Hm. Now that you say it that way, how would laughter come through? It would be nearly
trivial to add LoL or emoticons to the lexicon. My guess is we'd still see through it a
bit. I predict the bot would use it like a polite Boomer. GenX and later start getting a
bit wonky with when they use phrases like "lulz". And I can't even cap with the
kids on twitch these days. I have no idea what they're saying most of the time.
Surely, though, they've trained some large language models on texts with onomatopoeia,
though. Back in the late '80s, we'd use tokens in brackets like [grin], [ahem], or
[pffft]. I still do, though my use has changed. Interjections like "Awesome!"
are clearly there. It can't be much of a stretch to add mimicry of physiological seeming
things.
The Sophistry of dialogs like Socrates' is another matter. Recognition of the
absurd, riddles, paradox, would be an interesting thing to test. I just
prompted my GPT-3 (primed with biomedical literature):
gepr: "Interleukin and posture walk into a bar ..."
['The bartender says "We don\'t serve your kind here."']
On 6/13/22 15:06, Jochen Fromm wrote:
Yes, humor is important. Good point. Laughing is one of the things we do and
apes do not. For me this is where it starts to get interesting: when we look at
the things we do that apes do not, like language, culture, art, or writing
systems. I mean before the first civilizations in Mesopotamia, ancient Greece
and ancient Egypt appeared there were just clans fighting against each other to
determine their place in the pecking order. Primates do this too.
We have this constant drive to resolve inconsistencies which is related to the
confirmation bias. Every joke starts with an inconsistency that is resolved by
an insight. Maybe we need just one basic mechanism to create a self-supervised
agent that gets smarter bit by bit: artificial curiosity, i.e. a mechanism that
seeks new or inconsistent information and rewards the resolution of
inconsistencies. A bit like science itself.
-J.
-------- Original message --------
From: glen <geprope...@gmail.com>
Date: 6/13/22 23:14 (GMT+01:00)
To: friam@redfish.com
Subject: Re: [FRIAM] Google Engineer Thinks AI Bot Has Become Sentient
IDK, I wouldn't say the dialog was indistinguishable from a human. When I ask people
things like "Do you have feelings?", they respond pretty aggressively or
defensively. While I agree that all the sentences were well-formed and sensible (SSI),
they lacked the reflective quality of actual human responses. Plus, there wasn't any
humor as far as I could tell. You'd expect that in a conversation with such ridiculous
questions. That would be true even if, especially if, you were talking to a kid or a
typically blue collar sort. [⛧]
That's why it read, to me, like one of those fake dialogs intended to teach some lesson
or other. And it wasn't even Socratic. This is where Aaronson's comment ("but can I
run my own tests?") plays in. Meno or Euthyphro might *seem* indistinguishable from
a human ... but they're not, they're fantastically designed to render the just-so
condition the ideologue intends. Perhaps Lahontan's Kondiaronk was different?
[⛧] I once picked up a hitchhiker on my way home from work back in TX. Since the ride was
quite long, we discussed quite a bit. As we were driving through town, I commented that
most of the people looked, to me, like they were asleep ... metaphorically. The hitcher
said, "They look awake to me", literally.
On 6/13/22 13:40, Jochen Fromm wrote:
> I think the capabilities of large language models are really impressive. The
language of these models is not grounded, as this article says, but in principle
it is possible to do it.
>
https://techmonitor.ai/technology/ai-and-automation/foundation-models-may-be-future-of-ai-theyre-also-deeply-flawed
>
> Take for example a robot, connect it to the Internet and a large language
model, and add an additional OCR layer in between. The result? Probably creepy and
uncanny, but if it works we would most likely think such an actor would be
sentient. The replies in the LaMDA dialog transcript look indistinguishable from a
human.
>
https://cajundiscordian.medium.com/is-lamda-sentient-an-interview-ea64d916d917
>
> -J.
>
>
> -------- Original message --------
> From: glen <geprope...@gmail.com>
> Date: 6/13/22 17:14 (GMT+01:00)
> To: friam@redfish.com
> Subject: Re: [FRIAM] Google Engineer Thinks AI Bot Has Become Sentient
>
> "Remarkable" in the sense of "worthy of remark"? Yeah, maybe.
>
> LaMDA: Language Models for Dialog Applications
> https://arxiv.org/abs/2201.08239
>
> Personally, I think we can attribute Lemoine's belief in LaMDA's sentience
is an artifact of his religious belief. It's not exclusive to Christianity,
though. One of the risks of the positions taken by those who believe in the
reality of things like Jungian archetypes is false attribution. And it's not
limited to anthropomorphic attribution. To the person with a hammer, everything
looks like a nail. Even if such beliefs have some objective utility in some
contexts, that utility is not likely to be that transitive to other contexts.
>
> I suppose this is why I'm more sympathetic to the (obviously still false in
its extreme) behaviorist or skeptical position (cf
https://onlinelibrary.wiley.com/doi/abs/10.1111/phpr.12445). I.e. it's completely
irrelevant whether or not you *claim* to have feelings and emotions. What's needed
for knowledge (justified true belief) is a parallax pointing to the same
conclusion, preferably including some largely objective angles.
>
> An objective angle on LaMDA might well be available from IIT operating over
some (very large) log/trace data from the executing program. *That* plus the bot
claiming it's sentient would give me pause.
>
> On 6/12/22 08:28, Jochen Fromm wrote:
> > A Google engineer said he was placed on leave after claiming an AI
chatbot was sentient. The fact that he thinks it would be sentient is remarkable,
isn't it?
> >
https://www.businessinsider.com/google-engineer-thinks-artificial-intelligence-bot-has-become-sentient-2022-6
--
ꙮ Mɥǝu ǝlǝdɥɐuʇs ɟᴉƃɥʇ' ʇɥǝ ƃɹɐss snɟɟǝɹs˙ ꙮ
-. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
FRIAM Applied Complexity Group listserv
Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom
https://bit.ly/virtualfriam
to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
FRIAM-COMIC http://friam-comic.blogspot.com/
archives: 5/2017 thru present https://redfish.com/pipermail/friam_redfish.com/
1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
