On May 29, 2023, at 11:01 AM, Russ Abbott <russ.abb...@gmail.com> wrote:
I saw that Rodney Brooks video. He claimed that transformer-based
software has no semantic content. I think that's an exaggeration. The
semantic content is defined by the token embeddings. So many of the
explanations of token embeddings overcomplicate the basic idea. Look
up wordToVec
<https://www.google.com/search?q=word-to-vec+&tbm=isch&ved=2ahUKEwju7IP4i5v_AhUEK0QIHbXrDycQ2-cCegQIABAA&oq=word-to-vec+&gs_lcp=CgNpbWcQAzIECCMQJzIECAAQHjIECAAQHlD0EVj0EWDpGWgAcAB4AIABWogBrwGSAQEymAEAoAEBqgELZ3dzLXdpei1pbWfAAQE&sclient=img&ei=S-Z0ZK6bDITWkPIPtde_uAI&bih=956&biw=1781&cs=0&rlz=1C1RXQR_enUS1008US1008#imgrc=aSLVbcgBsmbHKM> and
read some of the articles. Word-to-Vec was around before
transformers. But transformers are based on that idea. (One of the
keys to transformers is that the embedding space, including the
features themselves, is generated as part of the training.) The
embedding of all tokens in the GPT embedding space is the semantics.
It's amazing the extent to which that idea can be pushed and the
results LLMs produce!
_
_
-- Russ
On Mon, May 29, 2023 at 10:13 AM Jochen Fromm <j...@cas-group.net> wrote:
Yes, Rodney Brooks said something similar. He said "GPTs have no
understanding of the words they use, no way to connect those
words, those symbols, to the real world. A robot needs to be
connected to the real world and its commands need to be coherent
with the real world. Classically it is known as the 'symbol
grounding problem""
https://rodneybrooks.com/what-will-transformers-transform/
One could argue that this form of connectedness and embeddedness
leads eventually to self-awareness. First physical embeddedness,
then social embeddedness and finally self-awareness
1. Physical Embeddedness:
Agents who are embedded in a physical world are aware of the
world and move in it. To be embedded they need to be embodied.
Embeddedness leads to a grounding and a unique point of view
https://iep.utm.edu/husspemb/
2. Social Embeddedness:
Agents who are embedded in a world of social actors are aware of
other agents in the world and interact with them
https://mitpress.mit.edu/9780262518581/the-new-science-of-the-mind/
3. Self-Awareness:
Agents who are embedded in two worlds, the physical world and the
world of language become aware that they are actors in a world of
social actors by creating a link between the embodiment in the
physical world (the body) and the embodiment in the world of
language (the name or I)
-J.
-------- Original message --------
From: Russ Abbott <russ.abb...@gmail.com>
Date: 5/29/23 7:08 AM (GMT+01:00)
To: The Friday Morning Applied Complexity Coffee Group
<friam@redfish.com>
Subject: [FRIAM] A distinguishing feature of living entities
While watching my two little dogs run around our house, it struck
me that a feature that distinguishes living from non-living
entities is the apparent effortlessness with which living ones
navigate the world. Imagine how difficult it would be to build a
robot that could navigate the world so effortlessly. To make the
comparison a bit simpler, imagine how difficult it would be to
build a robotic cockroach.
When I asked ChatGPT whether anyone has built a robotic
cockroach, it came up with these examples. (I haven't checked to
see whether these are real projects.)
*
DASH: The Dynamic Autonomous Sprawled Hexapod (DASH)
robot, developed at the University of California,
Berkeley, was inspired by the rapid locomotion of
cockroaches. It has six legs and can move quickly on
various terrains using a simple control mechanism.
*
Harvard RoboBee: Although not specifically modeled after
a cockroach, the Harvard RoboBee project aims to develop
small, insect-like robots. These tiny flying robots are
inspired by the mechanics and flight capabilities of
insects and demonstrate similar agility and maneuverability.
*
iSprawl: The iSprawl robot, developed at the University
of California, Berkeley, was inspired by cockroaches'
ability to squeeze through small spaces. It uses a
compliant body design and six legs to navigate tight and
cluttered environments.
*
VelociRoACH: Developed at the University of California,
Berkeley, the VelociRoACH is a fast-running robot
designed to mimic the high-speed locomotion of
cockroaches. It utilizes a legged design and has
demonstrated impressive speed and agility.
__These mainly explore locomotion. Besides locomotion,
cockroaches notice when someone enters an area where they are
exposed. They quickly scuttle off to some hiding place. How do
they sense the presence of a new being? How do they know where
the hiding places are? How do they know how to move in the right
direction? How do they know how to avoid small obstacles and
fires? Etc.
One can argue that these capabilities are hard-wired in. But that
doesn't make it any easier. These are still capabilities they
have, that would be a challenge to build.
I became amazed at how well-connected living entities are to
their environments. They quickly and easily extract and use
information from their environment that is important to their
survival.
Man-made robots have nowhere near that level of embeddedness and
environmental integration.
Was it Rodney Brooks who said that we should build that sort of
connectedness before worrying about building intelligence into
our robots? Today that struck me as an important insight.
__
-- Russ Abbott
Professor Emeritus, Computer Science
California State University, Los Angeles
-. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
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/
