Why is it so difficult to regulate technologies like artificial intelligence? In 1980 Professor David Collingridge described this dilemma as a paradox. To test the real impact of an innovation it is necessary to wait for it to develop and spread: intervening immediately would make the rules incapable of governing its future implications. However, once the technology is mature and entrenched, we risk losing the power and ability to regulate it. Collingridge’s dilemma is linked to the “pace problem” postulated by scholar Larry Downes: technology evolves exponentially, while legal, social and economic systems do so incrementally.
Getting out of this impasse is not easy. For this reason, for some years a new attempt has been being experimented around the world to bring technological development closer to regulation. These are the so-called regulatory sandboxes, and could soon also be adopted for artificial intelligence.
Digital rights Why is the Italian government on the side of those who don’t want to regulate AI? by Diletta Huyskes 24 November 2023
What are
Regulatory sandboxes are environments, physical or virtual, in which companies can experiment with new technological products or services under the supervision of industry authorities, for a limited period, and with the possibility of benefiting from exemptions from current regulations.
The term “sandbox” refers to what in computer science is the typical application testing environment. Equally pertinent is the reference to the sand boxes where the little ones have fun with a shovel and bucket: exactly like children, authorized by their parents to experiment with new forms of creativity in an isolated and controlled space and for a specific time, enjoying physical rules Completely exceptional, manufacturers participating in a sandbox can test new business models safely and with greater freedom of action.
At a procedural level, the traditional scheme provides for the establishment of these spaces through laws that regulate every aspect. Interested companies must then contact the responsible public body, nominating their innovative project. Once eligibility has been assessed, the trial can start according to the instructions and in constant dialogue with the relevant authorities.
By participating in a sandbox, companies have the opportunity to experiment with products or services in a real but risk-controlled environment, in a regime of temporary exemption from sector regulatory requirements, in close synergy with regulators and authorities. At the same time, they can gather new data and expertise to verify effectiveness and, where appropriate, improve the regulatory framework and supervision.
The idea An EU stamp for the quality of artificial intelligence by Brando Benifei* 26 March 2023
From fintech to self-driving cars
According to OECD data, there are around a hundred sandbox initiatives worldwide today. In any case, it is a paradigm born in more recent years. Similar projects began to emerge in 2012, with the first regulatory sandbox officially launched in 2015 by the UK Financial Conduct Authority in the financial services space. Fintech was the cradle of spaces for regulatory experimentation: many other countries followed the Anglo-Saxon example, such as Hong Kong, Singapore, India and Arizona. This unprecedented approach has also found widespread use in other sectors, such as energy, healthcare or transport, in particular with regard to self-driving cars. There is also no shortage of transversal sandbox programs: a line followed, for example, by Japan, Germany and Portugal. At the beginning of 2023, the European Union launched a regulatory experimentation space for blockchain (the European Blockchain Regulatory Sandbox).
The Italian experience
Italy has also introduced a space for regulatory experimentation in the fintech sector. This happened at the hands of Legislative Decree no. 34 of 30 April 2019 (Growth Decree) and the subsequent Decree of the Ministry of Economy and Finance n. 100 of 30 April 2021. This regulatory framework, which identified the Bank of Italy, Consob and Ivass as supervisory authorities, was also recently updated by Legislative Decree no. 25 of 17 March 2023 (Fintech Decree) with a view to simplification. Alongside the financial innovation sector, with Legislative Decree no. 76 of 16 July 2020 (Simplifications Decree), the “Sperimentazione Italia” project was also launched, which instead follows a multi-sectoral approach.
Interview The first Italian bill on AI is ready. Ascani: “He will protect us from deepfakes” by Pier Luigi Pisa 14 November 2023
Regulatory sandbox e IA
In light of the experiences and results collected, the use of regulatory experimentation spaces for artificial intelligence systems could not be missed. In Norway, for example, the privacy authority (Datatilsynet) has inaugurated a sandbox focused on the development of AI that respects the rules on personal data protection. A similar initiative was also undertaken in Colombia.
Last year, the European Commission and the Spanish government presented a pilot project for the creation of the first regulatory sandbox on artificial intelligence. This news anticipates what is foreseen by the Artificial Intelligence Act. In fact, the new European regulation on AI also includes among the measures to support innovation the establishment of regulatory experimentation spaces, defined as “a controlled environment established by a public authority which facilitates the safe development, testing and validation of innovative AI systems for a limited period of time before their placing on the market or putting into service in accordance with a specific plan subject to regulatory oversight”. Furthermore, with the version of the law proposed by the European Parliament, which was also responsible for the inclusion of the definitive proposal, each member state is required, alone or with others, to establish at least one space for regulatory experimentation at national level, which it must be operational at the latest on the day of entry into application of the regulation itself.
The importance that the European legislator intends to attribute to regulatory sandboxes for the future development of AI is therefore evident. Collingridge argued that “when change is easy, the need for it cannot be foreseen; when the need for change is evident, change has become expensive, difficult and time-consuming.” Only experience will tell us whether spaces for regulatory experimentation can actually contribute to resolving this paradox.
The OpenAI experiment: ChatGPT enters politics by Bruno Ruffilli 28 May 2023