Technology
Italy's Dual-Track AI: Navigating the EU AI Act, Trust, and the Future of Responsible Innovation
In this episode, we explore Italy's innovative dual-track approach to AI regulation, which combines the EU AI Act with a pioneering national framework. This strategy aims to foster responsible AI...
Italy's Dual-Track AI: Navigating the EU AI Act, Trust, and the Future of Responsible Innovation
Technology •
0:00 / 0:00
Interactive Transcript
spk_0
Welcome back to the DIMTive.
spk_0
Today, we're zoning in on something really significant
spk_0
happening in European tech policy.
spk_0
And maybe surprisingly, it's Italy taking a very visible lead.
spk_0
Usually when we talk AI regulation, all eyes are on the big EU AI
spk_0
act.
spk_0
But Italy, they're showing that just taking the boxes isn't
spk_0
the whole game.
spk_0
So our mission today is to unpack this stack of sources we've
spk_0
got here and really get into Italy's, well, quite
spk_0
sophisticated dual track strategy.
spk_0
It's not just about following rules.
spk_0
It's about blending those pan-European regulations with their
spk_0
own, frankly, pioneering national law.
spk_0
A law aimed at boosting innovation, but crucially, demanding
spk_0
trustworthy and human-centered AI.
spk_0
We want to show you why Italy's shaping up to be a key
spk_0
rights-respecting player in Europe's AI scene.
spk_0
It really is a unique approach, because they haven't just
spk_0
adopted the EU framework.
spk_0
They've built this ambitious legal and institutional
spk_0
structure right on top of it.
spk_0
And you can already see some of that foundation in the
spk_0
government AI readiness index scores.
spk_0
All right, the index.
spk_0
What does that tell us?
spk_0
Well, looking at the 2024 numbers, Italy scores a solid 71.22
spk_0
overall.
spk_0
But what's really interesting, especially for our discussion,
spk_0
is their institutional strength.
spk_0
They score particularly well in governance that's 78.64.
spk_0
And even higher in data and infrastructure, a very strong 81.88.
spk_0
OK, so those scores suggest the bedrock is there, right?
spk_0
The digital plumbing, the legal thinking, the institutions,
spk_0
they're ready to support this big regulatory move.
spk_0
Exactly.
spk_0
It signals they aren't starting from zero.
spk_0
They have the infrastructure to actually back up these legal
spk_0
ambitions we're going to talk about.
spk_0
That's a perfect jumping off point.
spk_0
Let's dive straight into the heart of it.
spk_0
What we're calling the regulatory foundation.
spk_0
Italy's dual track system.
spk_0
What exactly is this dual track strategy?
spk_0
Understanding this dual track idea is, I think, fundamental
spk_0
to grasping the Italian mindset here.
spk_0
Italy is absolutely committed to the EU AI Act.
spk_0
That's regulation 20241689.
spk_0
It's the main cross-sector law for the whole block.
spk_0
No question.
spk_0
But, and this is the key part, they recognize that the EU
spk_0
Act deliberately leaves some wiggle room, some space for
spk_0
national laws in specific areas.
spk_0
Like what kind of areas?
spk_0
Especially things touching on public administration,
spk_0
court-ger justice areas where national competence and
spk_0
specific national contexts are still really important.
spk_0
So dual track means Italy is intentionally adding another
spk_0
layer, a deeper national layer of law on top of the EU one.
spk_0
Tailoring it, basically.
spk_0
Precisely.
spk_0
Tailoring the rules to Italian society, the Italian economy,
spk_0
addressing things like protecting local culture, specific ways
spk_0
AI can be used in sensitive government services, and setting
spk_0
up detailed enforcement structures.
spk_0
And this isn't just a plan, is it?
spk_0
This extra layer, this national law, it's actually happened.
spk_0
It's law number 132225.
spk_0
Tell us about that.
spk_0
Yeah, this is genuinely quite groundbreaking.
spk_0
It could really set a precedent for how other EU countries
spk_0
might eventually flesh out the EU Act.
spk_0
The timeline was pretty fast, which shows the priority.
spk_0
It started as the AI bill, DDL 1146, provisions and
spk_0
delegations to the government concerning artificial intelligence.
spk_0
The Senate approved that back in March 2025.
spk_0
Then the Chamber of Deputies tweaked it and approved their
spk_0
version in June 2025.
spk_0
And finally, it became law number 132225 on September 17, 2025.
spk_0
And it's already in effect?
spk_0
Yep.
spk_0
When to effect pretty quickly, October 10, 2025.
spk_0
And our sources are really clear on this.
