afternoon it’s great to be here
so will not talk to you about today
robots robots have just been in
industrial applications but there’s now
a huge amount of investment in robots
and artificial intelligence and this is
meaning that robots are moving out of
those kind of environments and we’re
seeing in robots into all kinds of
environments in human activity so you
can see them robots in agriculture so in
farming you can see robots in
transportation you can encounter a robot
in a shop you can buy a robot for your
home and you can buy a robot for your
child to play with robots are coming so
my question really and what I’m
interested in is how do we understand
robots when we interact with them when
we just encountered them how do we make
sense of them how can we calibrate our
trust towards them how can we understand
when we just encounter a robot what it’s
doing what its purpose is what it’s
going to do next what kind of
capabilities
does it have so how do we understand
robots autonomous machines that means
rapidly well in order to answer that
question we have to understand a little
bit about humans and how humans work and
a little bit about how evolution has
evolved us to be the way we are suits
life ah society or our planet all living
organisms over that’s bacteria plants or
animals have sense the world and they’ve
sense their internal state and they use
this information to make choices so for
example even a tree
will shed branches of factories in the
forests that could share branches if the
leaves on that tree are not getting
enough light and it will grow branches
up towards the light
so that’s intelligent behavior and
that’s the kind of broad definition of
intelligence that computer scientists
like me would try and use on the
understan machines as animals came along
a involved actively sense the world so
they keep an eye out for food for
predators and possible mating
opportunities so it’s really more
animals tough to make because they to
get their genes into the next generation
and over time those genes change the
animals started to manipulate one
another another by sending signs and
signals to each other so for example
these vultures you can see him the
vulture the bull neck flapping his wings
so he’s signaling those other vultures
maybe attracting a mate or maybe trying
to scare them off a lion above roar when
a lion roars in the forest it’s making a
statement it’s trying to manipulate the
minds of the other animals in the forest
to say it’s dominant it’s in charge it’s
the line this anglerfish is even more
sophisticated so you can see here the
bony protrusion on the front at the
anglerfish okay so it’s a lure it lures
to praying so that the praise they
thinks that’s that’s some food that the
prank Anita when he gave us comes
towards it angler fish eats it so this
is manipulation communication as
manipulation and of course what was
really able to understand is that the
prey it’s not just going to continue to
be manipulated the prey becomes smarter
so it evolves to understand when it’s
being manipulated now we aren’t we
humans we’re fiber social animals okay
and we we hide the social animals
brains the main reason that we have
these large brains so that we understand
the interactions with one another we can
only can make models in our minds of how
we interact with one another and we have
a one you really unique ability as
humans and that’s language
so although animals can make signs and
signals and make certain kinds of
communications one another we humans
have this much more evolved system
called language so for example I can
manipulate your your mind and place
something into your mind just by saying
a few short words just imagine if you
will a warm summer’s day by the beach
did you do that okay so I don’t know
exactly what you’re imagining but I’m
sure you’re imagining the smell of the
sea the warm breeze on your face the Sun
on your face sound of children playing
you’re imagining all kinds of really
quite complicated things that go
together to make up this image I
manipulated your mind into doing that
just by saying some sounds with my mouth
and the cost of that to me was
incredibly small however if I’m going to
manipulate you to your detriment then
you’re not necessarily thanks believe me
so if I say to you guys hey give me a
hundred pounds now I’m Rob were on a
really honest guy I’ll give you a
thousand pounds Mac next week anybody
going to do that no you’re not gonna be
that easily easily manipulated however
if I tell you something that you really
want to hear and I’m perceived as a
person of authority then it’s likely
that your gates for these men or at
least some of you announce believe me
right
maybe 52% of you will believe one side
of the story in 15 easements or do tend
to believe the other side will support
that yeah so so just think about that
for a minute so why don’t we all lie to
each other all the time okay so we have
this nervous mechanism it’s called
reputation if I lie and you catch me
lying then you’re telling your social
network that’s called gossip now the
result of that is that your social
status will go up and my social status
will go down now North social status is
actually held in my mind and my social
status is held in your mind so we have a
theory of mind of each other and that
theory of mine is incredibly complex in
us hyper social animals it’s the way we
understand each other it’s the way that
we have been able to work together in
society but that model is the way that
we understand many many things we do a
thing called counter form or fising
which is that we understand things
around us as if they were people so that
might be animals your pets you treat
them kind as if they’re people we do the
same thing with inanimate objects my car
just doesn’t like me today doesn’t want
to start
they might computer you know its crash
so this it’s it’s destroyed all my
violence days it’s becoming an active
agent so these machines and inanimate
we humans are also bitter one other
really really good thing which is that
we make tools lots of animals make tools
but we make really really sophisticated
tools and starting 250 years ago any
dust or weather
we start to make tools that mean that we
don’t have to do some of the things that
we used to do by hand
now that machines to do these things
more recently it starts to build
incredible machines that allow us to do
things that we could never do before so
we could never go out into space before
now we can do that and many many other
things that we just couldn’t do before
we can now do but until really very
recently
all of those machines were controlled by
us humans okay they were like tools
sophisticated tools but we control them
we switch them on we sort of salaf we
press all the buttons we control
everything now that’s changing and it’s
changing fast with the rapid advances in
artificial intelligence so today we have
these things these are intelligent
autonomous machines you’ll all know that
they called robots right now these
things are not science pictured these
