Expensive AI blooper for GoogleBack
Google lost tens of billions on Wall Street this week. The price of shares fell by no less than 8 percent, when it turned out that an advertisement for the new chatbot perfectly reflected the limitations of this new form of artificial intelligence.
Google introduced the new service Bard this week, in response to Microsoft's integration of ChatGPT into the search service Bing. In both cases, artificial intelligence (AI) is used to present the search results in a coherent story.
Bard got it wrong with the question "What can I tell my 9-year-old about the discoveries of the James Webb Space Telescope". The chatbot replied, among other things, that the telescope "has taken the first pictures of a planet outside our own solar system". But such pictures were already taken in 2004 by the Very Large Telescope.
The flaw was discovered yesterday, just as Google was giving a presentation on all the AI innovations it is working on. In a response, Google says that testers are still working hard to remove teething problems from the system before it is widely rolled out.
Ironically, Google's trusty old search engine did give the right answer...
Hallucinating AI tools
Experts have long pointed to this kind of error, called "hallucinating", because AI can very convincingly deliver total nonsense. In this case, Bard was referring to the first time the Webb telescope captured exoplanets, but the program interpreted this as the first time ever.
An important blunder, because Bard must ensure that Google remains the market leader in search engines.
Incidentally, rival ChatGPT can also do something about it. The search engine told in detail about the result of the Superbowl, a match that still has to be played... The chatbot not only mentioned the result (31-28), but also made up some specific match statistics.
The launch of the new chatbots seems somewhat rushed and motivated more by competitive motives than reliability. Perhaps we should tentatively check a chatbot's answers against Google's current search engine before taking them for granted? In time, the chatbots will undoubtedly become more reliable, but until then healthy suspicion seems justified.