00:08Now AI is a great thing,
00:10because AI will solve all the
problems that we have today.
00:16It will solve employment,
00:19it will solve disease,
00:22it will solve poverty,
00:25but it will also create
new problems.
00:36The problem of fake news is going
to be a million times worse,
00:41cyber attacks will become
much more extreme,
00:46we will have totally
automated AI weapons.
00:52I think AI has the potential to
create infinitely stable dictatorships.
00:58This morning a warning about
the power of artificial intelligence,
01:03more than 1,300 tech industry
leaders, researchers and others
01:07are now asking for a pause
in the development
01:09of artificial intelligence
to consider the risks.
01:45Playing God,
01:47scientists have been accused
of playing God for a while,
01:54but there is a real sense in which
we are creating something
01:58very different from anything
we've created so far.
02:05Yeah, I mean, we definitely
will be able to create
02:10completely autonomous beings
with their own goals.
02:13And it will be very important,
02:15especially as these beings become
much smarter than humans,
02:20it's going to be important
to have these beings,
02:25the goals of these beings
be aligned with our goals.
02:45What inspires me?
02:52I like thinking about the very
fundamentals, the basics.
02:59What can our systems not do,
that humans definitely do?
03:06Almost approach it philosophically.
03:10Questions like, what is learning?
03:13What is experience?
03:16What is thinking?
03:20How does the brain work?
03:30I feel that technology
is a force of nature.
03:37I feel like there is a lot of similarity
between technology and biological evolution.
03:46It is very easy to understand
how biological evolution works,
03:50you have mutations,
you have natural selections.
03:53You keep the good ones,
the ones that survive
03:58and just through this process you are going
to have huge complexity in your organisms.
04:06We cannot understand how
the human body works
04:08because we understand evolution,
04:11but we understand the
process more or less.
04:14And I think machine learning
is in a similar state right now,
04:17especially deep learning,
we have a very simple rule
04:21that takes the information
from the data
04:23and puts it into the model and
we just keep repeating this process.
04:27And as a result of this process
the complexity from the data
04:30gets transferred into the
complexity of the model.
04:34So the resulting
model is really complex
04:36and we don't really know exactly how
it works you need to investigate,
04:40but the algorithm that
did it is very simple.
04:44ChatGPT, maybe
you've heard of it,
04:48if you haven't
then get ready.
04:51You describe it as the first spots
of rain before a downpour.
04:56It's something we just need
to be very conscious of,
04:58because I agree it is
a watershed moment.
05:01Well ChatGPT is being
heralded as a gamechanger
05:05and in many ways it is, its latest
triumph outscoring people.
05:10A recent study by Microsoft
research concludes that GPT4
05:14is an early, yet still incomplete
artificial general intelligence system.
05:35Artificial General Intelligence.
05:38AGI,
05:41a computer system that can
do any job or any task
05:45that a human does,
but only better.
05:54There is some probability the AGI
is going to happen pretty soon,
06:00there's also some probability
it's going to take much longer.
06:03But my position is that the probability
that AGI could happen soon
06:08is high enough that we
should take it seriously.
06:14And it's going to be
very important
06:15to make these very smart
capable systems
06:18be aligned and act
in our best interests.
06:27The very first AGIs
06:29will be basically be very,
very large data centres.
06:33Packed with specialised
neural network processors
06:37working in parallel.
06:40Compact, hot, power
hungry package,
06:44consuming like, 10m homes'
worth of energy.
06:48You're going to see dramatically
more intelligent systems
06:53and I think it's highly likely
that those systems will have
06:55completely astronomical
impact on society.
06:59Will humans actually benefit?
07:02And who will benefit
and who will not?
07:42The beliefs and desires of the first
AGIs will be extremely important
07:49and so it's important to
programme them correctly.
07:53I think that if this
is not done,
07:56then the nature of evolution,
of natural selection,
08:01favour those systems prioritise
their own survival above all else.
08:17It's not that it's going to actively
hate humans and want to harm them,
08:23but it is going to be too powerful
08:26and I think a good analogy would
be the way human humans treat animals.
08:32It's not we hate
animals, I think
08:33humans love animals and
have a lot of affection for them,
08:37but when the time comes to build
a highway between two cities,
08:42we are not asking the
animals for permission
08:45we just do it because
it's important for us.
08:49and I think by default that's
the kind of relationship
08:51that's going to be between
us and AGIs which
08:56are truly autonomous and
operating on their own behalf.
09:18Many machine learning experts,
09:19people who are very knowledgeable
and very experienced,
09:21have a lot of scepticism about AGI.
09:25About when it could happen and
about whether it could happen at all.
09:30Right now this is
something that just
09:31not that many people
have realised yet.
09:35That the speed of computers
for neural networks, for AI,
09:39are going to become
maybe 100,000 times faster
09:43in a small number of years.
09:51If you have an arms race dynamics
09:53between multiple teams
trying to build the AGI first,
09:59they will have less time make sure
that the AGI that they will build
10:03will care deeply for humans.
10:09Because the way I imagine
it is that there is an avalanche,
10:12like there is an avalanche
of AGI development.
10:15Imagine it, this huge
unstoppable force.
10:21And I think it's pretty likely the entire
surface of the earth will be covered with
10:26solar panels and data centres.
10:34Given these kinds of concerns,
it will be important that AGI
10:40somehow build as a cooperation
between multiple countries.
10:48The future is going to be
good for the AI regardless.
10:53It would be nice if it were
good for humans as well.
