AI warfare is global. Engaged in the race for artificial intelligence, Google could not do without the European Union, despite regulatory constraints: several months late, the group finally launched Bard, its answer to ChatGPT, in around fifty new countries including France. And like its OpenAI competitor, Google’s AI can be stunning… or tell anything with rock-solid certainty.
Bard can now speak about 40 languages including Arabic, German, Chinese, Spanish, French and Hindi, according to Google.
Good command of French
This conversational robot is able to generate a text in perfect French. Good luck detecting if it was written by a human or the machine. It can save time to compose a complaint email or even give ideas in case of blocking to write a social media post or a birthday card.
We can refine the result, for example by asking the AI to adopt a more professional tone, or to be accessible to children. On the creativity side, the machine is struggling for the moment to be original and still does not handle humor well. Bard is now able to read his answers orally, with a more natural voice than the original Google Assistant.
A more powerful search… When AI does not invent
It’s a problem that engineers can’t solve. These AIs tend to “hallucinate”. They invent or tell nonsense with a tone as certain as an expert. “Find me a restaurant in LA with a great view and vegetarian options that isn’t a restaurant vegan returns a promising list. Except that two of the six results do not exist or are not in the cited district.
When asked if he’s sure, Bard often apologizes: “I’m sorry, I was wrong. There is no restaurant named The Penthouse at the Beverly Wilshire. There is however a restaurant named The Penthouse at the Four Seasons Hotel Los Angeles at Beverly Hills. Except that’s also false.
Bard is also able to do summaries. That of an article by 20 minutes on Thursday’s last speech by Joe Biden starts off well on substance… except the US president was in Finland and not Poland – his Warsaw speech dates back to February. He shouldn’t be trusted as a medical assistant either: Bard recommends dancing as a recovery sport after a dislocated patella.
Fears in Europe
The American group had to delay the launch of Bard in the EU after questions from the Irish personal data regulator (the DPC). A worrying signal because its competitor ChatGPT, also accused of not respecting European regulations, had for its part been blocked at the end of March for a month in Italy.
“Google has made a number of changes ahead of launch, in particular improving transparency and changing some control mechanisms for users,” DPC spokesperson Graham Doyle said in a statement to the AFP. “Google has agreed to conduct an audit and provide a report to the DPC three months after Bard becomes operational in the EU,” he continued.
Artificial intelligence algorithms raise many fears in terms of privacy, disinformation or respect for intellectual property. Brussels wants to regulate the sector via a text currently under discussion and the European Data Committee has also launched a working group on the subject.
Bard is thus presented as a “creative tool”, “experimental”, which “can display inaccurate or shocking information”. A warning also present in its competitor.
Their economic model is still uncertain, between free for Bard or Bing AI, paid features for ChatGPT or access by subscription, while their operating costs are much higher than for a traditional search engine.