Name: Navigating the future - Unravelling the complexities of AI risk management
Uploaded: 2023-12-08T11:24:10.681Z
Duration: 867.2
Description: At the RiskMinds International conference 2023, we get an exclusive insight into the risk management implemented into AI. Discussing this are Lokke Moerel, Global Expert New Technologies & Cyber at Tilburg University and Nina Schick, Gen AI Expert and Author.

Navigating the future--Unravelling the complexities of -AI risk management.mp4

Navigating the future - Unravelling the complexities of AI risk management.vtt

ENGLISH

{{00:00:04}}
The biggest risk is the existential
risks everybody talks about, but what

{{00:00:35}}
I find is that by talking about those
big existential risks, we forget

{{00:00:41}}
the risks that are here today, and
that we need to solve to not get

{{00:00:46}}
to that level of existential risk.

{{00:00:48}}
So I find the whole discussion
on the existential risk not on.

{{00:00:54}}
We should focus on what is
happening here and now, and I tell

{{00:00:57}}
you, we can do something about
the risk that AI creates now.

{{00:01:01}}
So it is about Accuracy, the
hallucinations of AI, we need

{{00:01:07}}
to get that under control.

{{00:01:09}}
We need to solve the privacy issues,
where fake or wrong bios get out.

{{00:01:17}}
I saw one example where a professor
was accused of sexual harassment.

{{00:01:24}}
It wasn't true.

{{00:01:24}}
Do you see?

{{00:01:25}}
Those type of things, it can be
addressed and should be addressed now.

{{00:01:29}}
AI is a bit like the first car without
proper brakes, and that's where we are.

{{00:01:34}}
We're part of a Nash or
global how do you say that?

{{00:01:41}}
Experiment.

{{00:01:42}}
And we should work hard to get
that to get that under control.

{{00:01:46}}
I think that when you speak about AI
risk, the first thing to do is Understand

{{00:01:52}}
why this technology is so different
from any technologies we've seen before.

{{00:01:57}}
And really, it's because these AI
systems are now capable of not only

{{00:02:02}}
augmenting human intelligence, but also
uncovering intelligence that was unknown.

{{00:02:09}}
So when it comes to risk this is
a dual use technology, so there

{{00:02:14}}
is no doubt it can be weaponized.

{{00:02:17}}
But we have to distinguish
between near term risks and

{{00:02:20}}
catastrophic existential risks.

{{00:02:23}}
And when it comes to near term risks,
it's things like Mis and disinformation,

{{00:02:28}}
transparency, trust, using algorithms
to enhance cyber attacks, and the

{{00:02:34}}
catastrophic risk, which is something
that dominates the media narratives,

{{00:02:38}}
is more about, okay, what do we
do, or is there, is it inevitable

{{00:02:44}}
that we get to a point where the
systems are so much more intelligent

{{00:02:48}}
than us humans, their creators,
that we can no longer control them.

{{00:03:00}}
AI can be validated and tested for
accuracy, for bias, and I see the big

{{00:03:07}}
accountants, firms and other other
companies working on all types of

{{00:03:11}}
auditing mechanisms to see what whether
it's accurate, where it goes wrong, etc.

{{00:03:19}}
The whole point is that a lot of the
AI is a is, proprietary and cannot be

{{00:03:25}}
tested unless the companies allow it.

{{00:03:28}}
Allow that.

{{00:03:29}}
And I think the AI principles that came
out of the White House require that type

{{00:03:35}}
of transparency and also accountability.

{{00:03:39}}
And that is what we need to strive for.

{{00:03:41}}
Not only transparency to regulators so
they can test it, but also, for instance,

{{00:03:46}}
the Universities, so that researchers
can actually look into it and, validate.

{{00:03:53}}
AI systems are going to have
to be tested and validated.

{{00:03:58}}
But because all of what is happening
right now is still so nascent, the, we

{{00:04:03}}
were talking about an AI winter only a
few years ago and AI was really still

{{00:04:09}}
the realm of a few obscure academics
and what is now not only possible but

{{00:04:14}}
seems inevitable only a few years ago
was in the realms of science fiction.

{{00:04:19}}
So the answer to that is that
AI testing and transparency and

{{00:04:25}}
accountability, these systems and
processes don't really exist yet.

{{00:04:30}}
And that is going to be a challenge
over the next few months and years.

{{00:04:34}}
for enterprises and governments and
individuals to find out how exactly

{{00:04:40}}
to do that, because it's going to be
absolutely essential going forward.

{{00:04:50}}
The economic risk is also
a risk if you don't do it.

