In a marathon of closed-door negotiations, the EU has etched a deal for the world's first comprehensive laws for artificial intelligence (AI), a groundbreaking move with repercussions echoing far beyond the continent's borders. Simply named the "Artificial Intelligent Act", the legislation aims to ensure that the fundamental rights of democracy are protected from high risk AI whilst boosting innovating and making Europe a leader in the field.
The deal, a meticulous dance between ambition and caution, puts the most advanced AI foundation models under a magnifying glass. These models, like OpenAI's ChatGPT and Google's Bard, bred by tech titans, carry the potential to shape society. Accordingly, heightened scrutiny will be applied to these models that pose significant "systemic risks". EU regulators caution that such AI models could be harnessed for disinformation, cyberattacks, or even the development of bioweapons. Given the current lack of regulatory oversight, the EU noted that the developers of such models should be required to provide information on the data used to train the programs.
One of the most challenging points of negotiation in the deal was AI-powered facial recognition surveillance systems, with EU politicians calling for a blanket ban on its public use due to privacy concerns, with three specific exceptions. The police would be able to use the "invasive" technologies only in the event of:
an unexpected threat of a terrorist attack;
the need to search for victims; or
in the prosecution of serious crimes.
As European Commissioner Thierry Breton declares success with the "Historic!" development in a Tweet, the EU positions itself at the vanguard, setting a global precedent for AI governance. Unfortunately, the world will have to wait longer for the fine print of the legislation, expected to take effect no earlier than 2025.
The European Parliament still needs to vote on the Artificial Intelligence Act next year, but with the deal in place, approval seems likely.
In the past year, generative AI has been the showstopper of the tech world, capturing our imaginations while stoking fears of job loss, privacy invasion, and copyright conundrums. In a recent case, AI has found itself in the legal hot seat. In the UK case of Felicity Harber v The Commissioners for His Majesty's Revenue and Customs  TC09101, the First Tier Tribunal (FTT) uncovered that nine cases presented before Tribunal Judge Redstone were nothing but fabrications or "hallucinations" generated by an AI system akin to ChatGPT.
The appeal, which revolved around Harber's failure to notify the HMRC of Capital Gains Tax (CGT) on a property disposal, took an unexpected turn when it was revealed that the cases relied upon by Harber in her submissions were AI-generated. HMRC had assessed Harber for underdeclared gains related to undeclared rental income, slapping penalties on her for her failure to notify.
This isn't AI's first foray into legal fiction. The FTT highlighted a case across the pond, Mata v Avianca, Inc. 22-cv-1461(PKC), where barristers attempted to slip AI-generated fictitious cases past the Judge. They sought to rely on summaries that had "some traits that are superficially consistent with actual judicial decisions." The Judge was quick to sniff out the deception, identifying "stylistic and reasoning flaws that do not generally appear in decisions issued by United States Courts of Appeals."
Returning to the FTT case, the AI displayed a lack of legal acumen, failing to distinguish between the offences of failure to notify and late filing. The generated defense included a mishmash of cases, complete with American spellings in British judgments and suspiciously repetitive phrases. The FTT concluded that Harber was blissfully unaware that her defense was a product of AI creativity and otherwise dismissed her matter.
This case serves as another cautionary tale of using AI. While the technology has made remarkable strides in various fields, including law, relying blindly on its outputs may lead to legal misadventures that not even the best legal minds can untangle. That is why legislative and regulatory efforts like that of the EU's recent deal are positive steps toward an AI-assisted future, which at this stage, is looking inevitable.
By S Pettigrove, M Bacina and L Higgins