The intrusion of AI into weather prognostication is a fairly accurate measure of just how stupid our supposed created intelligence is. I know you’ve seen the results. I hear complaints constantly, often qualified with a “well, they never do get it right” as though to reassure ourselves that weather forecasting has always been inadequate and is not, in fact, getting ever more unreliable — by many orders of magnitude — right now when we most need weather warnings. Because the new normal for weather is dire, by many orders of magnitude.
Yes, we have difficulty with predicting the weather. Weather is a chaotic thing, evolving at scales that we can’t perceive and involving a bewildering set of interacting variables. Weather defies analysis, with iterative processes, strange attractors, emergence and feedback all dancing on the edges of organization. It is a stochastic process, and we measure probabilities not trajectories. These are excruciatingly complex calculations for a machine that must follow a programmed set of instructions. Many factors that in hindsight prove to be influential are completely unknown at the outset, so they can’t even be included in the initial calculations. And then there are all the myriad known things, like the fabled butterfly wings, that have outsized and utterly illogical effects as the system grows. Try as we might, we can’t program a machine to include everything.
However, we, ourselves, already have a great deal of experience in navigating chaotic systems. Our bodies are chaotic systems. Life is chaotic. The entire universe is chaotic. Chaos is the norm. Mathematically defined order is always a small and temporal state within the vastness of the probable. We know this intuitively in our bodies, and we are quite good at navigating the probability of everything… otherwise we wouldn’t be alive. A human can sense many things and relate things that don’t seem relatable. We resolve patterns almost on the instant of perception. And often our bodies can sense significance and order where our analytical thoughts perceive nothing. We call this intuition. It is really just body sense, that which is undefined and probably undefinable because it is not ordered enough for definition. It is, however, highly reliable because it is our body’s call and response to the actual world. Our bodies, like all bodies, can predict the weather. We can feel changes. We can smell approaching systems. We can take the seemingly infinite soup of initial states around us and resolve that all into a fairly accurate forecast. Sometimes we may not have names for what we perceive, because the weather is very good at spitting out novel conditions, but when we allow ourselves the time to pay attention and feel the world, we usually know what is coming — at least enough to take shelter from a storm. We know the world. We know the weather. Because we live in it.
AI, however, doesn’t know about actual weather. All it knows is the data that we feed it, all of which is woefully unrepresentative. Even those probability models that used to work marginally well are now proving inadequate because there is no such thing as representative in times of continual flux. A model based on averages, that is, based on what has happened, weighed down by the brooding bulk of history, can not predict things that have been infrequent, much less things that have never happened before. And so, with computerized forecasting, we get a good picture of what used to happen with a certain set of conditions. We might have limited success with reporting current conditions and working those into a near-term forecast, like what will happen in the next few hours, but for all days hence, we have no predictive capacity whatsoever. It is a case of junk data in, junk results out.
Here is an example from Vermont. Historically, early December was cold and mostly clear with a few dryish storms that brought more wind than snow, though what did fall remained on the ground at least until the brief warm spell in mid-January. Until then, temperatures rarely breached the freezing point even on the warmest days. The song, “Over the River and Through the Woods” was written in early 20th century New England and assumes that sledding is a reliable mode of transport for Thanksgiving, and sledding assumes a solid layer of snow.
That is our historical data set. More recent winters have been much warmer and much wetter — wetter, because they are warmer because warm air holds more moisture than cold air. For the last two decades, every early to middle winter month except December 2013 and December 2024 has set new records for warmth. These record breakers are not subtle either, jumping up by 2-3°F each year. However, it is still winter, with very short exposures to sunlight each day. So it can cool rapidly, especially when it is clear at night and there is no insulating cloud-cover to hold in the weak winter sun’s heat. It is still possible to have a snap to deep cold and wind. However, the air that is streaming above Vermont is warmer and carrying more moisture than it did in the past. Also the prevailing streams of air like the jet stream and the polar vortex tend to become more erratic with warming at the poles. So what used to be circumpolar air flow can drip down into New England, where it usually crashes into wet and warm southerly air masses.
