OpenAI’s thirst for data is coming back to bite it

In AI development, the dominant paradigm is that more training data is better. OpenAI’s GPT-2 model had a dataset consisting of 40 gigabytes of text. GPT-3, on which ChatGPT is based, was trained on 570 GB of data. OpenAI has not shared the dataset size of its latest model, GPT-4.

But that thirst for bigger models is now coming back to bite the company. Over the past few weeks, several Western data protection authorities have launched investigations into how OpenAI collects and processes the data that powers ChatGPT. They believe he took people’s personal details, such as names or email addresses, and used them without their consent.

The Italian authority has blocked the use of ChatGPT as a precaution, and French, German, Irish and Canadian data regulators are also investigating how the OpenAI system collects and uses data. The European Data Protection Board, the umbrella organization for data protection authorities, is also setting up an EU-wide task force to coordinate investigations and enforcement around ChatGPT.

Italy has given OpenAI until April 30 to comply with the law. This would mean that OpenAI would have to ask people for their consent to have their data deleted, or prove that it has a “legitimate interest” in collecting it. OpenAI will also need to explain to people how ChatGPT uses their data and give them the power to correct errors about them that the chatbot spits out, have their data erased if they wish, and object to the computer program spitting them out. used.

If OpenAI fails to convince authorities that its data use practices are legal, it could be banned in some countries, or even the entire European Union. It could also face hefty fines and could even be forced to delete models and the data used to train them, says Alexis Leautier, an artificial intelligence expert at the CNIL.

OpenAI’s breaches are so egregious that this case is likely to end up in the Court of Justice of the European Union, the EU’s highest court, says Lilian Edwards, professor of internet law at the University of Newcastle. It could take years before we see an answer to the questions posed by the Italian data regulator.

High stakes game

The stakes couldn’t be higher for OpenAI. The EU General Data Protection Regulation is the strictest data protection regime in the world, and it has been widely copied around the world. Across Brazil to California, regulators will be paying close attention to what happens next, and the outcome could fundamentally change the way AI companies go about collecting data.

In addition to being more transparent about its data practices, OpenAI will have to show that it uses one of two possible legal ways to collect training data for its algorithms: consent or “legitimate interest”. .