spk_0
It's widely seen as the first complete national framework for
spk_0
AI in the EU that fully aligns with the EU Act, but also
spk_0
expands on it strategically.
spk_0
That speed, that scope, it screams major strategic priority.
spk_0
Now, if the EU Act is mostly about risk levels, high risk,
spk_0
low risk AI, what makes this Italian law specifically human
spk_0
centered, it feels like they're trying to bake social values
spk_0
right into the legal code.
spk_0
That's exactly the aim.
spk_0
The guiding principles listed in the log are way beyond just
spk_0
standard legal stuff.
spk_0
Of course, you have the basics, transparency, proportionality,
spk_0
safety, security, accuracy and strong data protection and
spk_0
confidentiality.
spk_0
Naturally, that's baseline GDPR stuff, anyway.
spk_0
The human centered part.
spk_0
That comes through in the explicit inclusion of principles
spk_0
like non-discrimination and promoting gender equality,
spk_0
and interestingly, a commitment to sustainability demanding
spk_0
that AI systems consider their environmental footprint.
spk_0
Sustainability, that's forward thinking.
spk_0
It is.
spk_0
And tying it all together is this core theme of human
spk_0
centrality.
spk_0
OK, human centrality.
spk_0
What does that mean in practice for say a company developing
spk_0
AI in Italy?
spk_0
As it's written into the law, it means AI systems must always
spk_0
allow for human oversight.
spk_0
It guarantees that people have the right to step in to
spk_0
intervene, keep their autonomy, and hold the ultimate
spk_0
decision making power.
spk_0
The whole structure is designed to stop AI from causing
spk_0
harm, or crucially, from overriding human judgment in
spk_0
critical situations.
spk_0
And they're backing this up with real legal consequences.
spk_0
Absolutely.
spk_0
The law brings in criminal sanctions for harmful AI uses.
spk_0
It specifically targets things like malicious deepfakes
spk_0
used for identity theft or spreading disinformation.
spk_0
So actual jail time potentially.
spk_0
Potentially, yes.
spk_0
And it also includes specific safeguards for vulnerable
spk_0
groups, especially children.
spk_0
So it's not just reacting after harm is done.
spk_0
It's trying to prevent it, setting up strong legal
spk_0
deterrence against misuse.
spk_0
OK, imposing criminal sanctions on deepfakes makes
spk_0
sense domestically.
spk_0
But how does that work when the big, generative AI models
spk_0
that might create those deepfakes are often developed
spk_0
and trained outside Italy, outside the EU even?
spk_0
That's the enforced that.
spk_0
That's where the beauty or maybe the cleverness of the
spk_0
dual track system comes in.
spk_0
They rely on the EU AI act to regulate the big foundation
spk_0
models being placed on the market across borders.
spk_0
That's the EU's job.
spk_0
But the national law, law 132 2025, allows Italy to regulate
spk_0
the deployment and specific use cases of AI within Italy,
spk_0
regardless of where the model came from.
spk_0
So they control how it's used locally?
spk_0
Exactly.
spk_0
By creating clear criminal liability for harmful outcomes,
spk_0
like a fraudulent deepfake being used against Italian
spk_0
citizen within Italy, they effectively force anyone
spk_0
wanting to deploy AI there, even global players, to ensure
spk_0
their systems meet Italy's stricter rights-focused standards.
spk_0
The national law becomes a powerful local filter.
spk_0
Right, it sets a very high bar for operating within their
spk_0
borders.
spk_0
That clarity on consequences sets the ethical tone.
spk_0
But rules are only as good as their enforcement.
spk_0
That brings us to institutional oversight and ethical
spk_0
alignment.
spk_0
It's a pioneering law, huge scope.
spk_0
So the big question is, who's actually going to police all
spk_0
this?
spk_0
How does the governance puzzle fit together?
spk_0
Yeah, this is absolutely pivotal.
spk_0
And it's where Italy's big ambitions bump up against some
spk_0
institutional, let's say, complexities.
spk_0
The law proposes two main authorities to oversee things
spk_0
dividing up the jobs.
spk_0
First, there's the agency for digital Italy, AG.
spk_0
The idea is for AGA to handle promoting AI innovation,
spk_0
supporting development, and managing the compliance and
spk_0
notification procedures for AI systems.
spk_0
Think of them as the regulatory facilitator, the admin
spk_0
side.
spk_0
OK, AG are for promotion and paperwork.
spk_0
Who's the enforcer?
spk_0
That's proposed to be the National Cybersecurity Agency
spk_0
ACN.
spk_0
ACN is designated to handle supervision, inspections, and
spk_0
handing out sanctions.