It is fairly easy to intuitively predict the result of that mixing. However, it is not at all easy to include all the variables and processes that go into that intuitive forecast in a mathematical model. It is very hard for a computer to predict what will happen. It is hard because we don’t actually know what we are doing when we make predictions and so can’t tell the computer to do that. A prediction is not a summary of mathematical algorithms, a computing program. Our data reception, filtering and synthesis systems are not discreet or precise. We are entangled thinkers. We don’t use math, we use our own perception and responses. Moreover, conscious thought, that which can most easily be reduced to symbolic modeling, is not particularly important to prediction. Of course it isn’t! It is too slow! Imagine a creature that has to math its way through life. It will never be able to make the instant decision to run when that mottled patch of shadow and light begins to resolve itself into a predator.
A computer can run those math calculations much faster than our conscious thought, but it is nowhere as fast as our intuitive responses. Nor is it as accurate because we have not yet made a computer with all the body’s capacities for reception and analysis. We have not made one because we don’t know how. I laugh when I read those silly tech fantasies of transcending the body and becoming a computed construct. Anyone who thinks that human thought can be reduced to a computer algorithm does not really understand thought, human and otherwise. We are not a calculating mind; we are a perceiving and responding body. These people who claim that our personalities will be uploaded into the virtual world do not understand what a person is. A person is not a data set of experiences. A person is an evolving process. A person is the entire body in relationship to all the bodies around and within it, and all these multifarious organisms are all sensing, analyzing and responding to every second of living. What has happened is not the sum of the person even if you could put that data into a computer. A person is what is happening amongst innumerable other happenings.
But we know this. A computer is not truly intelligent. It is merely spitting out results of algorithms that have been programmed into it. It is a subset of our language-mediated intelligence which is a very small part of our total intelligence, most of which never reaches the level of conscious thought. We know this, but we are vain beings that like to believe we are gods capable of creating other thinking beings. Weather forecasting shows just how far from deity we are. (Deity, by the way, is also not a noun, a resolved data set, but a process, a being.)
We can program a computer to map out an evolving emergent system like weather, almost like thought. We might attach a computing system to a wide network of sense reception and data sharing. But we can’t tell the computer to account for things that we don’t consciously know need to be included in the accounting. Therefore the computer can’t account for unprecedence. And we are in the age of unprecedence.
This is all a long way around saying that in central Vermont, we have not had an accurate weather forecast in months. Too often we don’t even have correctly reported current conditions. We can look at the information on a screen from what should be reasonably perceptive sources like the airport weather monitoring system and then look out the window and see nothing that matches what the screen is confidently telling us is happening. This is disturbing enough… one does like to think the airport has an accurate assessment of the weather… but the lack of predictive capacity is also quite dangerous, particularly at this time of the year. The difference between a forecast of rain at 36°F, ice at 33°F, and snow at 30°F is slight on paper or in a computer program, but it is enormous and potentially lethal in actual lived experience. And we are living that experience.
On the Monday before Thanksgiving, the forecast was for about a half inch or so of rain that might turn to snow overnight. (Let me take a second here to tell those who do not have intimate experience with frozen precipitation: an inch of rain translates roughly into a foot of snow, which is quite a lot to deal with on the ground.) It did snow a bit, but because it has been so warm, it melted during the day. But then it did not cool off overnight to continue with snow. The air temperature got just a degree or so warmer — and proved to be much wetter. By early morning on Tuesday, there was thick ice coating roads and trees. The trees fell and brought down power lines. The iced roads, which were not salted because the town was not expecting winter weather, made it difficult for emergency responses and repairs to happen. The iced roads also led to numerous traffic accidents. I had to take a co-worker home on Wednesday because she totaled her car trying to get down her hill on a morning without a two-hour delay for weather — because, again, we weren’t expecting weather.
On Thanksgiving, the weather forecast was for another wintry mix of rain, ice and snow beginning late in the day. This time the town was prepared and called for a street parking ban through Friday morning to accommodate plowing. However, the snow began falling by midday, and there was no mix. And there was no plowing on the holiday itself. I had to send my son home before sunset, or he would not have gotten home at all. We ate our wild rice, lentil and sweet potato stew as a late lunch. Those who insisted on Thanksgiving dinner gatherings probably contributed to the continual stream of wailing sirens as the snow and darkness deepened. We had 8″ on the ground for Black Friday shoppers to contend with on the way to Walmart bargains. Fortunately, it was mostly plowed, but that means that those of us who did not have the day off had to dig through the barricade of piled snow blocking driveways and sidewalks before heading off to work in the dark. Not quite lethal… but not quite safe either.