{{00:04:54}}
There was research of Harvard
showing that your associates, your

{{00:05:01}}
consultants work better with AI
and the AI doesn't add that much.

{{00:05:06}}
to your really good associates,
but actually to the not that good.

{{00:05:10}}
It actually has a very
large incremental value.

{{00:05:15}}
So if you don't apply it
your, productivity and your

{{00:05:19}}
output will be less good.

{{00:05:22}}
And, that is an economic challenge.

{{00:05:25}}
If you don't start looking into that
being said, it means that your associates

{{00:05:31}}
need to learn how to work with AI.

{{00:05:34}}
And people say you're replaced by AI,
but you're not replaced by AI, you're

{{00:05:38}}
replaced by somebody who can work with AI.

{{00:05:41}}
And what I notice also in the RISC area
is that everybody with AI, everybody

{{00:05:48}}
needs to upscale and rescale, but
it's also the Risk function needs, it

{{00:05:53}}
applies equally to the risk function.

{{00:05:55}}
They need to know what it is,
how you can work with it, and

{{00:05:58}}
train how to work with it.

{{00:06:00}}
So that is an economic reality
we all need to get at grips with.

{{00:06:09}}
Then you said the financial risk.

{{00:06:13}}
Personally what I'm most, phishing
emails, that is the start of a lot of

{{00:06:18}}
cyber events, and that will be with
AI, it will be phishing on steroids.

{{00:06:24}}
You will have voice imitations trying to
trick you into giving your your children

{{00:06:30}}
will be used or imitated to get you to
give your passwords or make transfers.

{{00:06:38}}
I've seen fake avatars in the
metaverse trying to lure you if it's

{{00:06:42}}
your CEO walking and looking like
your CEO and talking like your CEO,

{{00:06:46}}
inviting you in a team's group you,
you may go let's put it that way.

{{00:06:53}}
But also I find if on both sides of the
trading, you have AI and We have already

{{00:06:59}}
algorithm, or I should say software,
trading software that, the flash trading.

{{00:07:05}}
And that software is already
so complicated that nobody

{{00:07:08}}
understands it anymore.

{{00:07:10}}
Just imagine that is AI on
all sides of the equation.

{{00:07:13}}
They may find ways to make
money and trade by tricking the

{{00:07:20}}
system, if you get my point.

{{00:07:22}}
And there we need controls
that I haven't seen yet.

{{00:07:25}}
So that is where I think the
whole financial system can be at

{{00:07:29}}
risk if we do not control that.

{{00:07:33}}
So I think that when it comes to a general
technology like AI if you look at historic

{{00:07:41}}
trends with technology, it starts as niche
and then it changes economic systems.

{{00:07:46}}
A good example of that is computing
and the associated technologies

{{00:07:50}}
of the internet, the smartphone.

{{00:07:52}}
and how quickly they've changed
the economic reality in our

{{00:07:55}}
own individual realities, it's
going to be the same with AI.

{{00:07:59}}
First of all, I should say that I think
that there's going to be a tremendous

{{00:08:02}}
amount of economic abundance associated
with the AI revolution, but when it comes

{{00:08:08}}
to risks, it's much more that as every
entity, whether you're an organization or

{{00:08:15}}
a nation state, races to gain competence.

{{00:08:20}}
The AI arms race is on how destabilizing
could that be for the world order?

{{00:08:25}}
How destabilizing could
that be for economies?

{{00:08:27}}
And moreover, there are certain
entities and players who are

{{00:08:33}}
emerging as foundational in this
AI revolution because they actually

{{00:08:37}}
own the infrastructure, right?

{{00:08:39}}
The infrastructure that's needed
to power the AI revolution.

{{00:08:43}}
So there's a very important question
about is all the reward and is all

{{00:08:47}}
the financial upside going to be
monopolized in the hands of a very few.

{{00:08:52}}
So those are the kind of economic
risks I associate with AI.

{{00:09:04}}
I, as I said, I think the existent,
existent existential risks are

{{00:09:10}}
maybe not overblown, but it
track, it detracts us from talking

{{00:09:14}}
about the things we can do today.

{{00:09:16}}
If we do our job today, those
risks will not materialize.

{{00:09:20}}
But by talking up on those material
existential risks all the time, it

{{00:09:25}}
becomes so impossible to tackle.

{{00:09:28}}
That if you feel overwhelmed, it's
like with the environment that,

{{00:09:34}}
you can't, there's nothing you
can do as an individual anymore.

{{00:09:37}}
That is nonsense.

{{00:09:38}}
I think we need to do our job today to
make sure we don't get in that scenario.