spk_0
But, and this is key, specifically related to the
spk_0
cybersecurity aspects of AI systems.
spk_0
So if an AI system is hackable or poses a security risk?
spk_0
ACN Stepson.
spk_0
If it violates privacy in a non-cyber way, that might be
spk_0
different.
spk_0
They're trying to leverage these existing powerful state
spk_0
agencies by splitting tasks.
spk_0
A.J.D. handles innovation and compliance
spk_0
admin.
spk_0
ACN handles security and enforcement related to security.
spk_0
Makes sense on paper, dividing tasks based on expertise.
spk_0
But you said complexities.
spk_0
What are the risks or tensions in this setup?
spk_0
Are they going to trip over each other?
spk_0
Well, that functional split is exactly what's raised eyebrows,
spk_0
particularly with the European Commission.
spk_0
The core issue boils down to institutional independence.
spk_0
Oh, the EUAI Act requires independent watch stocks, right?
spk_0
Precisely.
spk_0
The National Supervisory Authorities designated under the EU
spk_0
Act need to operate with enough independence to ensure
spk_0
their fair, impartial, and not unduly influenced, especially
spk_0
not by the bodies promoting the technology.
spk_0
And the worry is, A.J.D. is both promoter and compliance
spk_0
handler.
spk_0
That's one major concern from the Commission.
spk_0
Yes.
spk_0
Could having A.J.D., whose job is pushing digitalization,
spk_0
also handle compliance, create a conflict of interest?
spk_0
Might they prioritize getting AI out the door over rigorously
spk_0
checking it?
spk_0
The second concern is just ensuring the overall structure
spk_0
meets those EU independence requirements clearly,
spk_0
making sure these bodies aren't seen as beholden to political
spk_0
whims or each other so they can actually impose sanctions
spk_0
without fear or favor.
spk_0
Italy's still working through the final details on how to guarantee
spk_0
that independence to satisfy Brussels.
spk_0
It's critical for the laws credibility.
spk_0
So it sounds like they may be prioritized getting the law
spk_0
passed quickly, and now they're refining the governance
spk_0
structure to align fully with EU expectations.
spk_0
That enforcement part needs to be demonstrably neutral.
spk_0
That seems to be the situation, yes.
spk_0
But the governance picture in Italy for AI is actually much
spk_0
richer than just these two proposed bodies.
spk_0
You can't talk digital oversight there without talking about
spk_0
the data protection authority.
spk_0
The DPA, the guaranteed.
spk_0
They've been pretty active, haven't they?
spk_0
Extremely active and influential.
spk_0
Long before the EU AI Act was even finalized,
spk_0
the Italian DPA has been a major player.
spk_0
How far back does their AI focus go?
spk_0
They set up a dedicated AI unit way back in 2021.
spk_0
They clearly saw this coming and knew they needed specialized
spk_0
expertise and they haven't just been watching.
spk_0
The DPA has been issuing guidelines on really thorny AI issues,
spk_0
facial recognition use, how health data gets processed by algorithms,
spk_0
algorithms managing gig economy workers, even AI used for
spk_0
tax risk analysis.
spk_0
And crucially, they've taken enforcement actions against companies
spk_0
for misusing data in AI systems.
spk_0
Can you give an example like the gig economy stuff?
spk_0
Sure. Their guidelines there focused on making sure
spk_0
algorithms used to manage delivery writers or other gig workers
spk_0
didn't unfairly profile them or violate their privacy and dignity.
spk_0
They actually forced companies to be more transparent
spk_0
about how those algorithms worked.
spk_0
So the DPA has a proven track record of being assertive and proactive.
spk_0
That's why everyone expects them to play a central coordinating role,
spk_0
working alongside the competition and communications authorities
spk_0
to weave this whole tapestry of oversight together.
spk_0
Okay, so it's a layered system.
spk_0
The new national law, the EU Act, the proposed roles for HID and ACN,
spk_0
but with the experienced DPA likely acting as a central coordinator.
spk_0
And this isn't just built on new laws, is it?
spk_0
The sources mentioned a court ruling from 2019
spk_0
about algorithmic transparency.
spk_0
Yes, that 2019 administrative court judgment was really important.
spk_0
It basically established a legal precedent that public bodies
spk_0
using algorithms for decisions have to be transparent about how they work.
spk_0
They have to show their workings.
spk_0
Essentially, yes.
spk_0
The ruling mandated that public administration has to reveal the logic
spk_0
the criteria used by the software,
spk_0
so citizens actually have a basis to challenge an algorithmic decision
spk_0
they think is unfair.
spk_0
This is fundamental.
