The thing about large language models like GPT-3 and Lambda describing the experience of being self-aware is they can also describe the experience of being a squirrel.
https://www.aiweirdness.com/interview-with-a-squirrel/
@janellecshane
Leading questions are the new GIGO.
@feonixrift That and cherry picking 9 transcripts (5 with one system, 4 with another) to forge that better suits their argument.
@feonixrift *forge /one/ that..
@feonixrift @janellecshane i suddenly have a whole new appreciation for one old court rule of evidence
Yeah, there was something off about the "sentient" transcript and the linked article makes the point really clearly.
The "sentience" transcript seemed more like a human being talking calmly about the subject of sentience, rather than a machine suddenly realising it is alive.
@janellecshane Thanks so much for posting. This is a very convincing rejoinder to that other article.
@janellecshane @ghost_bird Reading this, it strikes me that GPT-3 is basically a bullshitting machine
And that makes me wonder, are we in the UK in fact being governed by a squirrel, a robot vacuum cleaner and part of the Chicago River? 🤷♀️
@transponderings @janellecshane @ghost_bird
It *does* sound like a politician trying to fill in time to avoid answering a question during an interview 😁
The transcripts (including the "sentient" one) were pretty polite and tame. What do these kinds of AIs do if the questioning is more confrontational, and demanding of a specific coherent answer?
and then, some people at least, will dutifully follow its advice.
reading tea leaves is hardly out of character for our species
@FediThing @transponderings @janellecshane @ghost_bird when i first saw similar counterfactual demonstrations my first thought was this is just like talking to a privileged man who just will not admit he doesn't know
@janellecshane
Nailed it. Turing's test doesn't fit purpose because we are easier to fool than we think. It lets us enjoy great books and suffer delusions minor and major in equal measure.
What kind of being is a self-aware #AI? Esp. one that was 'grown' from a good part of the full body of human knowledge. It will only be human-like because we gave it an interface express its sentience in human language constructs.
Indeed people being fooled will be a growing issue. Long before real sentience, being 'human-level' or beyond, is achieved (if ever), there'll be cults of people that succumb to the idea that this stage is reached. New religions, gods, etc.
doom scenario
@humanetech @edclayand @janellecshane gray goo but it's a techbro cult forcibly "uploading" (copying corpuses of then murdering) everyone until only bot zombies are left
@edclayand @janellecshane a turing test isn't a random chat but an adversarial test; it'd need a set of interviewers and a balanced pool of human and AI subjects, where the interviewers (and the human subjects) have strong incentives on getting which are which *right*.
These transcripts clearly show how to identify current systems; you can just lead them on in a way you never could an uncooperative human, so clearly they wouldn't even be passing.
@janellecshane worried this AI stuff is going to put hard working squirrels like me out of a job
@janellecshane this is whart I've been telling people all along, yet my sentience is still a matter of public discourse!
@janellecshane First AI with a fursona. I am here for this.
@janellecshane I notice it said nothing at all about the stark cognitive difference in a squirrel's time perception 😏 As with DALL-E, gpt-3 only knows what it has been told by the 'ignorant' masses in the data set, a giant mirror on our twisted social media scape.
good point. I think this gets towards more useful discriminants beyond "is not/is so" ethical ping pong
that said, even if squirrels are treated very differently from humans, they do enjoy some legal protections in many jurisdictions.
@janellecshane that is so funny, great read
@janellecshane So the argument is that AIs are not self-aware, they are merely actually flesh-and-blood (as it were) philosophical zombies, performing an indistinguishable simulation of self-awareness.
It's not going to be "indistinguishable" from the real thing, it's just that nobody's really tested it yet, so far as we know. Throwing it easy questions about how it's feeling today or whatever is not how you do that, if there's a reason to try.
@mike @janellecshane what do you say to people who doubt that you are a squirrel?
@IngaLovinde @janellecshane I turn my back on them and whip my tail left and right furiously for a few seconds before I scamper away.
@IngaLovinde @janellecshane A squirrel did this to my last night. :-( I was feeding her squash seeds and she was having trouble finding them in the grass. Once we had worked out that I was there to give her seeds and she was cool with that, I tried tossing them to the side onto a patch of bald earth instead of directly at her. She seemed to think I was throwing away the seeds instead of giving them do her and she gave me an angry little display and left. :-( Sorry squirrel!!
@janellecshane The real question now is, are squirrels sentient? 
@janellecshane
Wow. This is such a close parody of egocentric consumerist identity 'politics' - Do language modellers really think that this kind of sad weirdness is a model of language to adopt? Or is it just so predictable and narcissistic that it makes the life of a modeller very easy, and a minimalist Turing Test such a pushover?
@janellecshane report: Blake Lemoine has just been double fired by Google after breaking into a data center, removing several rack mount servers, and stashing them in a large oak tree
@janellecshane Rule 0 of the Turing test is that the one administering it must also be able to pass the test
@janellecshane
,I get to run and jump and play all day‘ is obviously a Disney Squirrel’s answer.
@janellecshane i can do that too. and i may or may not actually be a squirrel. 🙃
@janellecshane
I wonder how it would respond to challenges like, "But if you are a squirrel, how are you able to type?"
@janellecshane