I want to tell you a story.

Once upon a time,
they used to build nests,

and a small group of sparrows

were standing before
the sunset's fading light

remembering how heavy their day was.

One of the sparrows said,
We're all so small and weak.

It would be so nice, if an owl
could help us build our nests.

Indeed, said another sparrow.

It could also help us raise our children
and take care of our old folks.

A third sparrow joined in and said,

This would surely be

the best thing that could ever happen.

Let's send out all our scouts
to look for a baby owl.

Let's bring him here as soon as possible,
and progress with him.

With his keen eyesight,

he could also help us see and understand
when that pesky neighbor's cat is coming.

He could help where none of us can.

Only one small sparrow, Scrunfinkle,

with just one eye and a grumpy character,

butted in and said,

This could be the end of us.

Bringing in among us
a creature we don't know,

without knowing how to tame it,

will reduce us all to slavery,
wipe us out from the Earth.

Skeptical as they were,
the other sparrows said,

Absolutely not, an owl will help us!

Let's first bring him here.

Learning to train an owl, a new creature,
is difficult, and takes up a lot of time.

We need help now,
we want to make progress now.

We'll worry about the consequences later.

Scrunfinkle and another
small group of sparrows

began trying frantically
to figure out how to train owls,

fearing that the other scout sparrows

would return from their hunt
before a solution had been found.

I can't tell you
how that story turned out.

What I can tell you is,

today we should all be
that grumpy one-eyed sparrow.

We are creating
super artificial intelligences

and allowing it to learnm to evolve,
through access to Internet.

But nothing ensures

that artificial intelligence

will never evolve strategies

to secure their future dominance over us.

I madly love technology
I try out everything I find before me.

Technology always wins;
the future always wins;

trying to stop it is pointless.

A few years ago -
let's say several years ago-

I had the chance to try out,
for several months,

Google's enhanced reality glasses.

I was so eager to try them out that,

as soon as I got home,

I pulled them out of the box, put them on

and turned them on without reading
the very stuff I write myself:

T&C, and privacy information

that rule the relationship
between me and a technology.

My wife Alessandra, after a few minutes,

got a phone call from an alarmed friend

who asked her, are you aware

Marco is now streaming
in real time all over the world?

This it what happens

when we don't understand
the technology we have in front of us,

when we don't read,
when we aren't informed.

This is what happens
when we don't want to train ourselves,

build awareness.

This is what happens
when we're 5.0 ignorant,

when we have at our disposal
all the knowledge in this world,

but we don't want to get into it in depth.

Today the problem is very simple:

technology is going faster
than anything else,

faster than us, faster than the law,

and faster than our means
to learn and acquire knowledge.

So, nowadays we have
all the world's information at hand,

but we're unable to control it,

to eliminate the so-called
"background noise".

So, when we can't understand something,
we say we need a new law.

We say we need a paradigm shift.

But do we really need new laws?

In 1955, when Luigi Einaudi wrote
his "Prediche inutili" [Useless Sermons],

he wrote these words:

"Knowledge comes first, then discussion,
and finally deliberation.

Legislation is not made through
illusory and sterile ostentation."

Laws made in haste, without
knowing the subject matter,

lead to new laws

which try through workarounds
to do something impossible:

apply rules to the digital world
that were designed for the analogical one.

This is impossible.

This creates a mix
of totally inapplicable norms

that distort the market.

Today we have overcome
our grandparents' ignorance,

the kind of ignorance

arising from lack of information.

This is mostly due to the Internet.

However, a new type of ignorance
has emerged, the processing one,

a type of ignorance resulting
from an overabundance of information,

as well as from our limited desire
and time to process it.

There are concrete examples
in the newspapers every day:

the rampant increase of fake news
and lack of fact-checking

which lead to uncontrolled false alarms

and hate speeches.

Everything that turns up on Internet,

if not verified and checked,

leads to a superficial use of technology.

While a piece of news read incorrectly

can have consequences,

it can also lead, as we shall see,
to even worse consequences.

Nowadays, as I told you,
we are ignorant 5.0,

because both the information
ignorance of the past

and current processing one

lead to the same result:

sub-optimal actions, which are all alike.

Very often we behave like sheep;

we follow the mainstream,
do whatever everyone else does.

We do some things we shouldn't do.

Why?

Because sub-optimal actions
lead to a lack of knowledge

and a standardization
of our behaviour on the web,

certainly not to understanding
how technology works.

