The bloc’s omission hints at the broader power struggle between regulators and big tech in Europe. At the very least, the search giant needs more time to compare Bard with the EU AI Bill, says Henk van Ess, visiting professor at the Freetech Axel Springer Academy of Journalism and Technology in Berlin. “The proposed regulation emphasizes the importance of transparency and traceability in AI systems,” he says. “It can be challenging for large language models like Google Bard to fully meet this requirement, as the decision-making process in these models can be complex and difficult to interpret.”
The AI Act rules may also pose a problem if Google has trained Bard on a dataset that contains errors or biases, van Ess adds. In April, researchers found they could prompt Bard to deny climate change, mischaracterize the war in Ukraine and question the efficacy of the vaccine. “Google is playing it safe,” says Robin Rohm, founder of Berlin-based AI startup Apheris. “They recognize that Bard could be considered a product that enables high-risk applications, and this could expose them to risk under the proposed regulation. The delay could be an effort to gain time.”
Any high-profile mistake in the EU could cost the company dearly in the months and years to come. Google will be sensitive to the fact that what happens now will likely influence negotiations around the AI Act, according to Daniel Leufer, a Brussels-based senior policy analyst at digital rights group Access Now. “If ChatGPT, Bard, etc., in the next six to seven months, are responsible for serious public errors, then measures that would address those errors could very well find their way into the AI Act,” he says.
The draft AI Law rules are not expected to be approved until next year, but other EU regulations may already be giving Google headaches. Europe’s new Digital Services Act could also play a role when it comes to Google incorporating Bard into its search settings, says Harshvardhan Pandit, an assistant professor at Dublin City University’s Adapt Center. “Since Bard also acts as a search engine, Google may also be testing integrating ads into it and doesn’t want to be subject to the DSA at this time,” says Pandit. The DSA presents new rules about online advertising.
With competition to push the build of more generative AI services, Europe’s privacy laws are already causing problems for new services. “There is a lingering question as to whether these very large data sets, which have been collected more or less through indiscriminate scraping, have a sufficient legal basis under the GDPR,” says Leufer. At the end of March, Italy’s data regulator temporarily banned ChatGPT for not following the bloc’s GDPR privacy rules. OpenAI had “illegally” collected personal information from the web as part of its training data, the regulator said, as well as failing to tell people how their data was used or develop tools to prevent children from using ChatGPT.
The move resulted in OpenAI making changes to allow people to remove more data from ChatGPT, along with a host of other concessions. While ChatGPT is now available again in Italy, the country’s data regulator is still vetting the technology. The decision prompted other EU countries to create a joint working group to further investigate ChatGPT. The Irish Data Protection Authority, which handles GDPR matters relating to Google, Meta, Microsoft and Apple, did not immediately respond to a request for comment on whether it had discussed launching Bard in Europe with Google.
However, GDPR rules may be one of the reasons why Bard is not launching in the European Economic Area (EEA), a group of countries that includes the EU bloc, Norway, Iceland and Liechtenstein, says Judin from Norway. He adds that the Norwegian authority does not have any information as to why the EU and EEA have been excluded from the Bard launch.