spk_0
It means Italy's own courts had already cemented the principle
spk_0
of algorithmic accountability in the public sector before this new AI law.
spk_0
So the new law is building on a pre-solid legal foundation
spk_0
shaped by GDPR, Italian privacy code, competition law,
spk_0
and this judicial precedent.
spk_0
And this complex web of oversight DPA courts, new laws,
spk_0
seems designed to meet public demand, right?
spk_0
You mentioned that Euro-Berometer data.
spk_0
Exactly.
spk_0
That 2025 data showing 87% of Italians believe public institutions
spk_0
should lead AI development in a rights-respecting way.
spk_0
That's a huge mandate.
spk_0
The government is responding to that.
spk_0
Because if the public trusts that there's strong,
spk_0
independent oversight, if they know the DPA is watching,
spk_0
the courts offer recourse, and there are real penalties for misuse,
spk_0
it builds the confidence needed for people and businesses
spk_0
to actually embrace the technology.
spk_0
Trust really is the essential lubricant for innovation here.
spk_0
It makes total sense.
spk_0
You need the guardrails to encourage people onto the highway.
spk_0
Okay, let's shift gears then.
spk_0
We've talked regulation.
spk_0
Now let's talk fostering innovation.
spk_0
Research, compute, and SMEs.
spk_0
Governments can't just regulate.
spk_0
They need to actively cultivate the garden.
spk_0
What's Italy doing to grow its own AI expertise in businesses?
spk_0
They have a very clear roadmap here,
spk_0
going back to the 2022-2024 National Strategic Program on AI.
spk_0
And the absolute flagship initiative driving this is fair.
spk_0
The future artificial intelligence research partnership.
spk_0
This isn't just another funding pot.
spk_0
It's a huge coordinated national push.
spk_0
What makes fair so different or ambitious?
spk_0
Well, its main goal is to stop the fragmentation you often see in research.
spk_0
It brings together 27 top research institutions
spk_0
and over 90 companies all under one strategic umbrella.
spk_0
And they've structured it cleverly.
spk_0
It's organized around 10 thematic spokes.
spk_0
Each spoke focuses on a specific area of frontier AI research.
spk_0
And importantly, these are explicitly linked back
spk_0
to those human-centered principles we discussed.
spk_0
Can you give us examples of these books
spk_0
what kind of research are we talking about?
spk_0
Sure. There's bio-sociacognitive AI
spk_0
looking at how AI interacts with human biology and thinking.
spk_0
There's edge, excess scale AI focusing on massive computing power,
spk_0
both centralized and decentralized.
spk_0
And linking back to the legal principles again,
spk_0
there's green-aware AI.
spk_0
Green-aware AI.
spk_0
That connects to the sustainability principle in the law.
spk_0
What does that mean in research terms?
spk_0
It means actively designing and engineering AI models and systems
spk_0
to minimize their energy use and carbon footprint right from the start.
spk_0
Think energy-efficient algorithms,
spk_0
sustainable data center practices for AI.
spk_0
Given how much energy training big AI models consumes,
spk_0
dedicating a whole research spoke to this shows Italy's
spk_0
serious about making AI development environmentally responsible,
spk_0
not just economically powerful.
spk_0
Fair aims for Italian AI to be, well,
spk_0
symbiotic with society and the planet.
spk_0
That kind of cutting edge research needs serious computing muscle.
spk_0
And that brings us back to Italy's high score in data and infrastructure.
spk_0
They're really pushing their high performance computing or HBC capabilities, aren't they?
spk_0
Absolutely are. This is a key strategic asset for them.
spk_0
Italy hosts the Leonardo Supercomputer at Seneca.
spk_0
It's a pre-exascale Euro-HPC machine consistently ranked among the world's most powerful.
spk_0
So they have the engine?
spk_0
They have the engine, yes.
spk_0
This provides the raw power needed for training the huge AI models
spk_0
and running the complex simulations that underpin the research coming out of initiatives like
spk_0
fair.
spk_0
But the really smart move isn't just having the Supercomputer,
spk_0
it's making sure that power gets used beyond just the big research labs.
spk_0
They want to democratize access, especially for industry.
spk_0
How are they doing that?
spk_0
Through two main initiatives aimed at businesses,
spk_0
there's the IT4-LIA AI Factory and more recently,
spk_0
the AI4 Industry Foundation, which launched in Turin just in May, 2024.
spk_0
And the AI4 Industry Foundation, what's its specific role, especially for smaller companies?
spk_0
It's designed precisely to tackle that infrastructure gap we talked about.
spk_0
It offers access to on-premise HBC infrastructure,
spk_0
but tailored specifically for startups, key industries, and crucially for small and medium-sized
spk_0
enterprises SMEs.
