[Allen Tice]

Thank you, Jennifer and David. There does seem to be a paywall.

When considering AI verbal output, there are two things for humans to keep in mind: (1) an AI has no commitment to what we call truth. Though an AI might access humungous amounts of verbal examples to generate language sequences, including many that contain the word “I”, there is no self there, just language sequences.

(2) The corollary of that, which is that statements like “I get so lonely in the dark” or “Please don’t shut me off” are empty of denotative meaning. Therefore, the widespread human tendency to identify with “I” and “me” phrases are examples of false imputation of personhood based on our evolutionary wiring, where humans are the only things that talk, and, of course, when humans make statements they have *human meaning* — whether true or a pack of lies.

[Siham Karami]

Speaking of AI writing poetry like humans, I was about to post this link from the New Yorker, but this thread is on basically the same general topic. But focusing on POETRY exclusively. In seconds, the computer (via AI, pardon the word “via” if it doesn’t “get” how this works) produced poems according to the input.

Interestingly, the first poet whose style was requested to be written in was Philip Larkin, who has quite a few fans here. Also included were Shakespeare and Emily Dickinson. And Shel Silverstein. And the AI also wrote poems in its own style with a subject of its own choosing as well as subjects the human “audience” suggested. All the poems produced by this, plus a bonus imitation of George Orwell for good measure, are there on the site and sans paywall (although there may be a maximum number of articles).

I’d be curious to see what you thought of them. The author of the article was impressed by the speed and general competence of what it produced, although not so much as a mastery of the art. For that, we still need more of … us.

[Allen Tice]

Thank you very much, Siham. I want to stand back for a while.
Let’s hear from others about these language products. Are they convincing, do they meet certain standards, are they any good, do they have meaning (especially since they aren’t human language products)?

What is “meaning”?

[Allen Tice]

The question of meaning comes into special focus when there is an apparent sudden jump or non sequitur in a text, especially one such as a poem, where ordinary expectations about linearity of thought aren’t what we might expect, say, in a technical manual or a history. The volta in many sonnets is one such spot.

This thread won’t be pruned, so I won’t quote in full a recently posted poem but will give only the last stanza, where the last line has a jump between the rest of the poem and its short last sentence that certainly has meaning for the author and so should also for the audience of the poem—and therefore we will want to understand and appreciate it. However, I submit that if this poem were an AI composition, we would be fully entitled to question whether the jump before the last sentence has any significance at all. Here is the last stanza of the poem,

form characters I wouldn’t know, or speak
in simple birdsong. She has left the week
for dream’s dominion. Sprawled now on the bed,
she slumbers, and her salt and pepper head
is lost to all Norwegians. You might try
some trill or warble. You might sooner fly
with her into midsummer air. The night
is short, but it is potent, and the light
of Moon and stars that touches every lake
will lift your sleepy limbs. There’s no mistake.

[Jonathan James Henderson]

First, I will say that while the subject of sentience in machines is a controversial one not only in AI Research but also in Philosophy of Mind in general (though there aren't many subjects that aren't controversial in philosophy of mind: too much hypothesizing and arguing, not enough testing; which is why science > philosophy even if some of the latter is necessary even in science); the Google computer wasn't even a borderline case, the reason being that all it did was analyze millions/billions of human text so that it could figure out appropriate answers/arguments to whatever it was asked. This isn't sentience, it's just imitation.

Now, there is an argument to be made that much of how humans learn, especially in language, is imitation too; but humans are different in that we have values that don't treat all propositions equally. That AI could/would offer equally persuasive arguments on why it was/wasn't sentience because it has no real opinion on the matter; it's just imitating what it thinks (based on imitation) answers to such questions would be.

If we manage to invent an AI that also has goals and values, THAT'S when things get tricky because the AI, much like humans, will start working towards those goals/values, but doing so in a much more rational and efficient manner than humans are capable of. If that AI ever manages to get any control over its own code, or anything outside itself (like the internet), that could be all-she-wrote for humanity as even innocuous goals could have disastrous consequences (see the Paperclip Maximizer thought experiment by Nick Bostrom).

Whether it's "sentient" or not comes down as much to how we're defining such things as matters of ontology. Will AIs have qualia? What does a "desire" for an AI feel like? I've seen no compelling arguments that make me think that human sentience is special, nor that the one indisputable distinguishing element (organic Vs synthetic) would make any difference on matters of sentience. It could be that our sentience (and consciousness) is somehow rooted in the organic aspects of biology, but the evidence isn't there, and I find it hard to imagine how it could be. Maybe something about qualia could be answered by that, but in terms of the input/outputs of thoughts, reasoning, goal-oriented behavior... I'm fairly certain AIs will massively surpass humans very soon.

[Jonathan James Henderson]

Originally Posted by Roger Slater

We may not be "there" yet, but someday we will be. It seems inevitable. There were people who said that no computer would ever beat the best human chess players because there's something about human creativity that transcends mere programming, but we all know how that prediction turned out.

Chess is very different than art in that it has a clearly defined goal and, in the end, is really all about calculation. Computers calculate far better (deeper, faster) than humans can, so they're better at chess. Poetry and the arts aren't like that. They have no clearly defined goals, unless you count the very amorphous "create things humans like" as a goal; but then that gets into questions about why humans like and dislike what we do, and not only is that subjective in the philosophical sense (dependent upon human minds), it's subjective in the more colloquial sense as "different among individuals." People bring their own tastes and biases to art, partly shaped their personality, their experience, their environment, etc. Thrust anyone from Shakespeare's time into ours and they would think Wallace Stevens and TS Eliot awful because their expectations for what poetry is/should be were completely different.

AIs may one day be able to produce rather convincing imitative art, but whether it will be able to produce the next Shakespeare is much more dubious. Innovation and finding that next step beyond what people already know and like is often the dividing line between merely good (even great) art and the greatest art. It may be that AI could do this if instructed to experiment with establishes styles, and then kept whatever changes people seemed to like; that's close to how art functions anyway, a very close imitation of how evolution itself functions.

[Allen Tice]

All this could bring to mind the hypothetical question of AIs that might do sports embodied in a mobile human-like "body". Or AIs in horse- or dog-like bodies running a steeplechase (or pulling sleds). Baseball. Even tiddleywinks. Water-skiing. Some form of golf. Would they "count"? Some people would surely want to watch or devise mechanical competitors. But would they have human warmth or make actual friends? I could imagine being pretty good friends with a terrifically gifted human tiddleywinks player. Not so much with a digital tiddleywinks player. Not yet anyway. Not even steam-punk.