are all product you can buy any of these
robots there and these these machines
can interact with us in a variety of
ways so they can sense us they they have
vision they can make sense of our
language to some degree they can measure
our emotional state so that robot on the
right there could respond to you of
apparently appropriately to your
emotional state and just think about
that for a minute how are we going to
understand these machines how are we
going to interact with them we’re going
to use our theory of mind the theory of
mind that we with the gains over
evolutionary time that we still share
with our hunter-gatherer ancestors
that’s the only real one that we have in
our head
in our heads to understand so that was
my hypothesis and we wanted to test that
hypothesis so I built this machine this
little robot I call it r5 it’s a really
simple little machine it’s Arduino based
and we know because we know basic
machine and it can do some basic stuff
so it can navigate around without
bumping into things as a heat sensor on
the front we can detect humans I think
that’s up to human and then sort of goes
through a process of detecting whether
that’s a human or not and if it detects
a human it flashes a green light if it
can’t it flashes a red light and it goes
office just as easily can find some more
humans we did some experiments with this
machine in various locations and we do
some online experiments as well with
this machine to see what people make of
it when they just observe it they just
encounter it they give them no social
Clues no or any kind of clues about
what’s going on and we find that people
come up with all kinds of crazy ideas
about capacity of the machine what it’s
doing how intelligence it is what its
objectives are what it means when the
lights flash what does it mean when they
stay still people have really poor
that’s a really big problem now this
that’s a bit risky yeah this ridiculous
army armless but if it’s not kind of
smiley face painted on the front of it
but do we know what it’s doing do you
know what that’s machines doing that
what’s it doing it it’s kind of small
thing looking up moving around and we
could just be really helpful you know it
could be a really helpful machine that
wants to direct us you know wherever we
want to go but could have an internet
connected camera they could be streaming
that information remotely all the video
that’s that is governing to who knows
where what purpose it could be recording
the audio it could be trying to take us
off to somewhere you know anything with
it really
we don’t really know it’s uh it’s an
opaque machine it’s not transparent it’s
not really telling us anything about
itself or what its intended purpose is
so I’ve been doing some experiments
we’ll be looking at how we can improve
that it’ll are five robot that you saw
earlier to try and make it more
transparent so in this first experiment
here on the right hand side you can see
there’s a video there means Ratan to the
robot and on the left hand side there’s
this display and this is a very simple
graphical representation of what’s going
on inside the mind of the robot okay
inside the action selection system
that’s making the decisions about how
the robot does the things that it does
and we showed this to people online and
then after the questionnaire we looked
for the difference in their
understanding how good is they want
mental model understanding when we have
have this transparency display as
opposed to just letting them just
observe the robot directly and we see
what scientists call a very significant
difference okay a significant
improvement in the understanding of the
robot but actually you trustingly people
don’t think that they understand it any
better if you ask the people in those
two groups
do you understand the robot yep pretty
much the same answer so they think that
they understand the robofoot this
transparency display is really
beneficial in helping them actually
understand what the robots doing in the
most recent experiment we’ve just
completed we’ve done something slightly
different so rather than a graphical
presentation of the other processes
that’s going on inside the machines
brain as it runs we have there’s sound
so it produces kind of sentences so
what’s happening is we have a kind of
window into the side of the machines
brain and we’ve taken out the processing
and we use that to generate sentences
it’s not a behavior off the machine
itself it’s kind of window into the mind
of the machine and the reason that guys
with headphones on is because a lot of
noise answers at the Bristol science
learning center and there’s a lot of
background noise so it’s use of
Bluetooth and then they can they can
hear that you see the robot clearly and
we had a similar significant difference
in improvement of the robot so we’re
just demonstrating a how people open
stone robots and big how we can make
those machines much more transparent
which is really important now these guys
these guys are robot designers all
robots are designed by machines by
humans right they’re not evolved they’re
designed by class humans and so what we
what we need is some guidelines to try
and make sure that these robot designers
are building the kinds of robots that
are transparent not robots that are
there to deceive us to trick us into
thinking they’re smarter than they are
to deceiving us about their capabilities
or their intentions but transparent
machines that can help us all in our
lives to do all kinds of really useful
things just like with the other
technology we use now there are UK
actually leads in this because we have
the principles of robotics and academics
lawyers people got together in 2010 and
produce this document called the
principle of robotics which has set fire
principles in it for how robot designers
should build robots but they’re just ID
lines they don’t have any teeth the I
Triple E which is a global standards
body is also producing a set of
guidelines for AI and
Tomasz machines and the European
Parliament is looking at law there’s a
draft legislation of the moments at a
very early stage to try and produce
regulations for how people should build
robots but at the moment you can buy
robots completely unregulated there are
no rules or regulations at the moment
about these autonomous machines which I
think is a real concern which is why I’m
here today to tell you about that and
just to start your thinking about what
kind of robots do you want to see in the
future do you want to see militarized
robots autonomous machines fighting in
wars do you want to see robots caring
for old people providing companionship
and care those role people do you want
to see robots in shops and offices what
do you want these people need to think
about what kind of robots they want and
we all need to make sure that we get the
robots that we want and not necessarily
the robots that are easy to make cheap
to build they can manipulate us possibly
about that thank you very much
[Music]
[Applause] [Music]