Press "Enter" to skip to content

How Evolution can help us make better computers | Terence Soule | TEDxUIdaho


[Applause]
visitation with association exercise so
I have here two words of nature or you
can think of it as biology and computers
or you can think of it as technology if
you want and I’m going to associate some
images and some words with those terms
so first we have a picture here lots of
different creatures pretty clearly
that’s an example of what we see in
nature and this is a rack of computer
servers so this is what we get for
computer technology if we think about
the word adapts that’s nature right
nature is very adaptable fixed so
computers once you have a piece of
hardware it tends to be fixed right you
see lots of variety nature imagine a
forest of trees those trees may all be
the same species but each one
individually is different so even within
the same sets of trees the same species
we get lots of variety may contracts
computers tend to be repetitive and
identical so if you have a problem a
task that you want done over and over
and over again in exactly the same way
you give it to a computer to do nature
learns you can teach a dog even an old
dog it turns out new tricks whereas
computers it tend to repeat the same
mistakes over and over and over again so
why this dichotomy between what we see
in nature and biology person who we see
with computers and technology well a big
part of the reason for that difference
is that things in nature evolved and
computers are programmed and so this
raises an interesting question which is
what if we could sort of break that rule
and allow computers
all seems like sort of a strange
question and to address it we have to
think a little bit about but what really
is an evolution and we can think of
evolution as a process with four major
components so the first component is a
population imagine if you will a
population of rabbits and within that
population of rabbits we have lots of
variation so there are faster rabbits
and slower rabbits there are rabbits
that are more camouflaged and blend and
better because of their colors and
rabbits that stand out more so we have
variation and then based on that
variation we have selection the classic
sort of survival of the fittest
the faster rabbits are better at
escaping the foxes and what that leads
to then because they can survive is
inheritance so those rabbits they escape
the foxes either by being faster or
being more camouflaged get to reproduce
and pass their genes on to their
offspring so these four components are
what makes up the process of evolution
and the question becomes can we apply
these to computers and computer
technology and I want to begin with an
example that I suspect everybody is
pretty familiar with which is video
games and you may be thinking wait video
games live video games well in order to
answer that question we have to sort of
use an example of a video game so this
is a video game called Darwin’s demons
and you can see it’s sort of a classic
shooter type arcade game right we’ve got
a whole bunch of aliens and the player
is flying a ship and shooting at them
and so how does this relate to evolution
well if we think of those four
components we have a population here we
have a population of opponents of aliens
and within that population you can see
there’s already a lot of variation some
of them look a little bit different they
move differently so their behavior is
different they may shoot more rapidly or
shoot slow
or they may charge the player or hang
out in the corners and this is pretty
true of all video games right you rarely
in this kind of video game do you have
to face the same opponent over and over
again I would get boring so instead we
put in variation and so in terms of the
components of evolution we already have
two of them a population and variation
and then the third one was selection so
you might ask well is selection taking
place here and if so where and the
answer is yes that’s the players job
right there shooting at the aliens
they’re doing the selection and we can
go a little further and say we’re gonna
in terms of fitness say aliens that
survive longer or shoot more or charge
the bottom of the screen get better
fitness so we have already three of the
four components we need for evolution we
have our population we have variation we
have selection what we’re missing is
inheritance so here’s two of our
opponents two of our aliens from the
game and you can see they have variation
so they look different their eyes are
different color they have a different
shape if we were to put them back into
the game we could see that they behaved
differently maybe they move differently
or they shoot at different rates or
different types of bullets so we have
all of that variation well it turns out
of course because they’re
computer-generated aliens all of those
things are really just controlled by
their code which is a bunch of numbers
and so we can think of the numbers that
define how they look and how they behave
as their digital DNA and so now that
we’ve said the way these creatures these
aliens these opponents in our video game
look and behave is based on this digital
DNA now we can put in inheritance we can
say okay the aliens that last the
longest or shoot the best are the ones
that get to reproduce and they get to
take their digital
DNA and combine it and pass it on to
create a new alien and we can even if we
want throw in mutation and put in some
random variation as well and so now we
have all of the components of evolution
put in to in this case a video game so
what happens when you do this well you
end up with a video game that starts by
looking like this
and after you’ve played it for a while
it looks like this and you can see the
aliens have evolved issue faster there’s
all sorts of different patterns
different strategies that they’ve
adopted and this isn’t limited to just
sort of our simple arcade style game and
simple creatures we can have much more
complicated creatures in our games so
here’s an example of a 3d monster that
we could put into a game and you’ll
notice that it’s changing it has a
digital DNA that controls things like
how long its legs are or what color it
is
or how long it’s tail or its tentacles
are and this is again all controlled by
a series of numbers it’s digital DNA and
in this case we’re sort of showing
random mutations so you can see the
variety that can come from it nor is
this idea limited to just video games so
nASA has used these same sorts of
evolutionary techniques to actually
evolve antennas that they use on
satellites and the idea there is you
have a population of satellite designs
and each excuse me a population of
antenna designs and each antenna gets a
Fitness based on how well it does its
task there are companies that choose
stock portfolios and mutual fund mixes
using an evolutionary technique in which
Fitness is based on the risk and the
reward for that particular combination
of stocks their researchers were
applying evolutionary techniques to
music and so the digital DNA for music
is basically the musical score and the
fit
is how good that piece of music sounds
and the better pieces of music gets
selected and get to mix their digital
DNA to create new pieces of music can
also be applied to things like robot
design and robot behaviors so just like
we can see the virtual creatures in our
games change we can actually see
physical robots change and many more
applications of this idea of evolution
and so we might ask what can we expect
in the future and one possibility that
we should be thinking about is that we
have computers that can adapt and learn
and most of all evolve
thank you
Please follow and like us: