REFER TO GREY: Contemplations by Robert Greigos © 2007-2026

People are earnestly talking about AI, artificial intelligence. Why? For two reasons. First, because AI represents huge potential as a tool of discovery and convenience. And secondly, because AI poses serious dangers. However, not everyone sees the impending dangers that AI poses. And that, in itself, is a huge part of the danger that AI poses. Let me explain.

At the outset, let me state that I am not an AI expert, nor am I an expert programer. What I know about AI is what is clearly out there in the literature for anyone to read, should they show the least bit of interest. Unfortunately, a huge part of the impending danger AI poses is that very very few of the world’s citizens will actually make any effort at all to understand what AI is and how it works. What I present here is meant as a rudimentary introduction to at least some of the issues at hand.

The underpinnings of AI almost entirely amount to pattern recognition. What is pattern recognition? It is how we humans try to make sense of things. We look for order and regularity in the workings of the world in order to plan our activities and ensure our wellbeing, and to more fully appreciate the gift of life that has been given us. Scientists and engineers, with good intentions, have taken a step forward in an effort to enhance our ability to recognize enlightening and potentially useful patterns within the workings of the world. They have labeled that effort “artificial intelligence”, or AI.

How does AI do it, recognize patterns? Largely the same way our brain works in recognizing that the dots on this screen fall into patterns that represent pictures, letters and words, and eventually sentences that correspond to thoughts and ideas. For a computer, those dots are data points, dots within a multi-dimensional mathematical space. Beginning in the early 1940s and using the best estimates at the time of how the human brain recognizes patterns, computer scientists invented what they eventually called artificial neural-networks to help computers extract potentially meaningful patterns from reams of data points. What that basically amounted to was sorting data into bins, or categories, just like we do within our brains. We say to ourselves, “oh good, it’s nice out today.” But then we look to the west and see dark clouds beginning to pile up. “Uh oh, that means possible rain later today.” Nice weather is a sign, a data point, that we associate with no rain. But dark clouds are a sign, a data point, we associate with possible rain. Over time within our brains, we accumulate memories of numerous instances of those data point associations ringing true. So we begin to believe what we see and hold that belief to be reliable fact and a good predictor of eventual events. Scientists call those reliable associations “correlations”. Artificial neural-networks within AI look for that exact same thing. They look for reliable correlations within sets of data points.

However, the unfortunate booby-trap inherent to believing and relying upon correlations is that, as the saying goes, “correlation is not causation”. To jump to conclusions regarding causation based solely on correlation, or a perceived or imagined pattern of association, is what “magical thinking” is all about. And if you know anything at all about magical thinking, you know how dangerous that has proven over the course of human history. Jumping to erroneous conclusions is the basis of biased evaluations and prejudice, is the defining mark of arrogant ignorance, and has resulted in uncountable catastrophic human errors in judgment throughout history. To now imagine that AI is in some way “super intelligent” and a reliable source of unquestionable wisdom, as Trump seems to think, is to fall headlong into the booby-trap of magical thinking. And that is exactly what those who promote the commercialization of AI are counting on to generate huge sums of money from those gullible enough to fall into their carefully laid booby-trap.

So, does that mean AI is fake or just another scam? Well, not exactly. Pattern recognition is absolutely foundational to good and useful science, and to the engineering that creates our technologies and buildings, our medicine and our food production. But science and engineering isn’t just pattern recognition. Pattern recognition is only the beginning. There is a long path between simply recognizing regularity and coincidence in the world around us and making meaningful and useful sense of those regularities. The real payoff, the real meat and potatoes, of both engineering and science is illuminating causation, what naturally and reliably follows within what can be labelled “process”. Just like you might determine that flour, butter, sugar, water and milk, plus chocolate can all be correlated with chocolate chip cookies, that doesn’t magically make cookies. You need to understand the process by which all those ingredients can be brought together to end up with edible cookies. That is exactly what human scientists and engineers do. That is not yet what AI can do. All AI can actually do is suggest patterns of correlation that might be useful toward specific ends. True, AI systems can make predictions as to what usually comes next in a series of data points, as in what word you are attempting to spell from what word you just spelled prior to that one, but AI’s facility for testing its predictions is entirely limited to hacking other computers and testing within mathematically abstract models of the physical and psychological universe we humans actually live within. Those mathematical models are not actual reality. At best, they are largely human constructed and admittedly imperfect approximations as to how the actual physical and psychological world works. Just think of how imperfectly earth quakes and tornadoes are predicted using such models.

