The future of AI regulation: From data to algorithm deletion

OneTrust founder and CEO Kabir Barday warned that impending regulations may soon require AI companies to delete entire algorithms due to data errors.

Speaking at a recent Web Summit event, Kabir predicted there will be stronger regulations on AI, with companies such as OpenAI potentially having to delete entire algorithms if an error is found in the data that trains their LLMs.

“A lot of the problems facing data have always been there. There’s data everywhere. There are new laws. This stuff’s always been around,” said the CEO, while addressing the Centre Stage audience at Collision 2024 (our former North American event, succeeded by Web Summit Vancouver).

“What’s new with AI that’s driving the urgency is the fact that there is no machine unlearning. Once you put data into a machine learning algorithm … that data is there permanently.”

Kabir continued, “That’s different than in the business analytics and dashboarding era, where you can just delete data from your data warehouse and it’s done. You can’t do that in AI, and regulators have picked up on this.”

Kabir explained that enforcement actions used to consist of a fine and a data deletion order, whereas now, they may have to be full algorithmic deletion orders.

A new era of regulatory scrutiny for AI

“The FTC, in the last year, I think, just issued their first algorithmic deletion order to a company. Imagine, I don’t know, OpenAI – if one mistake they made in the data [is] put into that LLM, when these regulations pass, they’ll have to delete the entire algorithm,” explained the CEO.

“My prediction is that day is going to come [for OpenAI]. My prediction is they’re gonna have to delete their entire algorithm because of one mistake. I think there are lots of mistakes. All these companies are gonna have this problem, so you have to get this right up front.”

Big tech companies have a data advantage

Kabir also shared thoughts on legacy tech companies, such as Apple and Microsoft, going all in on AI in the past couple of years, arguing that success in AI tends to favour these established brands over startups.

“This is because established companies have more data, which gives them a competitive edge in developing AI solutions. Even if a startup is the first to create a new AI tool, larger companies can quickly catch up by integrating their own data through open APIs, effectively neutralising the startup’s advantage,” said Kabir.

“AI favors the incumbent, not the startup, because the differentiation comes from how much data you have, which is proportional to how many customers you already have.”

So, according to Kabir, if you’re a startup that manages to build a novel AI application, the minute a larger company builds an open API and uses its own data, “you’re dead”.

Web Summit 2024 is getting closer. Book your tickets now.

Main image: Web Summit (CC BY 2.0)