00:00Imagine a city where you never have to fill out the same form twice.
00:04where every euro of public money is visible to everyone, in real time.
00:07where artificial intelligence handles the administrative tasks,
00:10but where it is always humans who decide.
00:13No surveillance, no social score, no dystopia.
00:16Welcome to the solar city, an optimistic solarpunk scenario.
00:20but built on bricks that already exist, today, in the real world.
00:24Settle in, we're starting with the beginnings of artificial intelligence and we'll go all the way to 2045.
00:29It all stems from a feeling that many of us are familiar with,
00:32files that need to be sent again and again,
00:35decisions are made without explanation.
00:37and sometimes worse, the impression, or the certainty,
00:40that the rules are not the same for everyone.
00:42Corruption is first and foremost a matter of opacity.
00:45Take Ukraine in 2014.
00:47According to figures reported by the Wall Street Journal,
00:49Approximately 10% of federal public procurement was linked to some form of corruption.
00:53and another 10% was wasted due to a lack of competition.
00:56on a budget of approximately $11 billion.
00:59Remember one simple idea; it will run throughout this entire video.
01:02Corruption never thrives in broad daylight.
01:05To understand how AI could one day help manage a city,
01:08We need to go back to the origins.
01:10In 1943, Warren McCulloch and Walter Pitts proposed a mathematical model of the neuron.
01:15In 1950, the British mathematician Alan Turing posed a mind-boggling question.
01:20Can machines think?
01:22He imagines a test, the imitation game.
01:25Then, in the summer of 1956, at Dartmouth College in the United States,
01:29John McCarthy, Marvin Minsky, Claude Shannon and Nathaniel Rochester
01:32organize a workshop that gives its name to the discipline.
01:36artificial intelligence.
01:37In 1958, Frank Rosenblatt presented the Perceptron,
01:41a machine that learns to recognize shapes.
01:43The optimism is immense!
01:45But the history of AI is made up of spring and winter.
01:48In 1966, at MIT, Joseph Weizenbaum created ELISA,
01:52a program that mimics a psychotherapist by reformulating your sentences.
01:56In the 1970s, at Stanford,
01:58The MySyn expert system helps diagnose infections
02:01thanks to a few hundred "if, then" type rules.
02:05Then, the promises disappoint.
02:07In 1973, in the United Kingdom,
02:09The Light Hill report harshly criticizes the research.
02:12This is the first winter of AI.
02:14In 1987, the market for LISP machines collapsed.
02:17Second winter.
02:18However, one lesson remains valuable for our city.
02:21A handwritten rule is slow to produce.
02:24But it can always be explained.
02:25The real turning point comes with machine learning.
02:28In 1997, IBM's Deep Blue
02:31beats the world chess champion, Gary Kasparov.
02:33In 2012, the AlexNet neural network
02:36crushes an image recognition competition.
02:39Deep learning is taking off.
02:41In 2016, AlphaGo beat Icédol in the game of Go.
02:44In 2017, Google researchers published the Transformer architecture,
02:48the basis of almost all major language models.
02:50At the end of 2022, ChatGPT will be implementing conversational AI
02:53in the hands of the general public.
02:55And since 2024, we've seen the agents arrive.
02:58AI capable of planning,
03:00to use tools,
03:01to chain tasks together.
03:02Exactly the type of tools that could,
03:04Tomorrow, I will be processing an administrative file.
03:06Here now is the technical heart of our city.
03:09The law that becomes code.
03:11Even today,
03:12These are programs that enforce the law.
03:14In France,
03:15income tax is calculated
03:16by approximately 100,000 lines of a specific language,
03:19the M language.
03:20And social benefits rely
03:21across millions of COBOL lines,
03:23a language from the 1960s.
03:25Issue,
03:26when the code and the law diverge,
03:27Who realizes this?
03:29This is where Catala comes in.
03:30a language developed by INRIA
03:32and named in homage to Pierre Catala.
03:34Each paragraph of the law is translated,
03:36line by line,
03:37in verifiable code,
03:38that lawyers can review.
03:40In 2026,
03:41the CNAF,
03:42which pays out approximately 110 billion euros in benefits
03:44to more than 13 million households,
03:46joined the project.
03:47This movement is called
03:48Rules as Code
03:50with also Open Fisca in France
03:51and Better Rules in New Zealand.
03:53And what about the legal safeguards?
03:55They are building themselves.
03:56In 2016,
03:57the law for a digital republic
03:59requires that an administrative decision
04:00taken using an algorithm
04:02mentions it
04:03and that these rules can be communicated
04:05to the person concerned.
