Computational autism
This novel approach aims
to merge the traditional autism research community with the researchers who
study the reasoning and perception capabilities of humans and machines, which
are actually corrupted under autism.
We target the introduction
of ideas from computer science, artificial intelligence and mathematics on the
possible mechanisms of reasoning to the autism practitioners. Autism
phenomena needs the thorough study of autistic behavior and reasoning, backed
by the strict representation of what it means to reason in a “normal” versus an
autistic manner.
Computational autism in
the news:
NewScientist
Magazine on logic-based autistic rehabilitation Hope lies in flair for
logic Duncan Graham-Rowe July
20 2002 issue |
|||||
|
Autism is a developmental
disability that typically appears during the first three years of life. The
result of a neurological disorder that affects functioning of the brain, autism
and its associated behaviors occur in approximately 3 of every 1000
individuals.
Autism
interferes with the normal development of the brain in the areas of reasoning,
social interaction and communication skills. Children and adults with autism
typically have deficiencies in verbal and non-verbal communication,
social interactions and leisure or play activities. The disorder makes it hard
for them to communicate with others and relate to the outside world.
They may exhibit repeated body movements (hand flapping, rocking), unusual
responses to people or attachments to objects and resist any changes
in routines. In some cases, aggressive and/or self-injurious behavior may be
present.
Does the above description of autism
symptomotology sound like a set of bugs (or possibly even adaptive features)
characteristic of a broken software system? We think it does. To
put it more precisely, we believe that useful insights may be gained by viewing
autism in humans via computational
How to find these bugs?
Can one talk about fixing them?
Psychological, biological,
linguistic loci of autism is very hard to find. It seems productive to focus on
the reasoning-specific locus, that is more experimentally viable and can be
done with higher strictness and accuracy at least for individual patients.
Phenomena of computational autism is expected to provide a convincing
background for both diagnosis and rehabilitation. The latter is connected with
teaching autistic children the internals of mental world.
How hard is it to teach
autistic children?
Indeed, it is very hard to
reprogram. AI researchers know that it is always hard to make a software program
to perform a behavioral thing that seems very natural and easy for humans.
Indeed, it is frequently even harder to explain to an autistic child even the
simplest possible mental reasoning act, to understand that other persons have
wishes and how to know what is inside a bag.
At the same time, it is
easier to introduce an autistic patient to a formal or mechanical word than to
the mental one.
Why is it of interest to
the CS & AI community?
We, the programmers and
computer scientists, are all aware of how much sophistication have to be
involved to teach a computer how to
accomplish certain simple physical, mental or perceptual tasks. These tasks
stimulated development of rather abstract disciplines of formalisms-based AI,
from nonmonotonic reasoning to frame problem, from formal grammars to
simulation of speech acts.
It is the case for autistic learning: it may be extremely hard
to teach an autistic child to do something very basic from the usual
perspective. So the special learning methodology comes into play.
Why are the advances in
computational autism helpful for the patients?
Novel technology can
directly target the corruption of reasoning means with high consistency,
efficiency, and availability, freeing the parents and rehabilitation personnel
from the routine training responsibilities.
What will result from
merging the psychological and computer science efforts on autistic training?
Software rehabilitation
means:
Is there direct link
between autism and logic?
Because children with
Asperger’s Syndrome tend to be logical thinkers it is tempting to try and use
logic to argue them out of their autism. I know. I have been there with my son.
Believe me.
It does not work like that! If it did we would have cured Asperger’s Syndrome
by now.
Accepting that the mental reasoning is one of the central issues in autistic phenomena, what is exactly corrupted under autism: axioms vs rules of inference, knowledge structure vs access to memory etc.?
This question was communicated by M.Minsky, AAAI FSS 99 on Simulation of Human Agents. Autistic reasoning can be described in terms of pure/applied calculi, logic/non-logic (domain-specific) axioms, prepositional / first-order / second order axioms and rules of inference. Hence, since autistic children can satisfactorily apply the rules of inference in the domains other than mental, we conclude than mental axioms (e.g. BDI model) is what should be blamed.
People with autism have troubles relating to people. What they need is more contacts with people. Why do we want them to sit in front of computers any more than they already do?
Usually, when both human and automatic agents learn, they first need to be introduced to the principles / rules of the new domain, and then have a set of training samples. This is the case for autistic rehabilitation: children have to be first taught the principles of the construction of mental world and its inhabitants, and then have a practice with interaction with them. In contrast to normal children, autistic ones have already failed to adjust themselves to the real mental world, and a computer environment serves as a means to introduce the principles of understanding mental world. After the basics are acquired, autistic children are encouraged to develop mental skills practicing with other people rather than with a computer.
We may proceed, considering human
and automatic agents from the uniform prospective…
Persons with autism may
possess the following characteristics in various combinations in varying
degrees of severity (Autism
Society of America www.autism-society.org ) |
||||
Inappropriate |
No real fear of dangers |
Apparent |
May not want |
Sustained unusual |
Uneven physical or |
May avoid |
May prefer |
Difficulty in |
Inappropriate |
Insistence on |
Echoes words |
Inappropriate |
Spins objects |
Difficulty in |
Explanation of each characteristic in terms of reasoning machinery is an open problem in logical Artificial Intelligence! |
Cohen, I. L. (1998). Neural network
analysis of learning in autism. In D. Stein and J. Ludick (Eds.) Neural networks and
psychopathology, pp. 274-315. Cambridge: Cambridge University Press.
Gustafsson, L. (1997). Inadequate cortical feature maps: A
neural circuit theory of autism. Biological Psychiatry, 42,
1138-1147.
O’Loughlin. C. & Thagard, P.
(2000). Autism and coherence: A
computational model. Mind & Language, 15, 375-392.
Townsend, J., Westerfield, M.,
Leaver, E., Makeig, S., Jung, T-P., Pierce, K. & Courchesne, E. Abnormalities of topography and composition of
ERP response in autism during spatial attention (.pdf, 1.4Mb). Cognitive
Brain Research 11:127-145, 2001.
Oliver, A., Johnson, M. H.,
Karmiloff-Smith, A., & Pennington, B. (2000). Deviations in the emergence of
representations: A neuro-constructivist framework for analysing developmental
disorders. Developmental Science, 3, 1-23.