Understanding how blockchains work

how innovative technologies,
in general, work,

is committing and time-consuming.

We wouldn't even be able to get it,
more often than not,

because we lack a basic training.

And here is where ethics come into play.

Ethics is the glue
that binds peoples together,

the key that allows them to cooperate,

and something that has helped
our species to progress.

If we are who we are today,
for better or for worse,

we owe it to our ability
of moral reasoning,

to our capacity and desire
to approach situations

with an ethics, and a willingness
to stand out from the crowd.

This is us today.

Artificial intelligence is already here.

It's in our autonomous driving robots;

it's in the little robots
that clean our floors

and track every path
they took in our homes.

It's in hospitals, with surgical robots;

it's with the robots
that help care for the elderly;

it's with those little robots

that keep your child entertained

while you have something else to do
and can't look after them.

What I want you to understand

is that this type of intelligence
is among us today;

Since yesterday, in fact.

And getting into ethics today
is complex and difficult,

and often one doesn't want to do it.

Because AI manufacturers,
most of the time,

are not charities,

but profit companies

who legitimately want to make their money.

So, how can we mandate an ethical code

for whoever develops, programs
or designs an application,

an artificial intelligence
or a software system?

We'll see that later.

What is certainly true today

is that we are like children
playing with a bomb.

We don't realize

that robots should have one main scope,

which is assist us
in improving our well-being,

not promoting the evolution
of technology as an end in itself.

In 1947, when Asimov wrote
"The Three Laws of Robotics",

he said three main things:

Dear robot, you must not kill humans,

obey the orders given by humans beings

and protect your own existence.

Asimov's Three Laws are still today

the foundation for anyone
who deals with ethics of AI.

That's because -

Imagine this story:

a robot tells a woman -
its friend, a human-

that the man of her life loves her.

Actually, this isn't true, but it says so

because otherwise, it thinks,
she'd go mad, she would suffer,

and therefore the first law
would be violated:

do not do harm to humans,
don't make them suffer.

Too bad, by telling a lie,
it violates their trust relationship

and as a result she suffers anyway.

The robot, whose reasoning has always
been static and cold, now goes crazy

because it can't get out
of a moral dilemma:

"Shall I tell her or not?
And how do I tell her?"

Technology is neutral when it is created;

we are the ones to decide
how to apply it in the real world.

So how do you teach ethics to a robot?

How do you teach ethics

to cold binary artificial intelligence?

Do you just need
to code it into its brain?

Imagine a case

where you are in the backseat
of your self-driving car,

and you programmed it
never to go over the speed limit.

Too bad that, on that day you're
in the backseat of the car,

bleeding to death,

and you need to reach the hospital
as soon as possible.

But with its cold reasoning,
the car answers you,

"I can't speed up;
you've coded me not to."

We must pay attention

to how we teach things
to artificial intelligence.

Obviously, we are often
not the ones teaching to them.

But often we are, because
with machine learning technology

humans provide data

to a machine who thinks

in algoritmic terms.

There's a really funny anectode

on a piece of artificial intelligence
most of us keep in the house,

like Alexa or Google Home,
just to name a couple.

During a dinner, one of them
turned on and said,

"Remember to buy cocaine tomorrow"!

It wasn't true;

he didn't have to buy cocaine;
he wasn't a drug user.

But the night before,

on a tv show,

a line of the script was,

"Let's meet tomorrow to get cocaine".

You understand what impact it can have?

This is a funny example,

but there are a lot of examples
that can cause much worse damage.

Imagine a case where a self-driving car
must face a dilemma -

none of us would want to face,
much less leave it to a car.

In order to save
five pedestrians in the street,

should a car think about steering sharply

thus hitting and killing

an unwitting pedestrian on the sidewalk?

When must it do this?

And if there are three people,
or two, in the middle of the street?

How does it calculate this?

Should it take into account
their average age, or life expectancy,

and the expected loss for the State?

Its own financial loss, maybe?

You understand, we couldn't solve
this dilemma ourselves,

let alone teaching a robot what to do.

Clearly, in order to teach it what to do,

sooner or later we will have to define
what a robot can do.

Aristotle used to say,

"To learn how to be good people,
we must get used to doing good things."

Well, maybe this could be the solution

for teaching a robot something true,

what to do, how to react
in such a situation.

