Standards in Artificial Intelligence

Nick Maclaren nmm1 at
Tue Nov 7 12:44:27 EST 2000

In article <qasnp3n1m5.fsf at>, James Hammerton <james at> writes:
|> nmm1 at (Nick Maclaren) writes:
|> > Neural net programming, for example, is neither an attempt to study
|> > the way the human brain works nor the use of the brain's principles
|> > in computing.  
|> I disagree, the aim of neural networks is to employ abstractions of
|> how the brain works in computing, e.g. using architectures that are at
|> least broadly speaking inspired by the organisation of neurons in the
|> brain. 

A few people may have that aim, but the majority use the techniques
for general optimisation and analysis and have little or no interest
in how the brain actually works.  Also, remember that the current
fad for 'neural network algorithms' PREDATES most of even our limited
understanding of how the brain actually analyses data!

|> > It is a collection of techniques using ideas that
|> > were taken from hypotheses about how the brain might work. 
|> Certainly some of it is like that, but the findings of neuroscience
|> and computaitonal neuroscience are not ignored. E.g. recent models
|> have incorporated (in abstract form) pulsing neurons where if they
|> pulse in phase it indicates that the neurons concerned are all
|> referring to the same object in some manner and the suggestions
|> that low levels of nitric oxide (if I recall correctly) seem to be
|> used to facilitate communication between neurons indepdently of their
|> direct connections (e.g. look up "gas networks"). The idea of using
|> networks of units with modifiable connections is itself an abstraction
|> from the organisation of neurons in the brain anyway.

Sorry, but it isn't.  It predates that by a long way.  I can't tell
you when the first self-modifying network systems were invented,
but it was before my time - i.e. not after the mid-1960s, and I am
pretty sure that it was in the 1950s.

|> > Nothing
|> > wrong with that, but "artificial intelligence" it ain't!
|> This seems contradictory to some stuff you said earlier. If
|> "artificial intelligence" ought to be about building intelligent
|> machines, then ISTM that most neural network stuff IS about that, even
|> if it often involves using techniques based on how the brain might
|> work.
|> Which begs the question, what exactly do you mean by "artificial
|> intelligence" in the above paragraph?

As in the normal English meaning of the word - which may be
pretty vague, but is at least a reasonable starting point.

Nick Maclaren,
University of Cambridge Computing Service,
New Museums Site, Pembroke Street, Cambridge CB2 3QG, England.
Email:  nmm1 at
Tel.:  +44 1223 334761    Fax:  +44 1223 334679

More information about the Neur-sci mailing list