James Thorburn Can Make Mistakes
There are seven qualities that humans have, that AI doesn't:
Backstory
Friends and Colleagues
Humility and Self-doubt
Accountability and Propriety
Humour
Sense of taste, smell and tactility
A Point of View, rooted in all of the above
These absences each manifest in different ways to different degrees, but the lack of accountability has a particularly consistent substitution.
Every chatbot and chatbot mode has different vulnerabilities, but they generally provide information confidently, in an authoritative voice, in grammatical sentences, well-formatted and in some detail. Everything about this presentation makes you believe that it knows what it's talking about. Then, somewhere on the screen there is a version of this boilerplate phrase.
In this case it's, "Gemini can make mistakes."
Oh, okay - uh, anything specific you have doubts about or...
"so double-check it"
Hmm... I have to do that? How do I ... like with ChatGPT?
To err is human. The other thing that's human is owning your mistakes. Telling someone that *they* should check your work because you can make mistakes is not only an evasion of accountability, it's incredibly obnoxious behaviour.
We get it - AI runs on guesswork. It's a fundamental transition from the world of "2+2=4" to the world of "2+2 probably equals 4". When it makes factual errors or fabricates non-facts out of thin air, it isn't really 'hallucinating' - that suggests something different happened. When you are always guessing, sometimes you guess wrong. Sometimes you get caught.
Increasingly, chatbots are using techniques that involve live searches for reliable online information, which they then reformulate as a response to the user prompt, citing their sources. This is another fundamental transition - hopefully one that sticks.
But this is particularly acute because of the threat AI poses to the human authors of factual information. Wikipedia has an entire framework built around ensuring that the information someone has written in an article is accurate and sourced. There are communities of human moderators that check these things.
AI has ingested Wikipedia, regurgitates it on demand via chatbots, and made it unnecessary to ever visit the site again.
All those human writers, all that accountability, replaced with "AI responses may include mistakes"
I only use the free tier of the various chatbots, but to Wikipedia, I donate $11 every month. That's what accountability is worth to me.
Refs:
https://donate.wikimedia.org/w/index.php