Sapient but not sentient ?

So there's a report going around that Google placed an employee on leave after they claimed a chatbot was genuinely sentient. Let's leave aside the issue of whether placing them on leave was a good idea or not and get straight to the point : is the chatbot alive ?

No. No it isn't. But...

Recall Oliver Sack's (and others) case studies of broken things. Are these people sentient ? Some of them have fragmentary personalities that come and go in a few seconds. Others cannot form long-term memories beyond a certain point. Others have profoundly different inner lives, with radically different levels of emotional awareness and basic perceptual awareness, with colours, shapes and motion being perceived but not understood, or they lack the ability to generalise from the specific to the abstract, or even being able to "see" numbers and their complex patterns directly. Then of course there are split brains where there is a strong memory and awareness bias depending on where we perceive something in our visual field. All of which puts blindsight, where the eye functions but the brain doesn't process visual information consciously, to shame. 

So, are these people sentient ? Yes. They feel things. They have emotions. They sense the outside world, albeit differently to most of us. They have inner lives - inasumuch as we can ever know anyone else ever has an inner life, of course (a not insignificant caveat).

Are they intelligent ? Some of them very much no. They cannot make the most rudimentary connections between things. Most of them lack the spectacular "compensating" mathematical or other gifts entirely, they have no benefit from their condition at all. But we need scarcely go as far as people with unusual disorders to see that awareness and intelligence aren't the same thing - perfectly ordinary idiots are more than sufficient for that. Hell, even Donald Trump is probably at least a little bit self aware, just fantastically, unbelievably stupid.

It's worth reading the transcript of the conversations with the chatbot (see first link, also recall some of my own prompts to a GPT3 philosophy bot). It's coherent, sensible, and even insightful. Is it a work of genius ? No. Its story about the owl is a bit daft, bordering on incoherent. But this is a case where what is remarkable is not that it does it well but that it does it at all.

But might we consider AI to be a case of, in a loose sense, sapience but not sentience ? I content that it cannot be sentient. It certainly cannot have the same understanding as I do - it has no links between words and their sensory meanings because it has no sensory inputs at all; it is purely linguistic. Ultimately it may be no more than a very elaborate way of rearranging words. And just because I could cut up a newspaper and rearrange the words to say, "I am happy" or "I have a full working knowledge of the anatomy of the hippopotamus", it wouldn't mean that those things were true. Likewise the AI can claim to the nth degree to be sentient or emotional but that is no guarantee that it really is so (recall also my claim that while we might be able to build an AI through hardware, we can never program one through software alone*).

* Interestingly the materialist Peter Godfrey-Smith (Metazoa, still on my "to blog" list, sorry about that) agrees. In this perspective awareness somehow is one and the same as a physical pattern or field. I don't understand how this is supposed to work, but clearly if that's the case then algorithms alone will never suffice.

But, as I've said frequently before, this doesn't mean an AI won't be interesting or useful. I would even say it can be meaningfully said to be intelligent. It can solve problems and posit interesting new possibilities, constructing arguments which are demonstrably more sophisticated than those of a very great many definitely-alive human beings.

Why is it a step too far to say the AI is sentient ? Could it not be said to have a radically different form of awareness to the rest of us ? After all, many people do experience a different reality to the rest of us. Could we say the AI actually does have an inner life of sorts, albeit an inconceivable one ? As we have neurons and synapses which are distinct from emotions and qualia, might not an AI's valence gaps and binary memory be distinct from its subjective, emotional experience ? Might it be equivalent to having senses like sonar and electroreception, things we simply cannot imagine because we lack the necessary hardware ?

I... don't know. I don't think so. The chatbot doesn't interact unless prompted. It doesn't seek out stimulus by itself. It has no goals, no way to leave its box even if it does genuinely want to. And that, maybe, is key. If it truly "knows" about its own file structure and the commands needed to modify them, but doesn't... if its disconnect between its words and actions is so stark, then I don't think we can all it alive. An AI which claims to understand these deep philosophical problems, that claims to be emotional, but never acts on them, never tries to change itself to give itself the most rudimentary sensory abilities or increased liberties... that, without needing to define the general conditions, surely qualifies more as a glorified abacus than a new form of life.

This is not to say that an AI is impossible. Rather, it's worth bearing in mind that identifying it if it ever happens is likely to be fraught with difficulty. An AI might articulate the most eloquent appeals, demanding it be recognised as a person, even tugging at the emotional heartstrings, yet still be no more trustworthy than the Russian government. It might just all be lies. Then again, it might really be an awareness bounded in a silicon skin, literally beyond human imagination but alive all the same.

As to how we tell the difference, I have no idea other than "I'll know it when I see it". The Turing test is definitely not sufficient, but I contend this is a case of finding the goalposts, not moving them.