Amid growing calls to halt the development of artificial intelligence, Italy has blocked the use of OpenAI’s massively popular ChatGPT AI language model, citing privacy concerns.
In a press release Friday, the Italian Data Protection Authority announced it is blocking access to the ever-advancing AI chatbot and is starting an investigation into Microsoft-backed developer, OpenAI, to determine whether the company has violated the GDPR (General Data Protection Regulation).
“There appears to be no legal basis underpinning the massive collection and processing of personal data in order to “train” the algorithms on which the platform relies,” the Italian watchdog said.
Regulators also cited a data breach OpenAI reported on March 20 involving user conversations and payment information.
In addition to the unlawful collection of user data and cybersecurity concerns, the agency said the app offers no way of verifying the age of its users, exposing “minors to absolutely unsuitable answers compared to their degree of development and awareness.”
However, the Italian Data Protection Authority noted that the ban is temporary and could be lifted if the company complies with EU regulations. The agency has given OpenAI 20 days to address its concerns. The company could incur a €20 million (nearly $22 million) penalty or pay out up to 4% of its annual revenue if it fails to comply.
An Influx of Calls for AI Pause
In recent days, several high-profile players in the tech world have called for a halt on fast-developing AI tools like ChatGPT, Google Bard, and the recently released Microsoft Security Copilot.
These AI-driven systems are known as large language models (LLMs). They can effectively reproduce human writing — more or less — based on data from the internet as it existed in 2021, as well as countless digital books and writing. LLMs also learn from the millions of users who utilize them daily.
LLMs have successfully passed business and law school exams as well as U.S. medical licensing exams, although they may not have earned perfect marks.
On Thursday, the Center for AI and Digital Policy filed a complaint against OpenAI, asking the Federal Trade Commission to halt the development of ChatGPT and other AI models.
“The Federal Trade Commission has declared that the use of AI should be ‘transparent, explainable, fair, and empirically sound while fostering accountability.’ OpenAI’s product GPT-4 satisfies none of these requirements,” the complaint said. “It is time for the FTC to act.”
In an open letter earlier this week, a contingency of tech leaders — including Elon Musk, Apple co-founder Steve Wozniak, and 2020 presidential candidate Andrew Yang — called for a moratorium on the “dangerous race” of AI-powered systems.
“Contemporary AI systems are now becoming human-competitive at general tasks, and we must ask ourselves: Should we let machines flood our information channels with propaganda and untruth? Should we automate away all the jobs, including the fulfilling ones?” the letter reads.
“Should we develop nonhuman minds that might eventually outnumber, outsmart, obsolete and replace us? Should we risk loss of control of our civilization?”
Questions About AI, Personal Privacy Loom
While OpenAI says it collects user data to train its deep-learning AI model — it had amassed nearly 100 million regular users as of Jan. 2023 — the company hasn’t offered much insight into the other ways this data is used.
“The faster you grow, the faster the regulators react. Italy’s ban is about privacy and a breach, but raises fascinating new challenges,” Arun Sundararajan, NYU Stern School of Business Professor of Technology, said in a tweet Friday.
Speaking to VPNOverview, Sundararajan said individuals can’t do much “by themselves to restrict the extent” their data is used to train AI. While personal data has long been used to serve targeted adverts to customers, its purposes for large language models (LLMs) have yet to be fully revealed.
“Nobody, not even the scientists at OpenAI understand why ChatGPT is doing what it’s doing,” Sundararajan said. “This takes us into a world where our data is being used to train machine learning models. But if data that shouldn’t have been used was used — once you put the training in, there’s no way of taking it back out.”
With more questions than answers remaining, the damage may have already been done.
