A University of Washington linguistics professor described AI chatbots as “a terrible technology for information access” shortly after OpenAI rolled out SearchGPT for paying users.

The promise of a chatbot search engine sounds appealing: no need to endlessly click through pages of search results that can't guarantee you have the answer you're looking for; AI chatbots aim to give you specific responses to your queries in the manner you request. It's not an exaggeration to say that SearchGPT also rivals the offerings of features you can find on Google and Bing, like live data access to weather, news, sports, and stocks.

Want your latest medical discovery explained to you as if you were a five year old, no problem with SearchGPT, but Google's best bet is to send you a link to Reddit.

Despite these obvious conveniences, however, not everyone is convinced that AI chatbot search engines are necessarily the way to go. Professor Emily M. Bender of the University of Washington elaborates on her argument in X as to why she thinks LLMs, the brains behind chatbots like ChatGPT and Claude, are bad for search.

Vendor explained that LLMs are trained to generate sequences of words that sound like natural language, but the chatbot you are asking about does not really understand what it is saying.

“If someone uses LLM as an alternative to search and the output they get is correct, it is just a coincidence,” Bender wrote. She also argued that a search engine chatbot that is correct 95% of the time is more dangerous than one that is correct 50% of the time.

“People are more likely to trust its output and less likely to be 5% fact-checked,” Bender wrote.

Not all comments received on the vendor's claims agreed, but the vendor is not alone in questioning the AI search engine chatbot.

“Our lives will be more convenient and streamlined, but perhaps a little less full of splendor and wonder and illumination,” Matteo Wong wrote in The Atlantic.

Like Wong, vendors believe that the use of chatbots for Internet searches will reduce our ability to find, evaluate, interpret, and learn from a wide variety of different sources of knowledge. one Google search, but the midnight of strange gadgets presented at conferences such as CES may have ignited the spark that drives us to deep dive, but a chatbot may have satisfied the initial inquiry about when the next consumer electronics show will be held.

So even if tools like SearchGPT were not based on LLM, vendors would not be persuaded to use them, even if, by her own admission, traditional search engines have their own shortcomings that put commercial interests above the public good She stated that she would not be.

“But just because the current system is bad doesn't mean any change will make it better,” Bender said.