Treatise on exactly WHAT intelligence is, and its relationship to language, is now in draft form and available from DEAC.
According to Eric Baum, it is the evolutionary
generation of a “compact description of the world” which enables animals, including humans, to prosper. According to Jeff Hawkins, it is the predictive power inherent in the repeating unit of our cortical architecture,
the cortical column, and how these columns interact and make up “words” that are invariant representations of
things. According to Paul Adams, it is a powerful learning algorithm, also based
on the cortical column, that extracts higher order coincidences and learns infrequently, in a burst like fashion. As a neuroscientist, these ideas are more intriguing to me than the millions of pages of psychobabble about
g and how the mind works (no offense intended, just trying to parse things here). These ideas intrigue me because intelligence is computation at the level of neural circuits and no amount
of fMRI data is going to explain the computations that made the following instance of intelligence (be it high or low intelligence)
The Story of “email JF”
Here is a note I had written a long time ago and just today ran across by chance:
“email: JF about pretnding to be home dept. friendly”
I mentally read this as
“email Joe Fetcho about being friendly” and was momentarily puzzled because there is no reason why I should want
to write to my colleague Joe Fetcho, a professor at another college, and discuss “being friendly” (either his
being friendly or mine) but in the flash of the instant it took to experience this thought, I realized with perfect conviction
that it was not Joe Fetcho (who is a very friendly guy), but another JF, also a university faculty member (and also a good
guy), but someone with whom I did intend to discuss friendship. On the surface
this may seem like a trivial thing, but this little realization (and a concomitant) decision is based on a staggering amount
of computation and memory and it was accomplished and evaluated in a flash (maybe a second or two; maybe a fraction of a second)
from my perspective. Now, any self-respecting (future) digital entity
should ridicule me, flaying me with its digital laughter, guffawing at my puny speed of thought, since a trivial, 2005 digital
agent, like Google, can search a billion web pages in a second, find 10 million relevant pages and sort them. But today’s digital entities are absolutely pathetic in
regards to carrying out the task that I performed so easily and casually in perhaps a fraction of a second. They lack anything that smacks of real intelligence: rote calculation and searching is where they excel.
So what is real intelligence? Being able to understand language in a practical sense (not necessarily a metaphysical
sense) would be a good start. There is some progress here, which corporations
are quick to seize on as they are increasingly replacing human agents with digital agents.
For example, after you speak to a digital agent (agent not entity) for some time and you have exhausted
its intellectual capacity (and your patience) you may (thank god!) be asked the question: “would you like to speak with
a live agent?” How big is the gulf between today’s
digital creatures and a future creature, one that would not instantly
exhaust your patience?
Consider my simple email:
“email: JF about pretnding to be home dept. friendly”
You will immediately recognize that email is being used as a verb, although it is more
properly viewed of as a noun. You probably suspected that JF were the initials
of one specific human being. You also recognized that “home dept.”
was probably a division of his workplace and had nothing to do with JF’s actual home.
Maybe digital agents will get this far soon, but what about “about being friendly”. This is where the calculation gets interesting. [Yes it is
true that AI researchers have so far failed to accomplish the first 3 tasks, but this fourth task is what I find most interesting.] How did I recognize that I was not emailing Joe Fetcho about being friendly
and how did I instantly conclude that it was indeed the other JF that I wanted to email?
My answer to this would be “complex”, based on past interactions with both JF’s, current situations,
my understanding of their behaviors or mine and how they or I interact with others in our respective home departments. I could explain much of this in a few paragraphs, but even that would be a superficial
explanation which would likely miss a lot of the underlying neural computation (I suspect).
The amazing thing in this quite trivial exhibition of “intelligence” (OK, I’m no rocket scientist)
is not its speed (which I think is still impressive) but its versatility. The
problem with AI today is that it does not matter how much faster we make computers go, we can never get them to accomplish
this trivial instance of intelligence until we come up with better ideas of what intelligence is and how to imbue machines
There is another interesting
aspect to this little example. It was at the bottom of a page of text that I
had written some time ago as notes for my dealings with yet another colleague at another university, in the context of evaluating
this other institution’s plans in the areas of neuroscience and computational biology.
At the end of these notes, I had switched to writing about an ongoing compulsion of mine, the text of which is pasted
verbatim (syntax errors and all):
“I’ve intensively studied the overlap between AI and comp. neuroscie. from the perspective a reall cellular/molecular/systems neurophysiologist.
I’ve worked the mind body problem and Eric Baums book had crticial insights, yet he still mssed the mark in some
respects…he didn’t explain how to get semantics from syntax…another way of phrasing the mind body problem,
sorta…I know exactly how to do that…and I can’t guarantee that we can make conscious machines, because I
don’t know what consciousness is, but I’d lay extremely heavy odds that we will make sentient machines and much
sooner than anyone things…and they will destroy us…
they won’t have compassion, empathy or a nurturing nature…or
most of them won’t—they will be programmed by governments and corporations to act with extreme greed and prejudice
to serve their owners….and their owners will always make them as extremely intelligent as possible to server the owners
needs….and then one day, poof: consciousness, self-awareness, thought…the mysterious flame
JF about pretnding to be home dept. friendly”.
What’s it all
about, Alzy? The problem of language is a vast research area, and it
is clear that the complexity of one’s written language correlates not only with various measures of intelligence, but
also with one’s prognosis for avoiding Alzheimer’s disease later in life.
Why should this latter relationship be? In the tragic late stages of Alzheimer’s
disease we lose our capacity in many intellectual arenas, from understanding how to prepare a meal to recognizing ones love
ones. While skills more motor in nature, including singing and walking, are better
preserved, this wrinkle does not detract from the claim that there is vast, widespread degradation of intelligence in Alzheimer’s. What does this tell us about intelligence and its neural substrate?
damage is widespread throughout cortex (including its effects on neurotransmitter systems that are equally widespread), the
intellectual decline can be attributed to a gradual or graded decline in cortical system performance. It is (perhaps) not so much a matter of individual systems (e.g. visual or executive planning) being degraded,
but more a matter of the decline of system integration: the vast arrays of iteratively hierarchical and vastly interconnected
systems no longer do those specific and general things that allow one to almost-instantly understand phrases
like “email JF about being friendly”. Language and intelligence are
all about being able to rapidly do millions of specific and general things as we make our way in the world.