Google senior vice president and head of Google Search Prabhakar Raghavan warned users about "hallucinating" current versions of AI-powered chatbots. In his opinion, chatbots' responses should not be completely trusted because some of them may be fictitious, and solution makers do not fact-check the information generated by the bots.
"This type of artificial intelligence we're talking about now can sometimes lead to what we call hallucinations. It's expressed in such a way that the system gives a convincing but completely made-up answer," Raghavan told.
The head of Google Search added that one of Google's main objectives is not only to develop the Bard chatbot, but also to minimize the number of incorrect answers due to digital hallucinations in AI.
According to Raghavan, Google is faced with the current urgency to release its chatbot in order not to miss the market. But the company also understands the greater responsibility for this project. Developers don't want to mislead users and release a raw chatbot to the public right now, which in some cases can come up with a convincing answer and fool the user with expectations.
Scientist and astrophysicist Grant Tremblay agrees with Raghavan. He explained to the media that ChatGPT and Bard technologies look eerily impressive, but they are often "very confident" to make mistakes and fail to self-check for errors. According to the scientist, the main problem with artificial intelligence chatbots is their tendency to confidently state incorrect information as fact. "These systems often 'hallucinate,' that is, make up information, because they are essentially autocomplete systems," Tremblay believes.
On Feb. 6, 2023, Google unveiled its ChatGPT counterpart, an experimental dialog AI service called Bard. Astronomers and astrophysicists have since revealed that Google's Bard chatbot made a factual error during the first demonstration of its use. Its response indicated that the James Webb telescope took the very first images of an exoplanet outside the solar system. But it didn't. The first image of an exoplanet was taken in 2004 with the Very Large Telescope (VLT) complex in Chile using adaptive optics.
Subject matter experts explained to the media that instead of querying a database of verified facts to answer questions, chatbots learn from huge arrays of text and analyze patterns to determine which word to put next in any given sentence. In other words, they are probabilistic, not deterministic. Because of this, some in the industry consider chatbots to be "nonsense generators." According to experts, the Internet is already full of false and misleading information, but the problem is exacerbated by Microsoft and Google's strong desire to use chatbots as search engines. As a result, ChatGPT and Bard answers will soon gain the credibility of a potential omniscient machine when these projects are launched for general use.
Notably, after the hasty and, as it turned out, not entirely successful presentation of the chatbot Bard, Google employees criticized the actions of the head of the company Sundar Pichai on an internal forum, calling them hasty, a failure and comically short-sighted. It turned out that the presentation was made in such a hurry that many employees of the company learned about it at the last minute, and the presenter of the event did not have the necessary gadgets for the demonstration and he had to improvise in front of invited clients.
Earlier in February 2023
On February 7, Microsoft introduced new versions of the search engine Bing and the browser Edge using AI and opened registration for a new version of the search engine Bing based on ChatGPT bot. After that, the popularity of Microsoft's Bing mobile app increased 10-fold. In the U.S. the Bing app on the App Store climbed to number 12 in the top chart of all free apps for iOS, and the Microsoft Edge browser is now number 3.
ChatGPT had over 100 million active users in January. In two months, the service's audience grew 100 times. In the past 20 years, there has never been a service on the Web with a faster user base growth. It took TikTok about nine months after its global launch to reach 100 million monthly users, and other popular social networks up to 2.5 years.
ChatGPT can pass an MBA exam, solve a test assignment, create simple polymorphic viruses, as well as help in writing a diploma in higher education and even answer most of the questions of the USMLE medical exam for the right to work as a doctor in the U.S.