04:06In 2018,
04:07the GDPR,
04:08in its article 22,
04:09enshrines the right
04:10not to be the subject
04:11of a fully automated decision
04:13which significantly affects you,
04:15except in specific, regulated circumstances.
04:17In September 2020,
04:18Amsterdam and Helsinki
04:19launch the first public registers
04:21of algorithms in the world.
04:22In 2024,
04:24The European Union adopts the AI Act,
04:26who classifies artificial intelligences
04:27by risk level.
04:29And since February 2, 2025,
04:31the social rating of citizens
04:32is purely and simply prohibited
04:34within the European Union.
04:36This is important.
04:37because when we talk about a city managed by AI,
04:39Many immediately imagine the worst.
04:41A points-based driving license for citizens
04:43a social score,
04:44Cameras everywhere.
04:44So let's be clear,
04:46The solar city rejects the social score.
04:49She refuses the citizen's points-based driving license.
04:51She rejects mass facial recognition.
04:54She rejects black box decisions
04:55and any decision without human recourse.
04:58Here,
04:58a rule judges specific actions
05:01never the value of a person.
05:02And transparency works the other way around.
05:04It is not the citizen who is under surveillance.
05:06It is power that is visible.
05:09And what about France in all of this?
05:10She was a pioneer,
05:11and also vigilant.
05:13From the 1960s onwards,
05:14Pierre Catala,
05:15law professor,
05:16imagine legal informatics
05:18and creates databases of case law.
05:20State Councillor Lucien Mell
05:21He too is considering the automation of law.
05:24But in 1974,
05:26a project is causing a scandal.
05:28Safari,
05:28who wanted to assign a unique identifier
05:30to each Frenchman
05:31to interconnect the administration's files.
05:34The newspaper Le Monde headlines
05:35Safari, or the hunt for the French.
05:37The reaction is strong.
05:39and it culminated in 1978
05:41to the French Data Protection Act
05:42and to the creation of the CNIL.
05:44The lesson is essential.
05:46Connecting data without safeguards is frightening.
05:48Rightly so.
05:50Our city will therefore begin with the safeguards.
05:52Next stop, Estonia.
05:53From 2001 onwards,
05:54This small Baltic country is building X-Road.
05:56a secure data exchange layer.
05:58The idea is brilliant in its simplicity.
06:01There is no large central file.
06:02Each administration keeps its own data
06:05and they communicate with each other using encrypted methods.
06:07only when necessary and permitted.
06:09With a principle enshrined in law,
06:11the "only once".
06:12The state does not have the right to ask you again
06:14information he already possesses.
06:16And every citizen can see in access logs
06:18who accessed this data.
06:20Today,
06:21X-Road,
06:21That's more than 3000 online services
06:23and approximately 2.2 billion transactions per year.
06:26The software is even open source
06:28and used in Finland
06:29in Iceland or Japan.
06:31NOW,
06:31Ukraine and the fight against corruption.
06:33Before 2015,
06:35Public procurement was done on paper.
06:37Cooked, Prozoro is born.
06:38a name that evokes transparency in Ukrainian.
06:40The system is co-constructed by the State,
06:42Transparency International Ukraine,
06:44and digital companies.
06:46A 2015 law makes its use mandatory
06:48for public buyers.
06:50Its slogan?
06:51Everyone can see everything.
06:52Each call for tenders,
06:53each auction,
06:54each contract is public
06:55and civil society is monitoring
06:57thanks to Ozoro's portal.
06:58Result,
06:59in 9 years,
07:00approximately 301 billion rivnias in savings,
07:03including more than 51 billion in 2024 alone.
07:05It's not perfect,
07:07But it's proof.
07:08Digital transparency reduces corruption.
07:10Third pioneer,
07:12Taiwan.
07:12In 2014,
07:13The sunflower movement is shaking the country.
07:16A community of civic hackers,
07:17Gov,
07:18proposes to improve democracy
07:20rather than going around it.
07:21In 2015,
07:22V-Taiwan is born
07:23which uses a tool called
07:25Paul.Ease.
07:25The principle?
07:27Everyone reacts to proposals.
07:29The algorithm groups the opinions,
07:30and what comes up first,
07:32These are not the most aggressive messages.
07:34but the proposals that convince
07:36opposing groups.
07:37This is how it was worked
07:38the conflict between Uber and taxis.
07:40From 2015 to 2019,
07:4228 topics were discussed.
07:44with approximately 200,000 participants,
07:46And according to these promoters,
07:47a large majority resulted
07:48on legislative action.
07:50The momentum then waned,
07:52which brings to mind one thing,
07:53a tool can never replace
07:55political will.
