The next talk.
Artificial Intelligence, Ethics and Law:
Can EU Regulations Protect Humanity?
By Guiido Noto La Deiga,
chair of IP and Privacy Law
at University of Sterling,
and member of the European Commissions expert group
on AI and data in education and training.
Professor Noto La Deiga
will be here with us via Zoom
and he will analyze the effectiveness
of various initiatives created by
the European Commission in order to protect humanity
and humans from the threats and dangers of AI.
Enjoy.
[audience applauding]
Good afternoon, everyone,
and thank you so much for the very kind invitation.
I wish I could be there in person
unfortunately, I got Covid, which is not ideal,
but I really wanted to participate,
so I’m really thankful for to [indistinct]
and all the team at Photo Vogue
and Vogue Italia for giving me the opportunity
to intervene, to speak remotely.
So I’m gonna try and share my slides.
Let me know if you, if it doesn’t work.
I am assuming that they’re working.
You can see them right now.
So I am a lawyer by training.
I’m a legal scholar, so I’m an academic as well.
I’m Sicilian, but I live in Scotland.
And I had the great opportunity to work
with the European Commission
as part of the European Commission
expert group on AI in education.
But obviously, as lawyers love to do,
there is a disclaimer that here
I’m not representing the official views
of the European Commission,
I’m representing my own views and my own research.
So I wanna just share with you
a little bit about what’s going on
at the European level in terms of initiatives,
trying to minimize effectively
the risks posed by artificial intelligence.
I’m gonna, in the first part of the presentation,
I’m gonna consider the kinda ethical initiatives,
the initiatives around, AI ethics
and then I’m gonna move on to the law,
what the actual law, not just ethics,
but the law or proposed legislation
in the European Union is saying about AI.
And finally also learning from the title of the event.
I’m gonna try and chart new territories
and understand whether effectively
we can rely on just regulation to protect
humanity from AI.
Spoiler alert, the answer to the question is no.
There are so many different
ethical AI initiatives at the moment.
Ethical AI is a big and growing field.
You will see a lot of charters
and manifestos and guidelines
that try and effectively nudge developers
and other stakeholders so that
they develop AI systems that are ethical,
that reflect the values of our society.
This is easier said than done,
as you would expect.
I’m gonna share very briefly
my own experience as part of
the European Commission Expert group on AI and education.
So in September, 2020,
the European Commission published
the Digital Education Action Plan 2021-2027.
And as part of that action plan,
they set up an expert group and I was part of it,
whose main task was to publish
some the European ethical guidelines
on the use of AI in education,
which we did.
So you can find them online very easily.
You can reach out to me, I can,
if you don’t find ’em,
I’m very happy to share them.
And the purpose of these guidelines
is to kind of spell out some common fears
that we have around AI,
but also address the ethical considerations
and requirements for the responsible use of AI.
Provide practical advice to particular educators,
but more general stakeholders
and discuss emerging competencies.
What do we need to teach our students,
our children about AI?
We distilled four key ethical considerations,
four ethical considerations that we always need
to keep in mind when we think about
developing, adopting, using AI.
These are human agency, fairness,
humanity and justified choice.
I think these are pretty self-explanatory.
So I’m just gonna say a word about
the idea of justified choice,
because maybe that’s one expression
that might be less familiar.
Justified choice means that before,
before de developing an AI system,
before adopting an AI system, using it, et cetera,
we need to ask ourselves whether that choice is justified.
Has it, has the choice been taken in a transparent way?
But more importantly,
has the decision making behind
that choice been participatory?
I think a lot of you,
there will be a lot of photographers,
a lot of people working in the fashion industry,
the kind of a press, et cetera.
I think you’ve all been affected
and wondering how to,
the extent to which you’ve been affected by
ChatGPT and other forms of AI.
But I am pretty sure that no one
or nearly no one in that room has been asked,
what are your fears about it?
What would you like the EU to do for you?
Nobody has really consulted people about
those people that are mostly affected
by AI about what they want
in terms of legal change for example.
And I think that’s a
big issue that we should tackle.
