Ok, I’m on the warpath. I have spent the majority of my life caring about the meaning of words. The “AI” hype has gone too far. Time to put up a fight.
“Prompt Engineering”
Poking a village idiot in different ways with different sticks to illicit a response we hope will be useful to us is not “engineering”. LLMs are nothing more than glorified autocompletes and to think that our “cleverness” in prompting them can be called “engineering” is a disgrace to all true “engineers”. When you ask a question, you are “querying” a system. Quit quit quit calling this “engineering”.
“Generative AI”
Statistical mappings of text are anything but “intelligent”. We have come up with very “clever” ways to generate text and images and audio and video. Bravo! They are semi useful tools but very limited and they certainly are not intelligent. So stop calling it “AI”. It is generative computing.
“Agentic AI”
People are trying so damn hard! Trying to turn lead into gold. Statistical systems are never reliable. They are by nature “averages”. They are by nature not logical. They are probabilities. We cannot fix them using clever tricks including breaking them into smaller pieces and pretending they will do better. Agentic “AI” only exists because someone tried to fix LLMs. People who do this continue to ignore math. Those pipelines, not matter what “clever” glue we put in between them, still rely on non-logical, non-trustworthy statistics. They are agentic pipelines at best.
“AI Bias”
Come on! This one is way too easy: Training data bias. This reminds me of snake oil salesmen who take ordinary things and slap a fancy name on them and voilà: instant sale!
“AI”
The last program that I truly think was “artificial intelligence” was SHRDLU, Terry Winograd’s “blocks world” he created at MIT. It knew EVERYTHING about the block’s world and could interact with humans. It showed that if a computer program knows everything about some specific world just like a human, we could call it intelligent. That was back in 1968-1972. Since then, we claim “AI” when the program it is not “intelligent” at all. This one Bender and Hanna get right in their book “The AI Con”: describe the task. Throw away the phrase “AI”.
Emotion Versus Logic
All of this hype vocabulary exists because we humans think that computers that can run immense mathematical operations using immense computer farms that doing calculations that we humans could never do, will somehow magically become “intelligent”. That is what is happening with LLMs.
We are no different from our ancestors who believed in magic. We have not evolved. Our emotions still conquer logical truth. Trying to make statistical systems “intelligent” or think that some clever “trick” will turn them “intelligent”, is an emotional problem, not a logical one. Spock from Star Trek was right about humans and their problems with emotion – despite the episodes and movies trying to convince ourselves otherwise.
From now on, I will stop using the hype phrases and use truer ones.