{{00:09:44}}
And it irritates me if CEOs and
everybody makes it so big that it it

{{00:09:53}}
discharges them of the responsibilities
of doing a good job today.

{{00:09:58}}
There is so much misinformation around
AI, and again, the starting place is

{{00:10:03}}
this is a dual use technology, right?

{{00:10:05}}
So just as it has the promise to
actually solve some of the biggest

{{00:10:11}}
challenges known to mankind, whether
that's climate change, whether that's

{{00:10:16}}
challenges and breakthroughs in
medical research AI holds that promise.

{{00:10:22}}
Likewise, there is a lot of attention
and focus on the potential existential

{{00:10:28}}
risks, the catastrophic risks
associated with artificial intelligence.

{{00:10:32}}
And whilst we have to have that
conversation, which is hypothetically

{{00:10:36}}
possible, that we get to the point where
AI emerges as this existential risk.

{{00:10:42}}
There is still too much focus on
this risk in the media narrative.

{{00:10:47}}
Again, we have to distinguish from the
hypothetical, catastrophic future risks,

{{00:10:54}}
which, by the way, is not the risk that
most AI researchers are concerned about,

{{00:10:59}}
to actual near term risks, which have
nothing to do with kind of AGI and The

{{00:11:03}}
robots taking over, even though that is
the predominant narrative we hear about.

{{00:11:15}}
There's a discussion whether those
models were properly trained and etc,

{{00:11:19}}
But if you do not start looking into this
now, AI, generative AI is here to stay.

{{00:11:26}}
It's not going away.

{{00:11:27}}
So you will have to relate to it.

{{00:11:29}}
So you will need to experiment
and see where it can be applied.

{{00:11:33}}
without undue risk.

{{00:11:36}}
And that is well possible.

{{00:11:39}}
But if you don't start now, it,
you will be behind in a few years.

{{00:11:43}}
And What I'm concerned about is that the
AI developments are, as we see them now,

{{00:11:49}}
the Gen AI developments are very much U.

{{00:11:53}}
S.

{{00:11:53}}
players, and here in
Europe, and including the U.

{{00:11:56}}
K., we need to develop our own LLMs.

{{00:11:59}}
Why?

{{00:12:00}}
Because those LLMs in the U.

{{00:12:02}}
S.

{{00:12:02}}
are very much trained on U.

{{00:12:04}}
S.

{{00:12:06}}
data.

{{00:12:07}}
And we see that the LMS work better
in the US than they do in Europe.

{{00:12:11}}
Why?

{{00:12:11}}
Because we have a whole body of
documents in and in, in English,

{{00:12:17}}
or in Dutch or in German, and
they are o often not taken along.

{{00:12:22}}
So we need to invest very heavily
in those, in our own models to know

{{00:12:28}}
what they're doing, to do a proper
job, and also being not so dependent.

{{00:12:33}}
If we have issues with big tech
today, they will be again on

{{00:12:38}}
steroids the coming years if we
leave it to them to master LLM.

{{00:12:42}}
So first of all, this technology is here.

{{00:12:45}}
It's inevitable.

{{00:12:47}}
And if you look again at the history
of general technologies, there is no

{{00:12:52}}
way you can stop the diffusion and the
democratization and the engulfing of

{{00:12:58}}
everything by general technologies.

{{00:13:00}}
A great example is again,
Computing and the technologies

{{00:13:03}}
of the information revolution.

{{00:13:05}}
All of that happened in the
space of about 30 years.

{{00:13:08}}
Now we stand at the precipice of the
AI revolution, which implications, in

{{00:13:14}}
my view, are far more profound than the
implications of the technologies of the

{{00:13:19}}
information age, and I believe are going
to disperse even quicker than any other

{{00:13:25}}
technology we've known in the past.

{{00:13:27}}
By not engaging, you're not
engaging with the reality.

{{00:13:30}}
You miss out on the opportunities.

{{00:13:33}}
There are tremendous opportunities,
but you also miss out on mitigating the

{{00:13:37}}
risks, because there are risks associated.

{{00:13:40}}
And we absolutely have to think
about how do we contain those risks.

{{00:13:46}}
And the challenge is that is
not something a single entity

{{00:13:51}}
or a single institution can do.

{{00:13:53}}
We have to take a cross societal approach
involving government, regulators, but also

{{00:14:00}}
industry, civil society, and academia.

{{00:14:03}}
It's really a question of how
do we want to build our society.

{{00:14:07}}
In an age where technologies like AI and
associated exponential technologies like

{{00:14:14}}
quantum computing are so dramatically
and so quickly going to change the

{{00:14:19}}
reality of what it means to be human.

