[Dprglist] The dead end of Artificial Intelligence

Murray Altheim murray18 at altheim.com
Tue Aug 4 16:18:14 PDT 2020

Related to this topic there's an article in today's New York Times:

     For Robots, It’s a Time to Shine (and Maybe Disinfect)
     The pandemic has turned cleaning and other mundane building tasks
     into a challenge, stoking interest in machines as cost-effective
     Lisa Prevost, 4 Aug 2020, The New York Times

It looks like the robotic cleaning industry is doing well because of the

    “Neo has the artificial intelligence capability where, as it’s moving
     along its original path, if it sees something in its way, it will go
     around it,” Mr. Cobb said. “If the obstacle is there the next day,
     Neo will incorporate it into its map.”

If being able to go around an obstacle and note its location is "AI" then
we've all been doing AI all the time and the term is pretty meaningless.
More to the point, if a object is in the same location as something from
the previous day, that would signify nothing to the robot -- or shouldn't.
It might be the same object, a different but similar object, a different
and dissimilar object, it might be a briefcase or a cat. Maybe a cat on
a mat? And incorporating this information into the robot's map? What value
is that if the object is subsequently moved or removed? (Sometimes this
robot logic is simply too difficult for me to comprehend...)


While at least avoiding the common anthropomorphisation of calling the
act of storing as "remembering" or data as "knowledge", this still
mistakenly posits some value in building such a map world-view, when
clearly the robot can interpret nothing about that change in its
environment and shouldn't be making any assertions based on that data.
For such a robot, if a door is open or closed an area that won't be
cleaned. If within the design/requirements of the robot that room should
have been cleaned but the door was closed, the robot could signal to a
human or to a door mechanism to open the door. But is that "artificial
intelligence"? That'd merely be several lines of code. Is such a judgment
based on the complexity of the program or something else? What is the
threshold for "Artificial Intelligence"? Quantitative or qualitative?

I do think these kinds of issues important in robotics, but bringing
up the subject of AI usually just muddies the waters, since there is
no accepted, canonical definition of what AI actually is, simply
because we still have no accepted, canonical definition of intelligence
itself. Or knowledge. Nor really even a pragmatic (or Pragmatic)
definition of either that we can use.

I studied in the field of Knowledge Representation (a sub-field of AI),
but found out pretty quickly that neither word is clearly defined, even
within the KR/AI community itseelf. I read Wittgenstein, Peirce, John
Searle, Sellars, Rorty, and Robert Brandom among others, and it's
fairly likely I was not clever enough to find my answer, though I also
suspect there simply is no good answer.

...which brings me back to Rodney Brooks, whose position is summarised
on his Wikipedia page [2] (see also: [3]):

  "Instead of computation as the ultimate conceptual metaphor that helped
   artificial intelligence become a separate discipline in the scientific
   community, he proposed that action or behavior are more appropriate to
   be used in robotics."

I'm okay with that.



[1] https://www.pdcnet.org/jpr/content/jpr_2004_0029_0113_0124
[2] https://en.wikipedia.org/wiki/Rodney_Brooks#Academic_work
[3] Elephants Don't Play Chess
     Rodney A. Brooks, MIT Artificial Intelligence Laboratory 1990
Murray Altheim <murray18 at altheim dot com>                       = =  ===
http://www.altheim.com/murray/                                     ===  ===
                                                                    = =  ===
     In the evening
     The rice leaves in the garden
     Rustle in the autumn wind
     That blows through my reed hut.
            -- Minamoto no Tsunenobu

More information about the DPRGlist mailing list