23 / 84
were more likely to re-offend than
whites.
All this has taught us that we need
to look very carefully at the data
used as input for AI systems. It
also shows us that it is important
for us to understand at all times
how an algorithm arrives at a
certain conclu-sion – and that we
can make adjustments to it. The AI
system cannot be a ‘black box’. New
procedures and checks are required
to ensure da-ta and algorithmic
transparency.
Yet it will probably not be possible
to make all data input and algo-
rithms transparent – which means
there will be times when we do not
know why an AI system may have
come to certain conclusions. So it
is important to introduce regular
tests for AI systems, which include
pay-ing a great deal of attention to
ethical issues, so that any problems
can be identified quickly.
How can we trust
‘them’?
So, how can we ensure that
everyone is able to trust the AI
system they come across on the
workfloor, at the doctor’s surgery
or on the road? First and foremost
by allocating some sort of approval
certificate, based on a regular audit
of the system. One example of this
are the elevators that we use every
day. We have no idea about how
their technology works, but we trust
them to be safe and work properly
(and we can do this because they
are checked regularly by people
who know what they are doing).
The way humans and AI systems
communicate and their predictability
can also help to build a relationship
of trust. For instance, take the traf-
fic lights that we use to cross the
road as pedestrians: they have
a pushbutton that enables us to
provide our own input (while a
camera should also be able to
detect the pedestrian anyway); we
then receive a signal that our input
has been registered; next, we wait
patiently be-cause we know that we
are using a predictable system that
will turn green within a maximum of
2 minutes; and in some cases, the
system even tells us how long we
will have to wait before crossing.
Will we still have a job in
2035?
There is a good chance that many
routine jobs will be taken over by
AI systems and robots. That may
range from working on a conveyor
belt in a factory, to making certain
medical diagnoses or working as an
ac-countant or in legal jobs. Even
tasks where a bit more creativity is
re-quired can be carried out by AI
Fig 4:
We need to try and achieve the same level of trust in AI systems that
we already have in smart traffic lights.
Fig 5:
Chess computers are a great example of what is possible with
AI. They have gone so far that players can be suspected of using chess
computers if the moves they make are strikingly original and creative.
New-Tech Magazine Europe l 23