And I will go back to it perhaps later in the talk.
So that’s the kind of general framework.
Based on that general framework,
we have identified seven key requirements
for an AI system to be ethical.
And these are pretty much the same requirements
that the high level expert group on AI
had already considered a couple of years before us.
So we learned from lessons to a large extent,
I’m sure that you are,
well, you’re probably all familiar
with these concepts
and just gonna go through them very quickly,
maybe say one thing or about one of them.
So these key requirements for an AI system
to be ethical are human agency and oversight,
transparency, non-discrimination,
societal and environmental wellbeing,
privacy, data governance,
technical robustness and safety and accountability.
For example to explain a little bit
what some of these, or at least
what transparency for example, means.
A key question is, for example,
there are all these systems
that generate automatically in theory,
some beautiful or some beautiful pictures,
some beautiful photographs
or some beautiful videos.
Transparency means to ask
what kind of data has been used
to train that machine that is creating,
allegedly automatically these beautiful pictures?
Because chances are that type of AI
has been trained with photographs
that were covered by copyright
and nobody asked you for permission
and nobody compensated you about,
for this type of activity.
So transparency is really, really important.
And the difference, I suppose, between our work
and the work of the high level expert group was that,
they’re more focused on the idea of a checklist.
So, you know, you have a checklist
and you tick your boxes, as in,
yes, I’ve considered transparency
in developing this AI system, et cetera.
Whereas what we are focusing on
is actually on guiding questions.
We want people, educators, anyone
really dealing with AI to ask questions,
to be able to ask critical questions
before choosing and adopting some form of AI.
I’m gonna finish this initial part
of my talk related to AI
and ethics with some criticism.
So it’s true that it’s a good thing that we see some,
things like the ethical guidelines
that I just talked about
and all these different, different
ethical charters manifestos on AI,
they’re not a bad thing in and of itself,
but they’re often used as forms of ethics washing.
Companies that effectively behave
and continue behaving in ways that are unethical.
But they say, you know,
they use these ethical initiatives to say
and to pretend that they are better
than they actually are.
The most dangerous thing is when these companies,
the famous kind of big tech,
the larger corporations
that are behind main types of AI,
and they’re usually US based, or China based,
when these companies are saying we don’t need new laws,
we don’t need regulation
because we are already,
complying with our ethical principles.
But I dunno if there are lawyers in that room,
but I suppose for everyone that thinks about the difference
between ethics and the law, the main difference is,
that ethics is really not binding.
So these companies can say that
they are abiding by these high level ethical principles,
but in reality they don’t have to
because ethical principles are not binding.
Only the law is binding.
And there are other issues with ethical AI
that I don’t need to go into much detail.
Maybe it’s something that we can explore during the Q&A.
For example, issues around around
equality, diversity and inclusion,
the need to avoid forms of digital colonialism.
So again, ethical AI it’s a growing field,
it’s not all bad,
but it’s a dangerous thing to think
that we can replace ethics with the law
or the law with ethics.
So in the second part of this talk,
I wanna say share about something about what the EU,
what Europe is doing are in terms of legal interventions,
not just ethics, but something that is more binding,
something that can really change company’s behavior.
And that’s the AI Act.
The AI Act is just, it’s a proposal.
It’s not been adopted yet.
So I think that’s one thing to kind of keep in mind.
But we are getting there.
Many people hope that we can adopt it by before Christmas,
but I’m not sure that that’s gonna be possible.
But what is the background of this?
Why do we need to regulate AI?
For many reasons in general,
the main point is that
AI poses an unprecedented threat to our society
and to our key core values.
And probably this is particularly well expressed
through the examples related to algorithmic bias.
We know that in many legal systems such as in the US,
algorithms are used to predict whether
or not people are gonna become criminals in the future.
And many of you won’t be surprised to learn that
these systems are biased against black people.
So black people are more likely to become criminal
according to these types of algorithm.
So we clearly need better regulation
because we cannot allow our society
to become so unjust,
to become even more unjust than it currently is.
So the EU has decided to do something about it.
