glen wrote:
Loved the skit... and/but it made me think about the broader implications which made me "look in the horse's mouth" for an answer (to mix a parable or a metaphor):This is what I imagine y'all are doing when you chat with LLMs:Confusing ChatGPT With an Impossible Paradox https://youtu.be/1qbiCKrbbYc?si=V8U_mioTmlaDpynM At times, it smacks of a Monty Python skit.
The point was to take GPT beyond trying to relate it's own "gathered rumours" about how "a human claps hands" and consider what a (more broadly capable) AI system might need to do to get one step closer to actually "understanding" the clapping of hands. Many here may discount Hawkin's 1K brains model so this may be an(other) empty gesture but I find it a compelling model for thinking about how spatio-temporal (esp. tactile to visual) (functional) understanding of the physical world that brains with neocortices do. And maybe how that translates into understanding abstract concepts (a bit more of a stretch).
that's all (y'all)! ------------------------------------------------------------------------“What would it take for an AI to have an internal model, much like Jeff Hawkins' 1,000 Brains and Cortical Columns, to emulate the clapping of hands?”
00:14
ChatGPT said:
To emulate something like the clapping of hands using an internal model
inspired by Jeff Hawkins' "Thousand Brains" theory and cortical columns,
an AI would need a few key components.
First, it would require a distributed representation of the environment and the concept of hands, allowing it to form a model of how hands move and interact. This involves creating multiple parallel models (like cortical columns) that each predict and update their understanding of the hand's position, shape, and movement.
Second, the AI would need a sensorimotor integration mechanism, enabling it to predict and adjust its model based on sensory feedback—just as humans constantly refine their understanding of their hand's position through touch and proprioception.
Third, it would need a timing and sequence prediction capability to anticipate the sequence of events that constitute a hand clap: bringing the hands together, the moment of impact, and the resulting sound and feel.
Finally, it would need a way to compare its predictions to the actual outcome and refine its internal models accordingly, improving its understanding and emulation of the action over time.
In essence, it would involve a complex, layered network of predictive models working in tandem to generate and refine a coherent and dynamic representation of the action.
You said:
“How practical would it be to create such a thing?”
00:03
ChatGPT said:
Creating such a system is theoretically possible, but it's a massive
challenge in practice. Mimicking the "Thousand Brains" approach would
require a highly advanced architecture capable of running many parallel
models, each constantly predicting and updating in real time. This would
necessitate significant computational resources, advanced algorithms for
sensorimotor integration, and a robust feedback mechanism.
While some aspects of this are already being explored in neuroscience-inspired AI and robotics, achieving the full complexity of human-like sensorimotor predictions and adaptability is still at the frontier of research. Over time, as our understanding of brain-like computation and hardware capabilities improve, building something close to this concept could become more practical. For now, though, it's more of a long-term goal rather than an immediate reality.
OpenPGP_0xD5BAF94F88AFFA63.asc
Description: OpenPGP public key
OpenPGP_signature.asc
Description: OpenPGP digital signature
.- .-.. .-.. / ..-. --- --- - . .-. ... / .- .-. . / .-- .-. --- -. --. / ... --- -- . / .- .-. . / ..- ... . ..-. ..- .-.. 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/