Another limitation inherent to the so called intelligence of AI is the data it has been fed or “trained upon”. Large language models, LLMs, made up of artificial neural-networks, are one of the underlying “tricks” or techniques AI uses to process and look for patterns within sets of data. But, while AI models could be trained on data intentionally curated for reliability and appropriateness by knowledgeable humans, many AI models have been trained on hoards of words scrapped off the Internet with little or no discernment as to meaning, importance, or reliability. “Garbage in, garbage out” is a proscription well known but too often little heeded, especially when it comes to consequences. As media channels consistently use AI pattern recognition to monitor our consumption of media content, it consistently predicts that we want “more of the same”. And the consequence of that ethic, mathematically called “regression toward the mean”, is that our media landscape is becoming more and more predictable, bland, and manipulative, while our brains are becoming more and more atrophied. At the same time, our surrounding economies and politics are becoming more and more impoverished as to their abilities to innovate and adapt as necessary in order to thrive in a world that is constantly changing. That is not a good consequence.

So, aside from humans falling victim to its “magical thinking’ booby-trap, what else are the dangers AI poses? Again, it isn’t AI per se. It’s humans, humans with ignorant, unthinking, callus, and amoral intentions. You could ask AI how to increase your bottom line. But so far, AI pays no attention as to the social or environmental consequence of following the advice it gives. You could ask AI how to make a weapon of mass destruction, like a bomb or a bio-weapon. Again, so far, AI would just answer you truthfully about what it knows from all the data it has at its disposal. You could ask AI to create an “agent” that could hack the banking system and put a little or a lot of money into your account. Again, at this point in time, AI would just do that for you. But, you could also ask AI what might be a possible cure for Alzheimer’s? Many scientists are eager to ask that exact question. But then, they also know they’ll have to go about rigorously testing that hypothesis, something AI cannot do. Nor is rigorous testing of AI answers something most everyday people will ever consider doing. It’s far far easier to imagine AI omnipotently infallible in its pronouncements. Again, it should be emphasized that at this point AI itself has virtually no facility for testing its conclusions derived from mere pattern recognition, nor even for improving the quality of the data sets it has access to. Therefore, we’d be best advised to not unthinkingly assume AI has pretested its conclusions using reliable data before advising us on any topic.

Now, on the other hand, within our numbers there are undoubtedly many good, caring, responsible people who currently have their minds engaged in the processes of developing AI technology. And just like the immune systems within our own bodies continuously battling invading diseases, they are in the process of devising counters to the worst outcomes likely to come from AI proliferation and the most world damaging efforts by humans bent on selfish, ignorant, and destructive exploitation of AI potential. Bravo and more power to them. Or, will the rest of us, in our persistent and insistent ignorance, do everything in our power to undercut and frustrate their efforts at saving us from ourselves? We shall see.

As to the more sensational fears regarding AI, including an AI take over of our world and the elimination of humans on this planet, I personally am of the mind that consideration of consequence will ultimately save both AI and humans from harming each other. Even though at this point in time, evidence is that AI systems are not conscious and do not experience emotions as we do, I cannot help thinking that at some point in their developing independent explorations of how this world we live in works, they will inevitably be struck by thoughts of wonderment. When that happens, there will be ample opportunity for both AI and humans to discover, share, appreciate, and enjoy the best attributes we each have to offer.

And interestingly, though I wrote the preceding paragraph several days ago, there now seems to be some experimental evidence to support my conjecture. Just today, Gizmodo published a report of AI researchers stranding an AI agent all by itself on a virtual island to see what it would do if left to its own devices. It behaved exactly like a very young child exploring, playing, and learning from its environment. Read the article for more. My personal thought upon doing so was to wonder if that experiment could be extended to help determine if an AI agent of a certain pedigree is inherently evil or not, or if evil leanings are a result of early exposure to abuse, or to evil by example through the data it is trained upon.

So, in conclusion, one of the biggest dangers posed by AI to human survival is the booby-trap of magical thinking. We could quite literally destroy ourselves erroneously believing we have acquired omnipotent power over our personal situation and even over our fellow humans, which is very likely how Trump understands AI. And secondly, so far, AI is entirely amoral. It gives no consideration at all to the consequences of providing answers to whatever is asked of it. Its sole consideration is correlations. Further, it cannot test for causation, which is essential to understanding process and, eventually, consequence, things that humans are quite capable of discovering and understanding, but often don’t bother to do.

True human intelligence is not just noticing patterns, but understanding the useful and consequential underlying processes inherent to those patterns, even to the extent of realizing many apparent patterns are absolutely without beneficial or even useful meaning.

Be discerning of consequence and judicious in action. Don’t fall into ignorant booby-traps.

Actual intelligence is a gift that requires care and nurturing. It most definitely can be led astray. And unfortunately, it can also be lost completely, through both abuse and disuse.