Google launches generative AI model Bard in Europe

This article is part of our special report AI4TRUST – AI-based-technologies for trustworthy solutions against disinformation.

Bard, Google’s eagerly awaited response to ChatGPT, was launched in Europe on Thursday (13 July), following delays in complying with the EU’s data protection rules.

Google is a leading company in the field of Artificial Intelligence, especially in the rapidly growing category of generative AI, large language models able to produce human-like content based on user input.

Until last year, tech companies did not release their language models to the public for fear of possible societal risks. The situation turned dramatically when the start-up Stability AI publicly released its text-to-image model Stable Diffusion last August.

Just three months after, OpenAI released the chatbot ChatGPT to the public, which given the unprecedented human-like quality of the responses, quickly became the fastest-growing internet service.

In January, Microsoft poured $10 billion into OpenAI and integrated ChatGPT in its Bing search engine, throwing the gauntlet at Google over AI-powered online searches.

In response to rising expectations from users and investors of playing catch-up with Microsoft-backed OpenAI, in February, Google announced Bard, a conversational AI-assisted search tool.

Following some regulatory hurdles, the tool will now finally be available in Europe.

“We believe the terminology Chatbot is too limiting for the capability that it has, too limiting for our ability to create,” Google’s product director, Jack Krawczyk, told reporters ahead of the launch.

“We talk about Bard as a form of AI that we are calling ‘augmented imagination’."

Regulatory compliance

Krawczyk said they had held productive conversations with data protection authorities across Europe, not only the Irish.

This regulatory dialogue resulted in an approach that he defined as based on transparency over the use of data, the choice for users to allow Google to use the information and control to provide feedback requesting Bard to correct wrongful output.

Krawczyk said Bard’s privacy features will be the same also outside Europe.

“We will be continuing our engagement with Google in relation to Bard post-launch, and Google have agreed to carry out a review and provide a report to the DPC after three months of Bard becoming operational in the EU,” Ireland’s Deputy Commissioner Graham Doyle told EURACTIV.

At the same time, the European Data Protection Board, the body that gathers all EU privacy watchdogs, launched a task force on ChatGPT to find a coordinated approach to generative AI models following Italy’s temporary ban on data protection violations.

"Google’s launch of Bard in Europe causes several concerns. The US version we have seen thus far is even worse than ChatGPT or BingAI when it comes to revealing and acknowledging sources and providing links to them. Bard used millions of copyright protected works to train, without any consideration," Thomas Höppner, a partner at Hausfeld, told EURACTIV.

Generative AI systems are also set to fall under the scope of the AI Act, an EU flagship legislation to regulate Artificial Intelligence based on its potential to cause harm. The draft law is at the final stage of the legislative process, and how to deal with Bard and the like will be a key topic to hash out.

“I welcome that they took their time, and I venture to hope the reason for the delayed release was the understanding that, for the European market, consumers' interests and rights need to be fully protected. This shows our rules work and are feasible,” said Dragoș Tudorache, an EU lawmaker leading on the AI Act, reacting to Bard’s belated launch in Europe.

European data protection authorities launch task force on ChatGPT

The European privacy regulators decided on Thursday (13 April) to launch a dedicated task force to address the privacy concerns related to the world’s most famous chatbot.

New features

In addition, Bard introduced the possibility for users to maintain and revisit previous conversations, which can be continued at a later stage or deleted afterwards. Users might also fully or partially share the responses via a link.

Developers that use the language model for coding can export Python code to Replit, a service to create online projects.

Another feature, initially only available in English, will allow users to adjust the tone and style of the AI model’s outputs, such as long or short, professional or casual. A second English-only capability will be to use images as prompts by connecting Google Lens to Bard.

“In the future, Bard is going to introduce more information from other products across Google,” Krawczyk added, mentioning services such as Google Maps and Gmail.

AI Act: MEPs close in on rules for general purpose AI, foundation models

The European Parliament is set to propose stricter rules for foundation models like ChatGPT and distinguish them from general purpose AI, according to an advanced compromise text seen by EURACTIV.

[Edited by Zoran Radosavljevic]