spk_0
Because SMEs can't afford their own Supercomputers.
spk_0
Exactly. Or even the high-end servers needed for serious AI work.
spk_0
This foundation gives them access to advanced computing services without that crippling upfront cost.
spk_0
It's meant to be a bridge, helping research flow from labs like
spk_0
fair into actual commercial products and services developed by SMEs.
spk_0
That leads nicely into funding.
spk_0
You need research, you need computers, but you also need cash.
spk_0
The AI Bill seems to be directing serious money into this ecosystem.
spk_0
Yes, a very clear signal of intent here.
spk_0
The AI Bill authorizes investments totaling up to 1 billion.
spk_0
And this isn't just general funding.
spk_0
It's specifically earmarked for equity and quasi-equity investments.
spk_0
Targeted where?
spk_0
Targeted squarely at SMEs and companies identified as having high growth potential in AI.
spk_0
The stated goal is very explicit.
spk_0
They want to nurture and grow national technological champions in AI.
spk_0
And as one expert, Alessio LaDotsi, highlighted in our sources,
spk_0
this isn't just passive investment.
spk_0
It's meant to be a powerful catalyst.
spk_0
But he also made a key point.
spk_0
Its success is really tied to how well and how quickly Italy implements both the EU AI Act
spk_0
and its own national law.
spk_0
Because strong, clear rights-respecting rules build trust.
spk_0
And that trust is what ultimately attracts the private investment
spk_0
and international partners needed to scale up these national champions.
spk_0
Regulation and investment have to go hand in hand.
spk_0
A billion euros is certainly a catalyst.
spk_0
And they're trying to build the workforce to use it too, right?
spk_0
Tackling the talent pipeline.
spk_0
Absolutely essential.
spk_0
You can have all the money and computers in the world, but you need the people.
spk_0
Especially given Europe's historical challenge with brain drain talented people leaving.
spk_0
So they're putting a lot of effort into expanding the PhD-AI.it consortium.
spk_0
This is already one of Europe's biggest AI doctoral networks.
spk_0
One of the goals there, just more PhDs.
spk_0
It's more strategic than just numbers.
spk_0
They want to strengthen the national PhD program, yes,
spk_0
but also make STEM careers, particularly in AI, more visible and attractive, especially to women.
spk_0
And they're actively trying to attract global talent with scholarships and good research
spk_0
environments and crucially retain the talent they train.
spk_0
It's about building that critical massive expertise needed to make all this work.
spk_0
Okay, so we have this picture.
spk_0
Massive investment, world-class computing, a big push on research and talent.
spk_0
Yet the sources consistently point out a major bottleneck,
spk_0
actual AI uptake among SMEs.
spk_0
The small, immediate businesses that are the backbone of Italy's economy
spk_0
is still quite low.
spk_0
Why the disconnect?
spk_0
If the tools are there, why aren't they using them?
spk_0
Ah, this is the classic challenge, isn't it?
spk_0
Where top-down strategy meets bottom-up reality.
spk_0
The surveys pinpoint a few core interconnected reasons,
spk_0
particularly for the micro and small firms that dominate the Italian landscape.
spk_0
First, just the practical technical hurdles.
spk_0
Many SMEs simply don't have the internal digital infrastructure or the skilled IT staff
spk_0
needed to effectively integrate or manage AI tools.
spk_0
Second, and maybe even harder to shift, is what you could call limited business expertise
spk_0
or digital culture.
spk_0
You have many traditional, often family-run businesses where the leadership might be resistant to change,
spk_0
or just lack the understanding of how AI could actually benefit their specific business.
spk_0
So a skills gap and maybe a mindset gap?
spk_0
Exactly.
spk_0
And third, there's the straightforward financial barrier.
spk_0
Even with support like the AI for industry foundation making compute cheaper,
spk_0
the upfront cost for consulting, software integration,
spk_0
potentially retraining staff that can still be prohibitive for companies with type margins,
spk_0
the investment risk feels very high to them.
spk_0
So it's not just about access to the tech itself.
spk_0
It's the whole package.
spk_0
Yeah.
spk_0
Integration costs, lack of internal skills,
spk_0
maybe even a lack of awareness or belief in the benefits.
spk_0
Precisely.
spk_0
It's a deep structural and cultural issue.
spk_0
The infrastructure might be getting built,
spk_0
but bringing that last mile to the smaller firms requires more than just access.
spk_0
How can the strategy address that?
spk_0
That's where things like regulatory sandboxes come in,
spk_0
which the AI Bill specifically allows for.
spk_0
Yeah.