It’s the first organization, it’s not a country,
it’s not an international organization,
it’s a weird kettle of fish.
But the EU has decided to be the first in the world
to introduce a binding,
overarching comprehensive law around AI.
It applies to all industries,
including fashion and photography.
It’s even broader in its application than the
previous regulation that the general GDPR.
It introduces or it will introduce harmonized rules
for development placement on the market
and use of AI systems.
It introduces sanctions that are even higher
than the sanctions that we have
with the GDPR up to 30 million, euros,
or 6% of the global turnover of the company.
It’s a regulation that is similar to product safety.
I think all of you are familiar
with the model of product safety.
So when you build, when you make for example,
a product like, I dunno a hairdresser,
you’re gonna test it before putting it on the market
to make sure that it’s not gonna electrocute
the user of the hairdryer.
That’s the model of product safety.
And with the AI Act, we’re gonna see similar things.
That’s a sort of CE marking
for that is gonna say this AI system is actually safe
and can be put on the market.
That’s really good.
However, there are a number of issues
because the AI Act is focused on forms of narrow AI.
So what is saying, what the AI Act says
is that certain specific application
of artificial intelligence, are high risk
pause a high risk to fundamental rights, okay?
So the type of AI that the AI Act has in mind
is that type of narrow AI,
for example, we can think about Alexa
or Siri as examples of narrow AI.
And finally, another thing to keep
to know about the proposed AI Act is that
it will be enforced by
what the local’s market surveillance authorities
that’s in Italy,
that’s likely to be autorita’ antitrust.
However, I’ve carried out in the past research
about kind of fashion and power.
And what I found was, one of the things I found was
that the autorita’ antitrust in Italy
has done very little to address the power imbalance
in sectors such as fashion and photography.
Anyway, so this is a kind of,
some key elements of the AI Act
that I wanna share with you.
If you’ve heard about the AI Act before,
probably the main thing you’ve heard about it
is that it takes a risk-based approach.
A risk-based approach means that certain types of AI,
certain applications are regarded as posing
an acceptable risk to society.
For example, forms of AI used
for subliminal manipulation, they are unacceptable
and therefore they are prohibited.
They cannot be deployed,
they cannot be put on the market,
they cannot be used in Europe.
So it’s a very strong type of regulation, it’s a ban.
But that’s a very short,
the scope of this is very short.
There is not a lot of types,
there are not a lot of types of AI that are prohibited.
The core of the AI Act is higher risk systems.
High risk systems means systems that pose
a higher risk to health,
safety and fundamental rights.
For example, AI systems used to decide
whether or not your child can go,
can get into a certain university or not.
These are high risk.
That means that these systems are legal,
they are allowed,
but they are significantly regulated,
which means you can put them on the market,
you can use them,
but they have to be subject to certain
safeguards and standards.
For example, manufacturers of these systems
have to mark the AI system as safe
with the classical like CE marking
that you see on your hairdryer, for example.
And there are, for example,
some specific design requirements
and some duties that we don’t have
the time to go too much into them,
but let’s just keep in mind
that there is a risk-based approach.
Depending on the risk,
there are more or less restrictions
around different types of AI systems.
However, there is a big,
big problem with the AI Act.
The problem is the AI Act
was written before ChatGPT went viral.
Okay, ChatGPT and all the other
forms of generative AI.
And why is that a problem?
It’s a problem because the AI Act, as I said,
was focused on the idea of narrow AI.
It was focused on the idea
that certain specific applications
and certain specific domains of AI
are pose a risk, a high risk,
for example, to our fundamental rights.
Whereas ChatGPT
and similar forms of generative AI
are not so much forms of narrow AI
are forms of general AI,
because they can be used for a range of different things.
Some of these things will be very low risk.
Some of these things will be very high risk.
For example, you can ask ChatGPT
to write a birthday card to a friend of yours,
and that’s definitely no risk.
But you can also ask ChatGPT to write some codes
for a lethal autonomous weapon.
And that’s, I would say probably
even more than high risk.
The high risk, it should not be permitted.
