If 2022 was the yr the generative AI growth began, 2023 was the yr of the generative AI panic. Simply over 12 months since OpenAI launched ChatGPT and set a document for the fastest-growing client product, it seems to have additionally helped set a document for quickest authorities intervention in a brand new expertise. The US Federal Elections Fee is trying into misleading marketing campaign advertisements, Congress is looking for oversight into how AI corporations develop and label coaching knowledge for his or her algorithms, and the European Union handed its new AI Act with last-minute tweaks to reply to generative AI.
However for all of the novelty and pace, generative AI’s issues are additionally painfully acquainted. OpenAI and its rivals racing to launch new AI fashions are dealing with issues which have dogged social platforms, that earlier era-shaping new expertise, for practically twenty years. Firms like Meta by no means did get the higher hand over mis- and disinformation, sketchy labor practices, and nonconsensual pornography, to call just some of their unintended penalties. Now these points are gaining a difficult new life, with an AI twist.
“These are utterly predictable issues,” says Hany Farid, a professor on the UC Berkeley Faculty of Info, of the complications confronted by OpenAI and others. “I feel they had been preventable.”
Nicely-Trodden Path
In some instances, generative AI corporations are instantly constructed on problematic infrastructure put in place by social media corporations. Fb and others got here to depend on low-paid, outsourced content material moderation employees—usually within the World South—to maintain content material like hate speech or imagery with nudity or violence at bay.
That very same workforce is now being tapped to assist prepare generative AI fashions, usually with equally low pay and troublesome working circumstances. As a result of outsourcing places essential features of a social platform or AI firm administratively at arms size from its headquarters, and sometimes on one other continent, researchers and regulators can wrestle to get the complete image of how an AI system or social community is being constructed and ruled.
Outsourcing can even obscure the place the true intelligence inside a product actually lies. When a chunk of content material disappears, was it taken down by an algorithm or one of many many hundreds of human moderators? When a customer support chatbot helps out a buyer, how a lot credit score is because of AI and the way a lot to the employee in an overheated outsourcing hub?
There are additionally similarities in how AI corporations and social platforms reply to criticism of their unwell or unintended results. AI corporations speak about placing “safeguards” and “acceptable use” insurance policies in place on sure generative AI fashions, simply as platforms have their phrases of service round what content material is and isn’t allowed. As with the foundations of social networks, AI insurance policies and protections have confirmed comparatively straightforward to bypass.