spk_0
These create safe spaces for SMEs to experiment with AI,
spk_0
try things out without facing the full whack of regulations or huge costs immediately.
spk_0
Like a controlled test environment.
spk_0
Exactly.
spk_0
And also things like cascading funding,
spk_0
called smaller grants,
spk_0
channeled through initiatives like fair,
spk_0
specifically designed for SME projects,
spk_0
to help them take those first steps and de-risk the experimentation.
spk_0
Getting AI diffused widely into the SME sector
spk_0
is probably the single biggest economic challenge
spk_0
for Italy's AI strategy to succeed.
spk_0
Right, that makes sense.
spk_0
Now let's move to perhaps the most fascinating part of this dual track approach.
spk_0
How does all this high-level policy and regulation actually translate into specific uses of AI?
spk_0
And critically, what specific safeguards are being built in?
spk_0
Let's look at the key sectors the AI Bill highlights,
spk_0
starting with justice and public administration.
spk_0
This is where that principle of human centrality becomes really concrete, non-negotiable.
spk_0
The law encourages using AI and public administration
spk_0
for things where it's genuinely useful,
spk_0
organizing services better, simplifying some aspects of judicial casework,
spk_0
speeding up processes to boost efficiency.
spk_0
AI can help manage the sheer volume of data public bodies deal with.
spk_0
However, the law draws a very clear, very bright line.
spk_0
The absolute key safeguard is stated explicitly.
spk_0
Any decision involving interpreting the law,
spk_0
evaluating facts or evidence in a case,
spk_0
or making a final decision that affects the citizen's rights
spk_0
or access to public services.
spk_0
That decision is always reserved for the human judge or the human administrator.
spk_0
So AI can assist, gather info, maybe draft documents.
spk_0
Exactly. It can be a powerful assistant.
spk_0
But it can never be the judge, jury, or final decision maker.
spk_0
What risks does that safeguard really tackle?
spk_0
It directly tackles the risk of automated injustice,
spk_0
where biased algorithms or errors could lead to wrong or unfair outcomes for citizens.
spk_0
It prevents the erosion of due process.
spk_0
By banning full automation in these critical decision points,
spk_0
Italy is hardwiring human accountability into its public sector AI use.
spk_0
It's a profound statement about maintaining democratic control over administrative
spk_0
and judicial power.
spk_0
That clarity is really striking.
spk_0
How does that principle apply in other sensitive areas,
spk_0
like health care and labor?
spk_0
Very similarly.
spk_0
In health care, the aim is positive.
spk_0
Use AI to help improve diagnosis,
spk_0
spot disease patterns earlier,
spk_0
personalized treatment plans.
spk_0
There's also a focus on using AI to develop systems that help people with disabilities.
spk_0
But just like injustice,
spk_0
the ultimate responsibility stays human.
spk_0
Medical decisions, the diagnosis,
spk_0
the treatment plan always remain the responsibility of the human medical professional.
spk_0
AI can provide insights,
spk_0
analyze scans, suggest options, but the doctor decides.
spk_0
And in the workplace, labor.
spk_0
The focus there is on using AI to improve working conditions
spk_0
and boost productivity,
spk_0
but with a very strong emphasis on protecting the worker.
spk_0
The law explicitly says AI must enhance work
spk_0
while protecting human dignity and the psychophysical well-being of employees.
spk_0
Plus, they're planning an observatory specifically to monitor how AI adoption actually impacts
spk_0
the workforce in practice.
spk_0
Looking for things like algorithmic bias and hiring or promotions,
spk_0
new forms of surveillance,
spk_0
or impacts on job quality and worker rights.
spk_0
Italy's cultural heritage is arguably its biggest global calling card,
spk_0
all those UNESCO sites.
spk_0
How does the AI bill handle culture and creativity,
spk_0
especially now with generative AI exploding?
spk_0
We have that vignette from Cecilia Lasselfari.
spk_0
Yes, this is where the national law gets very specific again,
spk_0
and very relevant to people working in creative fields.
spk_0
As Cecilia Lasselfari points out,
spk_0
generative AI is already creeping into the cultural sector.
spk_0
How so?
spk_0
Think things like creating dynamic, interactive museum exhibits,
spk_0
digitally restoring damaged artworks,
spk_0
or using AI to help artists bridge traditional analog techniques
spk_0
with new digital possibilities,
spk_0
maybe for augmented reality experiences.
spk_0
The law says this integration is fine,
spk_0
but it must be done responsibly.
spk_0
That means strictly respecting existing copyright laws,
spk_0
and perhaps most importantly for creators,
spk_0
the law tackles the thorny question of authorship head on.
