The race is on for the tech overlords to incorporate AI into everything possible. Google now offers a Gemini AI overview at the top of the page of search results whether wanted or not, and users can even get Gemini AI summaries without any web links whatsoever, while Bing has Copilot ready to listen and answer questions. The problem is that as search engines go, AI chatbots are all pretty much crap.
Study Finds That AI Search Engines Are Wrong an Astounding Proportion of the Time
A recent study found that AI chatbots gave out wrong information to an average of 60% of queries. Perplexity, from a start-up of the same name, was most accurate with only 37% wrong, while Musk’s Grok 3 got it wrong an amazing 97% of the time. That’s way more than simple error alone would create, one would think. After all, as the old saying is, a broken clock is right twice a day. So was it deliberately trying to lie?
AI Search Has A Citation Problem
The tests were remarkably simple too. Ten articles from various publishers were chosen. The chatbots were given quotes and asked to identify the original article’s headline, date, original publisher and URL. Pasting the same information into Google’s engine returned the source within the top three results. However, the tests showed some highly disturbing results.
The chatbots seemed unrestrained by content licensing deals to guarantee information quality as well as exclusions of data from search engines. They took information wherever they found it, true or not, private or not. Yet, the paid chatbots were even more confident in their mistakes than their free counterparts. In general, bots would not admit any inability to answer, but reply incorrectly or speculatively without eveadmitting it. In other words, they did not say “it appears”, “might”, or “maybe”. So just like the president, AIs easily make up stuff and pronounce it with utter confidence, even doubling down when questioned further.
At least the bots have an excuse — they are programmed to generate answers. They don’t know any facts, they have no self-consciousness, conscience, or emotions: all they really know is how humans put one word after another. They are essentially sales agents, basically, and lousy ones at that. No wonder Elon loves them.
Hopefully, this will prove to be a fad just like connecting everything to the net was during the Internet of Things craze a few years back, when even waterbottles were online. Fortunately, there are a few workarounds for some, but not all, of Google’s AI slop. But with truth under such assault these days, this could not be happening at a worse time. As it was most likely intended.
