Could an artificial intelligence (AI) be sentient? How could we tell? A recent opinion piece in the New York Times claims that AIs are not sentient.⁽¹⁾ The article raises an interesting question but does not adequately answer it, in part because it conflates sentience and intelligence and in part because its language is confused. Here is an example of the confusion:

There is no evidence this technology is sentient or conscious — two words that describe an awareness of the surrounding world.

This sentence is little more than a tautology, since in English “conscious” and “aware” mean the same thing.⁽²⁾ It basically says that “sentient” means “conscious” and “conscious” means being conscious.

Here is another:

Sentience — the ability to experience feelings and sensations — is not something easily measured. Nor is consciousness — being awake and aware of your surroundings.

So sentience is the ability to experience — that is, the ability to be conscious of — feelings and sensations, and being conscious is (substituting equivalent words) being awake and conscious. That doesn’t tell us much.

Confused as this language is, we can agree that “sentient” means being conscious. But in what sense? Does it mean having the capacity to be conscious or actually being conscious during some period of time? Or both? Clearly if something has no ability to be conscious, it can’t be conscious in any given duration of time, so let’s say it means both. The author is claiming that AIs have no capacity to be conscious and are never conscious of their world. Unfortunately, he offers little evidence for his assertion.

Before we get there, let’s dispose of a related question, whether an AI can be intelligent. Here is a definition of intelligence:

The ability to learn — the ability to take in new context and solve something in a new way — is intelligence.

Obviously many AIs are intelligent, albeit artificially so. That’s the whole point of the AI enterprise, to create things that can learn and solve problems in a new way. When Google Assistant or Alexa hears what you say and gives a relevant and useful response, that’s AI at work. And we can judge how intelligent they are by the relevance of their responses. Siri, for instance, comes in a poor third in such a contest. But are they conscious?

We need to distinguish two aspects of being conscious, first pointed out many years ago by philosopher Ned Block.⁽³⁾ He calls them “phenomenal consciousness” and “access consciousness,” which I prefer to restate as being conscious in phenomenal mode and being conscious in access mode. In phenomenal mode we see colors and shapes, we hear sounds, we smell aromas, etc. In access mode we have ideas or mental representations of what we are phenomenally conscious of, and these ideas enable us to do something with it, such as reason about it, say something about it or take some action on it. We are able to do these things with something of which we are phenomenally conscious because we have a representation of it in our mind.

In most cases the phenomenal aspect and the access-enabling aspect occur together. That’s why many use the term “conscious” to mean both. But some AIs are clearly conscious in access mode even though we doubt that they are conscious in phenomenal mode. An example is a self-driving car or a robot. Such devices detect and respond to their environment. They can make decisions, for instance whether to stop or proceed or to slow down or speed up. They can speak; think of Amazon’s Alexa and Google’s Assistant. But we doubt that the world appears to them phenomenally in any way at all.

And that’s the issue at hand, whether AIs can be truly phenomenally conscious. Are they — can they be — conscious of things in phenomenal mode, as we are when we are awake and alert? Does a world appear to them as it does to us?

The problem is that we can tell only inferentially. As the author says, whether something is sentient is “not something easily measured.” We have no direct access to someone else’s mind, let alone the mind of an AI. Hence, we can only ascertain the presence of mind from behavior.

Unfortunately, the author cites as evidence researchers who seem to conflate sentience, the ability to be phenomenally conscious, with intelligence.

“A conscious organism — like a person or a dog or other animals — can learn something in one context and learn something else in another context and then put the two things together to do something in a novel context they have never experienced before,” Dr. [Colin] Allen of the University of Pittsburgh said. “This technology is nowhere close to doing that.”

Alison Gopnik, a professor of psychology who is part of the A.I. research group at the University of California, Berkeley, agreed. “The computational capacities of current A.I. like the large language models,” she said, “don’t make it any more likely that they are sentient than that rocks or other machines are.”

The argument seems to be that current AIs are too dumb to be conscious. But that doesn’t follow at all. Plenty of conscious organisms are quite stupid.⁽⁴⁾ Lack of intelligence is no evidence for lack of ability to be conscious, nor for failure to actually be conscious at any given time.

More likely, although not stated explicitly, is that the author and the researchers he cites are a bit narrow minded. They think that only living organisms can be phenomenally conscious and arrangements of silicon and metal obviously can’t. But that is just prejudice. We think other people are conscious because they are like us, and we know that we are conscious. Animals such as dogs and cats appear to be conscious because they behave like us and we can imagine inhabiting their point of view and seeing the world as they do. There are certain limits to such imagination, of course. Can we imagine how the world would appear to a bat?⁽⁵⁾ Or an amoeba? It is even harder to imagine being an AI. But reality is not limited to what we can imagine.

