The Instinct to Question AI
My first shift working alone as a locomotive engineer 30 years ago took place on an utterly moonless night. I was out of the cab and on the ground. An unattended 30-ton railroad boxcar slowly crept by me in the pitch black, silent as death, just a foot away. Who knows where it came from?
I already knew that a crawling car made less noise than a mouse. No lights, nothing. I never heard this one. I just saw it as it passed by my elbow.
I grimaced, but was still in one piece. The car wasn’t supposed to be loose. I radioed my conductor, also on the ground, about a mile behind me, to warn him what was coming. He replied, “Thanks.” Minutes later he followed up with, “That’s scary, isn’t it?”
Although the boxcar was barely moving, a human wouldn’t have had a chance. Its 30 tons would cut a person in half between the flat-headed steel rails and the overlapping flanged steel wheels.
That car couldn’t have cared less. It would’ve rolled through me without a thought and kept creeping downhill until something else stopped it.
If an AI system wiped out humanity—or just a noticeable part of it—it wouldn’t care one bit. Like that boxcar, it has no feelings about anything. At least not yet.
AI pioneers Geoffrey Hinton and Yoshua Bengio joined hundreds of scientists last week in calling for international oversight and containment standards. Hinton has warned AI could threaten humanity. Former Vice President Kamala Harris has separately urged safeguards.
In this nascent Artificial Intelligence Age, we need citizens to think for themselves and prevent the unexpected by acting quickly. Education gives us the knowledge and helps train the instincts to question what AI is doing and act in the moment.
We will need those instincts most when an uncaring but apparently all-wise computer gives us unfamiliar facts, tells us what to think about them, and ultimately what to do. Or it might just skip us and take action on its own.
Humans design AI. We build it, install it, program it, and set it in motion. We are responsible for the results, whether the system follows our instructions, makes an error, or ignores us. When AI inevitably runs amok, the chickens will come home to roost on us -- if we are around to take the blame.
I’ve caught AI mistakes. The warning has always been instinctual: “That just doesn’t seem right.”
Sometimes I had facts in my head that exposed the error. At other times, it was just a gut reaction. I then asked AI to point me to a source and double-checked it myself.
Those instincts came from 70 years of reading and education, formal and informal. All of it gave me more than facts; it taught me how to use them, when to doubt them, and when to follow up.
When one system gives a troubling answer and another confirms it, contradicts it, or shrugs its digital shoulders, pleading ignorance, no machine merits unsupervised reliance.
AI doesn’t need to be smarter or more ambitious than us to be a menace. It need only be intimidating or more persuasive. We already risk surrendering our judgment because AI is fast, articulate, and seems to relieve us of responsibility for our decisions. Or so we may think.
Competing AI systems might catch one another’s errors. That competition doesn’t guarantee truth. We will have to use well-developed instincts to sense and investigate what seems wrong. Humans must stay deeply involved at every stage and act quickly when necessary.
By human monitors, I mean broadly educated, trained professionals whose full-time job is to monitor AI systems while they operate. They must have the authority and responsibility to intervene when a system’s behavior raises an alarm.
We bear final responsibility for AI. Otherwise, we surrender our freedom to machinery we cannot hold accountable.
Andrew Warren is a (re)tired criminal defense lawyer and erstwhile college professor. Despite his better judgement, he writes carefully crafted opinion pieces for the Silver City Daily Press and the Deming Headlight every other week. He has a Master's Degree in American History and Government and teaches such classes for the Western Institute of Life Long Learning, also in Silver City. Given the current state of discourse, he prefers reading over political dialogue. He can be reached at alwarrenesq@msn.com.