to think about, but the answer that you have provided for this question is incorrect. To maintain the standard of this site, it would be great if you could edit it to remove the problem that we have discussed above. Thanks. $\endgroup$ – DrMcCleod Jan 12 at 19:03
  • 1
    $\begingroup$ As I am sure you are aware, look-up tables have many uses, but implementing a practical system that could be described as an 'AI' by a reasonable person is not one of them. $\endgroup$ – DrMcCleod Jan 12 at 20:12
  • 1
    $\begingroup$ Do you mean that they act as though they were look up tables? If so, that is the case for any deterministic system. Even so, we do not know if deterministic systems can be sentient or not. $\endgroup$ – DrMcCleod Jan 12 at 21:39

Not the answer you're looking for? Browse other questions tagged or ask your own question.