A more plausible reason for doubting AI sentience, not mentioned by the author of the New York Times piece, is that a certain complexity of material substrate — neurons and brain cells in living beings — seems to be required for an organism to be conscious. And the more complex that substrate, the more vivid and intense the world appears to the organism and the more intelligent is its repertoire of behavior. On this view, because AIs lack such a complex substrate of living cells they can’t be conscious. But perhaps such complexity can be mirrored in non-living form. Perhaps it’s not the nature of the material substrate that counts but the complex patterns embodied in the substrate.

The authors disparage science fiction and accuse AI zealots of failing to distinguish science fiction from reality. But science fiction has much to offer. Consider the novels of Iain Banks known as the Culture Series. They are set in a utopian, post-scarcity society of humans, humanoid aliens and advanced superintelligent AIs living in artificial habitats spread across the galaxy.⁽⁶⁾ The non-human AIs are characters in the stories as much as the humans are. They give every indication of being not only highly intelligent but also quite conscious of their world.

If such a world were to come to pass, we could distinguish between AIs and humans only by means of their appearance, not by whether they were sentient or not. That time has not yet come, but it may well be on its way. We may agree that today’s AIs most likely aren’t sentient, but we have no infallible way to decide for sure. And we certainly can’t be sure that future AIs won’t be. A bit of humility is called for here, as well as a sense of wonder, which underlies science fiction and philosophy both.

 

---

 

Notes

(1) Metz, “A.I. Is Not Sentient.” All quotations unless otherwise cited are from this article.

(2) The English language has two terms that mean roughly the same thing, “conscious” and “aware.” The former is from a Latin root, and the latter is from Old Saxon. (See Dictionary.com, “Conscious” and “Aware.”) Many other languages have only one: “bewusst” in German and “consciente” in Spanish, for instance. The two English terms are interchangeable. The only exception to using them interchangeably is that sometimes “aware” connotes being informed or cognizant in a way that “conscious” does not. If you want to say that someone knows the rules, “She is aware of the rules” sounds better than “She is conscious of the rules.” But that is not the meaning in the sentence quoted.

(3) Block, “On a confusion about a function of consciousness.”

(4) Unattributed, “Top 10 Dumbest Animals in the World.”

(5) Nagel, “What Is It Like To Be A Bat?”

(6) Wikipedia, “Culture series.”

 

References

Block, Ned. “On a confusion about a function of consciousness.” Behavioral And Brain Sciences (1995) vol. 18, pp. 227-287. Online publication https://www.nedblock.us/papers/1995_Function.pdf as of 6 August 2022..

Dictionary.com. “Aware.” Online publication http://www.dictionary.com/browse/aware, as of 4 May 2016.

Dictionary.com. “Conscious.” Online publication http://www.dictionary.com/browse/conscious, as of 4 May 2016.

Metz, Cade. “A.I. Is Not Sentient. Why Do People Say It Is?” Online publication
https://www.nytimes.com/2022/08/05/technology/ai-sentient-google.html as of 6 August 2022. If that link doesn’t work, try this one instead:
https://www.nytimes.com/2022/08/05/technology/ai-sentient-google.html?unlocked_article_code=AAAAAAAAAAAAAAAACEIPuomT1JKd6J17Vw1cRCfTTMQmqxCdw_PIxftm3iWka3DFDm4fiPgYCIiG_EPKarskbtp2xDmdWN5MNqNqS_t1wetSeUxxTg3i6r21pKM4GQRn44SiQjFxmJvXQbEz9TKtZWC3dr0jlLf65hjXPX3tWaTbzSYrcw16pJVmJQn73yNfxaaWAfc1joclpYopA5l8RzkEYjDb_KW7TkUjZ6jVK03U-QI0WOpGWDDMnNH6674IcwVaC12uX2ooqC9nq4saYIVLSf65ex0we8P-gqETAnhqKOqqA54xT4vVkNZ6oP65_uCA6uRBZWYDCbN5PK4&smid=url-share.

Nagel, Thomas. “What Is It Like To Be A Bat?” The Philosophical Review, Vol. 83, No. 4 (Oct., 1974), pp. 435-450. Online publication http://www.jstor.org/stable/2183914 as of 29 April 2015.

Unattributed. “Top 10 Dumbest Animals in the World.” Online publication https://www.animalsaroundtheglobe.com/top-10-dumbest-animals-in-the-world/ as of 6 August 2022.

Wikipedia. “Culture series.” Online publication https://en.wikipedia.org/wiki/Culture_series as of 6 August 2022.