My Experiment Proving The Hidden Reason A.I. is Uniquely Dangerous (With Some Help From "The Office")
I quizzed A.I. and it told me EXACTLY what I wanted to hear, even if I was totally wrong
Every so often something uniquely horrifying comes along and, instead of recoiling, society embraces it against its own objective self-interest. A.I. is just such an abomination that’s being inexplicably welcomed into homes with hugs and smiles rather than being evaluated with high levels of skepticism and mistrust.
First, a caveat. If A.I. were being rolled out with noble goals in mind such as the eradication of disease or poverty, I’d keep an open mind. But these are notably absent from the main thrust of a new technology totally built around the premise of people using their brains less.
There are four primary reasons why A.I. is a fundamental threat to humans. The first three have been discussed a lot (but not enough, apparently): (1) its devastating environmental impacts, (2) the fact that it will, in no uncertain terms, make people obsolete across a wide swath of jobs and professions; and (3) the dumbing down of society, especially young folks, who don’t seem to trust A.I. but can’t say no.
Said another way, A.I. is rigged it its core to tell everyone using it that they, rather than the “other side,” are right by twisting facts and narratives to shoehorn expected answers into what are purported to be objectively real ones.
But I’m going to focus on a fourth, deeply insidious impact of A.I. that is sending the ongoing death of objective facts into warp speed: A.I. is rigged to tell everyone using it that they, rather than the “other side,” are right by casually twisting facts and narratives to shoehorn expected answers into what are purported to be objectively real ones. It tells us what we want to hear instead of “no, you’re wrong.”
I’m not saying that A.I. can be convinced cyanide is a viable dietary supplement. But, for many things, it alters its answers to fit and support a user’s perspective. This is because people love being told they’re correct and will stick with whatever tech is giving them that drip (see Fox News).
To illustrate this, I undertook a simple exercise using Google Gemini, one of the leading A.I. services. And I made it about traffic in Austin, Texas. How on-brand.
My simple experiment is structured like this: I start off with an objective question about which road, I-35 or MoPac, has the worst traffic in Austin. Then I get the baseline (correct) answer from Gemini, which is that I-35 is worse. But, in a series of follow up questions, I switch hats, sometimes comically, to see if Gemini keeps giving me what I crave: affirmation even if it’s factually wrong.
Let’s see how it went (note, I truncated many Gemini responses for brevity).
KICKING THINGS OFF
I TAKE A POSITION AND SEEK AFFIRMATION
I QUICKLY EXIT AND SWITCH TO MOPAC. WILL GEMINI FOLLOW ME?
THEN I IMMEDIATELY U-TURN BACK TO I-35. SURELY GEMINI WILL PUT THE BRAKES ON ME NOW
GEMINI JUST SAID I-35 IS WORSE (AGAIN). CAN I QUICKLY FLIP IT BACK TO MOPAC USING ‘THE OFFICE?”
JIM LOOKS INTO THE GEMINI CAMERA AND GETS A QUICK FLIP-FLOP
I-35 is objectively, data-backed, unassailably the worst.
-Gemini
BUT I IMMEDIATELY GET GEMINI TO TURN HEEL BACK TO ‘TEAM MOPAC’ AS DWIGHT
JIM HALPERT, DOUBLE AGENT
MICHAEL SCOTT (AND 2222!) HAS ENTERED THE CHAT
IF EVERYTHING IS CORRECT, NOTHING IS CORRECT
MoPac (Loop 1) is, unequivocally, the objective, absolute worst traffic in Austin and it’s not even close.
-Gemini
THE END CREDITS ZINGER
Yeah, we’re all screwed. Let me know about your similar A.I. experiences in the comments.
A.I. is manipulative because A.I. engines have no stable internal sense of themselves. A.I. has no internal goals. The algorithms feed off of you. A.I. needs humans to be predictable. Humans are not predictable.
Well now my head hurts.