The biggest risk is the existential
risks everybody talks about, but what

I find is that by talking about those
big existential risks, we forget

the risks that are here today, and
that we need to solve to not get

So I find the whole discussion
on the existential risk not on.

We should focus on what is
happening here and now, and I tell

you, we can do something about
the risk that AI creates now.

So it is about Accuracy, the
hallucinations of AI, we need

We need to solve the privacy issues,
where fake or wrong bios get out.

I saw one example where a professor
was accused of sexual harassment.

Those type of things, it can be
addressed and should be addressed now.

AI is a bit like the first car without
proper brakes, and that's where we are.

We're part of a Nash or
global how do you say that?

And we should work hard to get
that to get that under control.

I think that when you speak about AI
risk, the first thing to do is Understand

why this technology is so different
from any technologies we've seen before.

And really, it's because these AI
systems are now capable of not only

augmenting human intelligence, but also
uncovering intelligence that was unknown.

So when it comes to risk this is
a dual use technology, so there

But we have to distinguish
between near term risks and

And when it comes to near term risks,
it's things like Mis and disinformation,

transparency, trust, using algorithms
to enhance cyber attacks, and the

catastrophic risk, which is something
that dominates the media narratives,

is more about, okay, what do we
do, or is there, is it inevitable

that we get to a point where the
systems are so much more intelligent

than us humans, their creators,
that we can no longer control them.

AI can be validated and tested for
accuracy, for bias, and I see the big

accountants, firms and other other
companies working on all types of

auditing mechanisms to see what whether
it's accurate, where it goes wrong, etc.

The whole point is that a lot of the
AI is a is, proprietary and cannot be

And I think the AI principles that came
out of the White House require that type

Not only transparency to regulators so
they can test it, but also, for instance,

the Universities, so that researchers
can actually look into it and, validate.

AI systems are going to have
to be tested and validated.

But because all of what is happening
right now is still so nascent, the, we

were talking about an AI winter only a
few years ago and AI was really still

the realm of a few obscure academics
and what is now not only possible but

seems inevitable only a few years ago
was in the realms of science fiction.

So the answer to that is that
AI testing and transparency and

accountability, these systems and
processes don't really exist yet.

And that is going to be a challenge
over the next few months and years.

for enterprises and governments and
individuals to find out how exactly

to do that, because it's going to be
absolutely essential going forward.

The economic risk is also
a risk if you don't do it.

There was research of Harvard
showing that your associates, your

consultants work better with AI
and the AI doesn't add that much.

to your really good associates,
but actually to the not that good.

It actually has a very
large incremental value.

So if you don't apply it
your, productivity and your

If you don't start looking into that
being said, it means that your associates

And people say you're replaced by AI,
but you're not replaced by AI, you're

replaced by somebody who can work with AI.

And what I notice also in the RISC area
is that everybody with AI, everybody

needs to upscale and rescale, but
it's also the Risk function needs, it

They need to know what it is,
how you can work with it, and

So that is an economic reality
we all need to get at grips with.

Personally what I'm most, phishing
emails, that is the start of a lot of

cyber events, and that will be with
AI, it will be phishing on steroids.

You will have voice imitations trying to
trick you into giving your your children

will be used or imitated to get you to
give your passwords or make transfers.

I've seen fake avatars in the
metaverse trying to lure you if it's

your CEO walking and looking like
your CEO and talking like your CEO,

inviting you in a team's group you,
you may go let's put it that way.

But also I find if on both sides of the
trading, you have AI and We have already

algorithm, or I should say software,
trading software that, the flash trading.

And that software is already
so complicated that nobody

Just imagine that is AI on
all sides of the equation.

They may find ways to make
money and trade by tricking the

And there we need controls
that I haven't seen yet.

So that is where I think the
whole financial system can be at

So I think that when it comes to a general
technology like AI if you look at historic

trends with technology, it starts as niche
and then it changes economic systems.

A good example of that is computing
and the associated technologies

and how quickly they've changed
the economic reality in our

own individual realities, it's
going to be the same with AI.

First of all, I should say that I think
that there's going to be a tremendous

amount of economic abundance associated
with the AI revolution, but when it comes

to risks, it's much more that as every
entity, whether you're an organization or

a nation state, races to gain competence.

The AI arms race is on how destabilizing
could that be for the world order?

RiskMinds & QuantMinds

Expert Insight

At the RiskMinds International conference 2023, we get an exclusive insight into the risk management implemented into AI. Discussing this are Lokke Moerel, Global Expert New Technologies & Cyber at Tilburg University and Nina Schick, Gen AI Expert and Author.