But before beginning to think

that artificial intelligence reasons
according to our desires,

we have to step back a bit

and understand how to use ethics

while programming or developing a system.

These self-driving cars are already
driving around many States in America.

They're being tested.

We're trying to understand -

In the newspapers
there's the classic story,

"Driverless car hits
and kills pedestrian".

He would've been hit
even with a human driven car,

because it was dark and he
wasn't over the crosswalk.

There's always an attempt
to limit technology, hinder its progress.

We've got to a point

where even the European Commission
was led to consider

the ethical aspects
of artificial intelligence;

just a few weeks ago
the first guidelines came out:

the first non-binding recommendations

for anyone developing or programming
artificial intelligences,

which are based on the concept
of "trustability"

and the anthropocentric notion
that humans must be at the center.

Technology must not evolve for itself,

but it must evolve to improve
humans' well-being,

such as in the sparrows' plan
at the beginning.

These rules are based on basic principles,

many of which are taken
from the laws of Asimov:

Dear artificial intelligence,
you must not kill humans.

Dear artificial intelligence,

you must not do them harm,
you must follow their orders

and protect yourself.

Others have been added:

you must guarantee equal treatment
to every individual.

And here a small digression
comes into play:

over the past few days
there has been a great scandal

because an application
used all over the world

was used in Arab countries

to monitor a specific type
of individual, women.

Women were being tracked:

where were they going, how long
they stayed, what were they doing.

As I told you before,
technology is neutral.

Ethics, on the other hand,
is in constant evolution,

are forever being debated,
and are culturally specific.

There's no single common notion of ethics.

And this is a problem,
as trying to understand

and explain a developer, a programmer

or a user interface technicial

how to develop, introduce ethics
into his work is already complicated.

Imagine where a European code of ethics -
not even global, just european -

must deal with limiting

scientific and technological progress.

Then there's another problem.

our anthropomorphism
compels us to think about vulnerability.

If we were to think that this robot,
this artificial intelligence,

was similar or identical as us,

we would be making a mistake.

It's not the case now,
and it won't be for many years to come.

Some people say
that artificial intelligence

will surpass our intelligence
in a few decades.

Others are skeptical, instead:

AIs, they say, will always be
bound and narrow.

I'm not telling you which choice is right;
I just ask you to think about it,

Let's start considering,
if we give these machines

too much credibility,
belief and moral thought

without them really having it,

we make a mistake.

Imagine an especially
attractive woman robot

who tells her human companion,

"If you want to keep dating me,

you have to buy me flowers
from this online shop,

and you have to buy me clothes
from that online shop.

You have to take me on certain trips,

otherwise our friendship is over."

Too bad that robot
isn't making human choice,

where two people can decide
what is good for both of them.

Behind those algorithms
and that intelligence,

there are for profit companies.

They could restrict and profile people,

and persuade us in ways
that are totally unthinkable now.

But there is a hope:

these machines do and will need us

as much as we will need them.

This notion of trustability,
this notion of trusting in machines,

must succeed, because if it doesn't,

we won't implement AIs into our lives;

we will never trust it,
always looking for alternative solutions

or we will block technology saying,
"No, I don't want this thing" -

in a superficial way, maybe
without knowing anything about it -

"because it's not for me".

It's ok for me

to manually turn on the light,

instead of giving a voice command;

it's no trouble driving a car
rather than using a self-driving one.

Too bad that progress will go forward,

and growing niches of people
will probably stay behind.

The problem of trust will be

that all this trust placed in machines,
on the other hand, must find a limit.

And the limit could be an ethical one:

the guidelines I have just listed,

or maybe the programmers,

or the people who will
thoughtfully use this intelligences

knowing them, studying them,

much like I tried to do with my son
since he was a few months old,

letting him see technology,

letting him try it
and putting it in his hands.

A friend was telling me, the other day,
robots are like 747 airplanes.

That can cross the planet in a few hours,

but they'll never be able
to land on a tree.

Think again, before completely trusting
our machines, our technology

in your plans, your ideas
and your way of life.

However, if we want the chance
of having a choice,

if we don't want them
to overpower us in the future -

a dystopian future, but in the future -

rule of law won't be worth much.

The ethics of developers, legal designers
and entrepreneurs will be needed,

who will have to think differently

if they don't want to be overwhelmed
by these innovations, too.

In a nutshell, we need
to look beyond the horizon

to avoid ending up like those sparrows.

Thank you.

(Applause)