So what the other problem is that the types of AI
that the AI Act considers are mostly types of predictive AI,
whereas generative AI is about generation of new content.
So it’s quite, it’s quite different.
So why does it matter?
Why should we also regulate
not just the types of AI that we had before,
but also these new types on AI?
And why should you care people in that room in Milan?
Well, because I assume that many of you
are concerned and if not should be concerned
and may have read headlines
such as the one that you see on these slides
from The Guardian biggest act
of copyright theft in history.
Effectively, what happened was that
some Australian writers found out that
thousands, thousands of their books
had been used to train an AI model.
And not only they had not been asked for permission,
they were not even sort of consulted,
and they were not even compensated for this.
That’s a proper theft, that’s a huge, huge problem.
And imagine if you’re a photographer,
if you just found out that
all of your photographs have been used to train an AI model
and they didn’t ask for permission
and they didn’t compensate you.
So that’s a huge problem.
And that applies across the creative industries.
So it’s clear that there is a need for regulation,
but how do we do this?
Because we said that the AI Act
really was crafted and drafted before ChatGPT.
So what’s going on at the moment
is that the European parliament and the council
are trying desperately to change
and change this, the AI Act,
so that it accounts for these types of new generative AI.
But it is a mess,
because obviously that’s not the purpose
for which this law was written.
So I’m not gonna go too much into details
because I’ve seen that I’ve finished my time.
But effectively the problem is that they’re trying to,
to shoehorn,
this law to adapt to a new type of AI
that really requires quite a different type of legislation,
which to me means that we should stop, we should pause
and we should probably rewrite this AI Act from scratch.
But in law, one of the big problems in law
is the politics behind it.
So the EU has an interest in being the first
sort of country in the world
to have binding AI regulations,
and therefore they’re trying desperately to
find a consensus.
But at the moment the main problem,
the main reason why we cannot find a consensus
is exactly the problem around generative AI
and foundation models, how to regulate them,
it’s really quite unclear.
I can see that I have finished my time.
So I’m just gonna briefly mention this,
that the question of whether,
the law can really help us
to protect humanity in a world of AI.
The answer is probably no.
In particular in the fashion industry,
but also beyond that.
I mentioned that, in previous,
in the past conducted some research around fashion.
And what I found was that
the law really cannot fix the power imbalance of fashion
because in the fashion industry
there are unwritten rules, social norms
that prevail even on the law.
So there is that problem in the fashion industry.
But more generally, the law really struggles
with new technologies.
And in my book,
in terms of things and the law,
I argue that the law will always fail
to thoroughly regulate new technologies.
And to do that, I take a Marxist approach
that I’m not gonna go too much into it now,
but what I will say is that,
even though the law tends to fail when it comes to,
the law itself is not sufficient
in regulating new technologies,
it’s an important part of the puzzle,
but it’s not sufficient.
What we need alongside the law
is really conscious action.
We really need to organize all the people
that are potentially affected by AI.
They really need to organize
and fight against these threats.
So to conclude,
and I promise this is gonna be 20 seconds,
there is an abundance of ethical guidelines,
manifestos, et cetera, in the field of AI.
They’re not all bad.
They are conversation starter,
they can help raise public awareness,
but cannot, they cannot be a substitute
for actual regulation of AI.
The EU is the most advanced place in the world
when it comes to AI regulation,
despite all the problems that we’ve seen.
But we need to slow down
and we need to rewrite the rules
to account for types of AI
like generative AI like ChatGPT.
The law I do think will always struggle
in the regulation of new technologies.
Well, fundamentally,
because it’s the law itself is a capitalistic product.
And the law struggles in particular
in the fashion industry and in the tech sectors.
The law, however is part of the puzzle.
It’s part of a strategy to tackle the issues
that we have in this field.
What I really would like to see, as I said,
is collective action.
I would like people like photographers
whose work are being used to train AI,
would talk about these problems
and organize to make their voices heard.
For example, do they want forms
of fair remuneration for the use of their works?
And I think things like this symposium
can be definitely the start of that conversation.
Thank you.
[audience applauding]