The thing that drove me was this
question, right: how does the brain work? And sort of
the engineering part of me at some point just took over and
said, you know, well, it’s a whole bunch of little devices,
and we actually know a fair amount about how each of
those individual devices work. SPAUN is a model, so it’s a piece
of software. And what we’re trying to do is simulate how the brain functions.
So just like you might simulate a circuit, a digital circuit or something,
we want to make sure that each of the devices — which,
in the case of the brain, is a neuron — is simulated properly.
So we write some equations down that describe how it functions,
and then we put hundreds or thousands or millions of
them together and make them communicate in a way that
is appropriate to reflecting the kind of structures that you
find in the brain and to performing some specific function,
which we know that the brain performs. On the medical side, it gives us
an understanding of how biology actually results in cognition.
Right? So what is — what is it that neurons are doing together
to give rise to people being able to remember lists of items,
for example. And so we can think about what kinds of
interruptions at the biological level — so what
kinds of diseases — might change some of the behaviours
that we observe. In recent work, we have taken
out some of the very simple cells and replaced them with
much more sophisticated cells, where the neurons actually
have spatial structures. And what this lets us do is
new kinds of experiments with models like SPAUN.
So for instance, we can essentially virtually introduce
a drug. And because SPAUN has very high-level psychological
functions, even though we’re introducing them at
the biological level, we can look at the effects
at the behavioural level. We can’t do this now, but at
some point we may be able to scan your brain or do
some behavioural tests, and that will let us build a model which
is specific to you. And then we can actually tailor the kinds of
neurological interventions to the things that we know
about your own brain. On the more technological side,
when we’re building models like SPAUN, we’re actually
sometimes getting functions out which are not very
natural for computers to do. If we extract those functions
and put them into machines, then we have to start worrying
about, you know, how we’re going to interact with those
machines, what kinds of machines we do want to build
and we don’t want to build, and other sorts of sociological
questions about what happens if those machines start taking
over a lot of the jobs that we typically expect
people to do. All of the simulation tools that
we’ve built, all the models that we’ve built, all of the designs
for robots and so on are things that we just release to the public.
And the reason is that what we’d really like to do is to have
not only our lab but many labs around the world adopt these
kinds of methods so we can build something like a grand,
unified model across many labs across many countries.
Because I ultimately think that’s the only way we’re going
to be able to tackle the kind of complexity that we observe
in the biological brain.