Paul Cantrell
I like this general approach from @twsh, both as an antidote to magical thinking and hype, and also as a framework for having a useful conversation.
What exactly “contentful” means is an open (newly reopenes?) and extremely sticky question — but also a usefully focusing one.
https://scholar.social/@twsh/112898189891851050
Paul Cantrell
The Turing Test has been so widely muddled as an idea, it’s easy to lose track of its original point: not that there is in fact an objective test that ends the debate about intelligence, but rather that we should not rely on probing the mechanism to delineate what is or is not intelligence. “Intelligence is as intelligence does.”
Paul Cantrell
That approach, however, all but invites the current AI hype tactic of bamboozling humans into imagining they see intelligence: “It’s intelligence if it can fool you.” Thus the last ~60 years of intelligence mirages in AI. It’s always just around the corner!
3/
Paul Cantrell
The approach @twsh is taking (based on my ignorant, off-the-cuff reading of two posts, so…grain of salt here) seems to point the way toward being able to examine the internals of the allegedly intelligent thing while still respecting Turing’s argument that mechanism shouldn’t matter. Beauty may be entirely in the eye of the beholder, but conversation has •two• sides. Seems like there’s promise lurking here — at least to this philosophy dilettante’s ear.
/end
Paul Cantrell
@pointlessone @twsh
At the risk of speaking for Thomas:
The OP is about what it means to “say“ something, not what it means to “hear” something. See downthread re “eye of the beholder.”
PointlessOne :loading:
@inthehands @twsh I’m not convinced this framing is useful.
I agree with points 1 and 2, and even accept 3 as a logical conclusion. But there’s a counterexample: AIs produce coherent language. It’s often hard to tell the nature of the author of a piece of text produced by a LLM, sometimes impossible. So from the reader’s point of view the text says things. The fact that AI has no communicative intent, or that it has no coherent world model is irrelevant.
This is not unique to AI. People on a regular basis gleam meaning that was not intended by a speakers/writers.
Rich Felker
@inthehands That point is kinda wrong though. If there is any reasonably simple mechanism you can discover, such that the "intelligence" unravels as a con, that's important information.
Paul Cantrell
@dalias
Yes! That’s more or less the point of what’s downthread.
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.