I’m told that I should remain skeptical despite having read something on the Internet. To mix metaphors for the purposes of this post: Big if always true.
The common retort I’m alluding to is used as a cudgel as it is too often and too easily invoked. Rather than engage with the presented “fact” found on the Internet, someone simply dismisses the assertion with just the declarative statement alone.
It is usually good advice; so, fair enough. But it shouldn’t substitute for actual argument and truth investigation.
In the age of AI including its tendencies (ever decreasing) to hallucinate, this takes on greater importance. Again, fair enough.
But it doesn’t end with hallucinations. A lot of what is interesting in the world is far beyond simple fact. We want to know reasons as well as context. Simple thinkers hate this. They want “The” answer(s).
“It depends” is a swear phrase to many. And diving into nuances only enrages the person asking what they think is a simple question.
They ask, “What is the tallest building in the world?” Oh, if it were only so simple.
It is nice to see that AI tends to want to provide context to even simple questions. And it is always looking to dive deeper—of course, this is also part of those companies’ business model.
AI might hallucinate some, but even those hallucinations have value. Consider this in two facets:
The threat of hallucination (should) keep us from blindly following it.
The awareness of hallucination as well as the general questioning/thinking required by us could be a way to both stay human as well as grow as a human thinker.
In the first case we are encouraged to always think about rather than just accept the output. This benefit is in part fairly straightforward. There is a deeper level, though. The act of thinking about it forces us to question not only the output but the premises—from the prompt we entered to the meaning we attach to the output.
For the second case I have in mind a partial solution to what one might term “virtual bowling alone”. As an example, perhaps AI offers a solution to the skills and culture gap for young workers or workers who otherwise need training but who want to work from home.
In addition to a lack of quantity of interaction with coworkers and mentors, young workers who work from home don’t get the spontaneous and informal training that comes from being in the workplace with others. Now with AI, they can have those conversations if they will at least take some initiative to engage with it.
Again, hallucinations in this facet are actually a feature not a bug. A big part of developing wisdom is honing judgment through trial and error learning who to believe and how strongly to believe them.
AIs often come across as authoritative, but how much more confident do they actually seem than the most confident humans? Yet we learn in engaging with confident humans that we can and should still question them and the answers they give—even if we don’t do so explicitly in response.
If I am correct that there is value like this in interactions with AI including casual interactions, then a push rather than just pull model of AI might also be valuable. So rather than just passively waiting for a human to initiate an interaction, an AI might be enabled to interrupt like a coworker would proverbially showing up in the doorway. This could as it does in real life range from watercooler talk to wanting to get one’s thoughts on something work related.
We want AI to continue to get better (smarter and more accurate). But we don’t want perfection, and this is good because perfection is an impossible quest. It will never obtain omniscience (what we want to know will always exceed what we can know), and omniscience conceptually is often a category error (there often is no single answer).