Re: smart people doing stupid things

On 5/17/2024 7:11 PM, Edward Rawde wrote:

"Don Y" <blockedofcourse@foo.invalid> wrote in message
news:v28rap$2e811$3@dont-email.me...

On 5/17/2024 1:43 PM, Edward Rawde wrote:

Not sure how he managed to say master debaters that many times while
seemingly keeping a straight face but it reminds me of this:
https://www.learningmethods.com/downloads/pdf/james.alcock--the.belief.engine.pdf
>
One thing which bothers me about AI is that if it's like us but way more
intelligent than us then...

>
The 'I' in AI doesn't refer to the same sense of "intelligence" that
you are imagining.

 Strange that you could know what I was imagining.

People are invariably mislead by thinking that there is "intelligence"
involved in the technology.  If there is intelligence, then there should
be *reason*, right?  If there is reason, then I should be able to inquire
as to what, specifically, those reasons were for any "decision"/choice
that is made.
[Hint:  you can't get such an answer.  Just a set of coefficients that
resolve to a particular "choice".]
Additionally, these "baseless" decisions can be fed back to the AI to
enhance its (apparent) abilities.  Who acts as gatekeepers of that
"knowledge"?  Is it *really* knowledge?
I can recall hearing folks comment about friends who were dying of cancer
when I was a child.  They would say things like:  "Once the *air* gets
at it, they're dead!" -- refering to once they are opened up by
a surgeon (hence the "air getting at it").
Of course, this is nonsense.  The cancerous cells didn't magically react to
the "air".  Rather, the patient was sick enough to warrant a drastic
surgical intervention and, thus, more likely to *die* (than someone
else who also has UNDIAGNOSED cancer).

Have a look at this and then tell me where you think AI/AGI will be in say
10 years.
https://www.youtube.com/watch?v=YZjmZFDx-pA

"10 years" and "AI" are almost an hilarious cliche; it's ALWAYS been
"10 years from now" (since my classes in the 70's).
Until it was *here* (or, appeared to be)
Where it will be in 10 years is impossible to predict.  But, as the genie is
out of the bottle, there is nothing to stop others from using/abusing it
in ways that we might not consider palatable!  (Do you really think an
adversary will follow YOUR rules for its use -- if they see a way to
achieve gains?)
The risk from AI is that it makes decisions without being able to articulate
a "reason" in a verifiable form.  And, then marches on -- without our
ever "blessing" it's conclusion(s).  There is no understanding; no REASONING;
it's all just pattern observation/matching.
I use AIs to anticipate the needs of occupants (of the house, a business,
etc.).  Based on observations of their past behaviors.
SWMBO sleeps at night.  The AI doesn't know that she is "sleeping"
or even what "sleeping" is!  It just notices that she enters the bedroom
each night and doesn't leave it until some time the next morning.  This
is such a repeated behavior that the AI *expects* her to enter the
bedroom each night (at roughly the same hour).
Often, she will awaken in the middle of the night for a bathroom break,
to clear her sinuses, or get up and read for a while.
If she takes a bathroom break, the AI will notice that she invariably
turn on her HiFi afterwards (to have some music to listen to while drifting
BACK to sleep).
If she reads (for some indeterminate time), the AI will notice that she
turns on her HiFi just before turning off the light by her bedside.
It doesn't know why she is headed into the bathroom.  Or, why the bedside
light comes on.  Or, why she is turning on the HiFi.  But, HER *observed*
behavior fits a repeatable pattern that allows the AI to turn the HiFi
on *for* her -- when she comes out of the bathroom or AFTER she has turned
off her bedside light.
Due to the manner in which I implemented the AI, *I* can see the
conditions that are triggering the AIs behavior and correct erroneous
conclusions (maybe the AI hears a neighbor's truck passing by the house
as he heads off to work in the wee hours of the morning and correlates
THAT with her desire to listen to music!  "'Music'?  What's that??")
But, as you get more subtleties in the AIs input, these sorts of
causal actions are less obvious.  So, you have to think hard
about what you provide *to* the AI for it to draw its conclusions.
OTOH, if what you provide is limited by the relationships that
YOU can imagine, then the AI is limited to YOUR imagination!
Maybe the color of your car DOES relate to the chance of it
being in an accident!
An AI looks at tens of thousands of mamograms and "somehow" comes up
with a  good correlation between image and breast cancer incidence.
*It* then starts recommending care.  What does the oncologist do?
The AI is telling him there is a good indication of cancer (or,
a likelihood of it developing).  Does *he* treat the cancer?  (the
AI can't practice medicine)  What if he *doesn't*?  Will he face a
lawsuit when/if the patient later develops cancer and has a bad
outcome -- that might have been preventable if the oncologist
had heeded the AI's advice?  ("You CHARGED me for the AI consult;
and then you IGNORED its recommendations??")
OTOH, what if the AI was "hallucinating" and saw something that
*seemed* to correlate well -- but, a human examiner would know is
NOT related to the Dx (e.g., maybe the AI noticed some characteristic
of the WRITTEN label on the film and correlated that, by CHANCE,
with the Dx -- a human would KNOW there was no likely causal relationship!)