Doh, don't know why I didn't spot this glaring connection between ML generated outputs and Frankfurtian bullshit.
Conversation
And thinking about it, this is a restatement of point that humans who are not concentrating are not general intelligences. "Not general intelligences" may be synonomous with bs producers/consumers. "Concentration" = truth-looped?
2
2
24
Bullshit in this epistemological sense is not a moral failure, but an incomplete cognition mode. It corresponds to the upstream part of what Daniel Dennett called the "multiple drafts" view of consciousness. First you confabulate, then you discriminate.
1
1
20
Ie the bullshit is necessary but not sufficient. You free-associate to produce output that has a cosmetic coherence, and then close the truth loop somehow in a downstream discrimination step before actual output. Bullshitters output intermediate output indiscriminately basically.
3
1
18
This is why embodiment is important. It is obviously important in closing "physics" based truth loops (if you think there is no obstacle where there is an obstacle, you will experience a collision). But it is also important in logic-based internal coherence truth loops
1
10
Even pure mathematicians working with symbols use subtle things like visual symmetry in written equations and rewrite rules that draw on intuitions from mechanical manipulation... ie physics-like embodied cognition.
2
1
9
In fact, I'd even argue that there is no variety of complete cognition (ie a generation + discrimination production pipeline with a truth loop) that can be disembodied. At the very least, your disembodied AI will need a simulation to live in to grow beyond pure bullshitting.
2
15
Replying to
yes, I think that's probably the difference between truth-tracking thinking and BS/confabulation. if you're truth-tracking, you have one or more "check: does this match [source of ground truth/itself]?" loops running.
1
1