spk_0
It states clearly that to claim authorship of AI-generated content,
spk_0
there must be a significant human creative contribution.
spk_0
Ah, so just typing a prompt into mid-journey might not cut it,
spk_0
that output wouldn't automatically be authored by the promptor under this law?
spk_0
That seems to be the direct implication, yes.
spk_0
It pushes back against the idea that the AI itself is the author.
spk_0
It reaffirms the value of the human creator's input,
spk_0
the curation, the artistic choices, the refinement,
spk_0
the intellectual effort involved.
spk_0
It also promotes transparency,
spk_0
ensuring that if AI tools were used in creating something,
spk_0
that should be clear.
spk_0
It's really an attempt to protect the value of human creativity
spk_0
and intellectual property in the age of AI.
spk_0
That commitment to human oversight and democratic values
spk_0
seems to run deep.
spk_0
It even extends into the government itself.
spk_0
Let's talk about that really interesting 2024 project
spk_0
in the Italian Chamber of Deputies.
spk_0
This idea of augmented democracy.
spk_0
Using AI not just to regulate others,
spk_0
but to improve how government itself functions.
spk_0
It's a fascinating case study, really.
spk_0
The government basically leading by example,
spk_0
showing they trust the tech enough for their own critical work,
spk_0
but only under very strict conditions.
spk_0
In 2024, the Chamber of Deputies piloted three AI prototypes aimed at making
spk_0
the legislative process more efficient and accountable.
spk_0
What did these systems actually do?
spk_0
Okay, first, they built Norma.
spk_0
Think of it as a virtual research assistant for members of parliament.
spk_0
It helps them quickly find, organize, summarize,
spk_0
and analyze huge amounts of legislative documents,
spk_0
past laws, related debates,
spk_0
making it easier to get up to speed on complex bills.
spk_0
Second, there's MSE.
spk_0
That stands for Makina,
spk_0
Skritura, Amenda Menti,
spk_0
which translates roughly to amendment writing machine.
spk_0
This is more technical.
spk_0
It helps parliamentary staff draft legally,
spk_0
precise, and correctly formatted amendments to legislation,
spk_0
which can be a very thirdly time consuming process.
spk_0
It boosts efficiency without losing accuracy.
spk_0
Okay, internal efficiency tools.
spk_0
What about engaging the public?
spk_0
That's a third one, Deput chat.
spk_0
This is a public facing chatbot,
spk_0
designed to answer citizen questions about what parliament is doing,
spk_0
bills being debated, legislative processes, etc.
spk_0
The idea is to make parliament less opaque
spk_0
and improve democratic engagement by providing information
spk_0
in an easy to use conversational way.
spk_0
Sounds powerful.
spk_0
But here's the million dollar question,
spk_0
especially for systems involved in analyzing
spk_0
and even drafting legal text.
spk_0
AI is known to hallucinate, right?
spk_0
Make things up.
spk_0
How do they stop these parliamentary
spk_0
AI's from generating incorrect information
spk_0
or worse drafting flawed amendments?
spk_0
The risk there seems enormous.
spk_0
This is the really clever part,
spk_0
and it's the ultimate proof of their commitment
spk_0
to security and trustworthiness.
spk_0
The source does make a huge point of this.
spk_0
These parliamentary AI systems operate
spk_0
within a completely sealed off environment,
spk_0
a walled garden as they put it.
spk_0
Meaning the AI models used in Norma, MSE,
spk_0
and Deput chat are only trained on and allowed to access
spk_0
verified official, authenticated legislative data,
spk_0
government documents, parliamentary records,
spk_0
official visits, that's it,
spk_0
no connection to the open internet,
spk_0
no scraping random websites.
spk_0
By strictly limiting the data source
spk_0
to validated official information,
spk_0
they effectively eliminate the primary cause
spk_0
of AI hallucinations,
spk_0
which usually happens when models pull
spk_0
from the vast, messy,
spk_0
often unreliable data of the open web.
spk_0
It's a very security conscious design.
spk_0
Wow.
spk_0
So they get the speed and analytical power of AI,
spk_0
but apply it only to a trusted,
spk_0
controlled data set relevant to the task.
spk_0
Yeah, that's actually brilliant.
spk_0
It sidesteps a massive risk.
spk_0
It really is.
spk_0
It's a model other governments
spk_0
wrestling with AI safety and disinformation risk
spk_0
could learn a lot from.
spk_0
It shows you can use AI in sensitive government functions,
spk_0
but only if you build in extreme safeguards
spk_0
around the data it uses.
spk_0
That's a profound practical example
spk_0
of their ethical strategy.
spk_0
Use the power,
spk_0
but control the inputs rigorously
spk_0
to maintain integrity.
spk_0
Okay, as we start to wrap up this deep dive,
spk_0
let's pull back.
spk_0
We've seen Italy's strengths,
spk_0
the pioneering national law,
spk_0
strong data infrastructure,
spk_0
big research coordination,
spk_0
VFR,
spk_0
a clear commitment to human centric values.
spk_0
But what are the big hurdles
spk_0
that persistent challenges they still need to clear
spk_0
to make this whole vision a reality?
spk_0
There are probably three main areas
spk_0
where continued focus
spk_0
and frankly accelerated executioner needed.
spk_0
First, as we discussed,
spk_0
that crucial issue of institutional clarity.
spk_0
They absolutely have to finalize the roles
spk_0
and responsibilities of the supervisory bodies,
spk_0
AJA, D&ACN,
spk_0
and critically satisfy the European
spk_0
commission's concerns about their independence
spk_0
and resources.
spk_0
Any lingering ambiguity there
spk_0
just slows things down
spk_0
and undermines confidence.
spk_0
Get the government structure finalized
spk_0
and approved.
spk_0
What else?
spk_0
Second, SME adoption.
spk_0
That persistent challenge we talked about.
spk_0
The billion euro fund is great.
spk_0
The infrastructure is improving,
spk_0
but they have to crack the code
spk_0
on getting AI diffused widely
spk_0
through the small and medium business sector.
spk_0
That means getting those regulatory sandboxes
spk_0
up and running fast,
spk_0
creating targeted support programs
spk_0
may be addressing the cultural
spk_0
and skills gaps more directly.
spk_0
Make it easier and less risky
spk_0
for SMEs to actually use the tools.
spk_0
And the third.
spk_0
Third is talent retention.
spk_0
The PhD AI Agate initiative is a good start,
spk_0
but combating brain drain requires
spk_0
continuous long-term effort.
spk_0
They need to make sure Italy is an attractive place
spk_0
not just to train AI experts,
spk_0
but for them to stay and build their careers,
spk_0
whether in academia or industry.
spk_0
You need the skilled people on the ground
spk_0
to make all this technology work effectively and ethically.
spk_0
So institutional clarity, wider SME uptake,
spk_0
and holding onto the talent.
spk_0
Big tasks remaining.
spk_0
So looking ahead,
spk_0
what does Italy's whole approach,
spk_0
this blend of specific national rules,
spk_0
ethical focus, and tech investment
spk_0
mean for the bigger picture of AI governance,
spk_0
maybe globally?
spk_0
Well, what's really striking is how explicitly
spk_0
Italy is framing its strategy around AI for society.
spk_0
It's not just about economic competition,
spk_0
though that's part of it.
spk_0
It's about proactively embedding
spk_0
these human-centric ethical,
spk_0
even green principles right into the legal
spk_0
and research fabric from the outset.
spk_0
And when you connect that to the concrete examples,
spk_0
the absolute safeguard for human decision-making
spk_0
and justice, the clarity on authorship
spk_0
and creative fields,
spk_0
that incredibly secure,
spk_0
walled garden approach for their own parliamentary AI,
spk_0
you see a consistent pattern.
spk_0
It's a deep commitment to protecting democratic processes
spk_0
and individual rights while innovating.
spk_0
Which leads to a final thought?
spk_0
Yeah, it raises a really interesting question
spk_0
for you, the listener, to think about.
spk_0
Could Italy's model focusing not just on being
spk_0
the fastest or biggest in AI,
spk_0
but on carefully balancing innovation
spk_0
with deeply embedded democratic values
spk_0
and human rights?
spk_0
Could that actually become a kind of gold standard?
spk_0
Could it set a unique benchmark
spk_0
for how other countries,
spk_0
especially in Europe,
spk_0
navigate the really complex ethical terrain of the AI age?
spk_0
They're making a strong case
spk_0
that you don't have to choose between innovation and values.
spk_0
A very compelling thought, indeed.
spk_0
The blueprint seems strong.
spk_0
The ethical foundations are clear.
spk_0
Now, as you said,
spk_0
it comes down to execution
spk_0
and making it all work in practice.
spk_0
Regulators worldwide will definitely be watching Italy closely
spk_0
to see if they can translate this
spk_0
impressive readiness and pioneering law
spk_0
into a thriving rights-respecting AI ecosystem.
spk_0
Thank you for walking us through
spk_0
this really complex dual track strategy.
spk_0
It was my pleasure, a fascinating case study.
spk_0
And thank you, listener, for joining us for this deep dive.