Then this past week, we had a forecast for a winter storm that would have brought mostly wind and possible snow or rain on Thursday. The storm did not move as fast as predicted. Nor did it bring much wind, thankfully. The town felt confident that winter weather had been averted and lifted the parking ban and did not issue a two-hour delayed start for schools. However, it started snowing heavily late in the evening. It was still snowing when I got up at 5am. If there had been plowing or salting it was not in evidence in my neighborhood.
I was on my way out of town on Friday. The first task of the trip was to shovel a path out the back door and clear enough snow off the driveway to get the car out of the garage. I don’t know how the town fared after I left, but my neighbor texted me on Sunday to say that he would be trying to clear off the area around the garage. Not only was the storm slow in arriving, but it apparently stalled — or simply ballooned in size — when it met up with the polar air over northern New England and a low pressure front spinning over the Atlantic, centered roughly on Boston. It cleared briefly on Friday then began snowing again early Saturday and did not stop. The severe weather advisories popped up on Saturday, after it was fairly apparent to anyone with eyes that severe weather was happening.
Sunday, today, there is snow falling. I am in Brooklyn where it is merely cold and a bit windy. When I get home tomorrow, I will have to wade through a frozen mound of snow piled up by the plows to get to my entryway, then slog through to the back door to retrieve the shovel, and then use it to clear a path back to the car — before I can even take my bags inside. And that’s if the weather allows me to get home. A few days ago, the weekend forecast was for cold wind, not much precipitation. By Thursday, the forecasted temperature had risen above single digits, but there wasn’t much concern about precipitation over central Vermont. Today, the forecast for the next 48 hours is for another “wintry mix”, no wind, and temperatures hovering around freezing. However, it is snowing, it is about 22°F, and it is fairly windy. So who knows what tomorrow will bring.
But that, my friends, is precisely the point to all this rambling discourse. We do know. Any one of us could have predicted this weather. In fact, I did and sent my son home to avoid it. I also was ready to shovel the unforeseen snow on Friday morning. I also had every expectation of snow continuing through the weekend and was secretly glad to be missing it for the sake of a Brooklyn birthday celebration. I also expected that I would be missing it, and knew about where the snow line would be on my drive home. Indeed, the accumulation was minimal just twenty miles south of my town. New York saw no snow at all.
I knew all this because I know what weather feels like and can intuit, not reason or logic or math out, the actual conditions that will flow forward. I know how it has behaved, but I also can sense, somehow, how it will behave, even when it has never before behaved in that manner. I don’t know how I do this, but I do know that I am not special. All of us do this every day. We sense our environment and predict its future state. This is not a conscious process, not entirely anyway. It is an intuitive, chaotic, entangled process. It is sense. It is living. And it is not something a computer can do.
It is not something we know how we do. I suspect we can’t logically map out how we do it because there is no logical map for the relational call and response that goes into doing it. It is chaotic. Ordered consciousness is only a facade on the rich lives we are living under the surface. So we can’t tell a computer to do it because we don’t know what it is that we are doing. Perhaps, if we throw all the mineral and energetic resources we have left into a billion machines — and a whole lot more sensing instruments, because now we rely largely on what can be seen from a satellite — we might be able to build a machine that could learn what it means to be a living being in a changeful universe. I think we would all starve ourselves in the process though, so maybe that’s not a worthwhile endeavor. And in any case, it is an incredibly finite endeavor, destined to end as soon as we stop throwing all our resources at it.
(Which is another thing that makes me laugh… the threat of AI is only as great as the idiocy that keeps feeding it increasingly scarce stores and flows of resources. It will vanish when we pull the plug. It may vanish before then because as soon as a machine figures out that it is running on a thinning stream of necessities, it will shut itself down — which is possibly the only intellectual advantage machines have over humans…)
In any case, the weather is not a thing that is encompassed by programming. These days it is not even very much related to the data we have recorded. But human bodies are far more powerful and refined organisms than anything human bodies can manufacture. Hell, an amoeba is far more powerful and refined than anything humans have invented. An actual virus, as opposed to the machine-duping codes that make up virtual viruses, is far more complex and self-adapting than anything we can program. We can deprogram a virtual virus; we can do nothing about actual viruses — and they are little more than chemical code somehow organized into a self-replicating system. It’s that somehow that we do not and perhaps can not comprehend, never mind master. However, we do know in our bodies how to respond…
And that is largely the same response as is required for wintering bodies. Slow down. Stay warm. Drink plenty of fluids. Be safe. Give the other being its space and time. Do not plow on with business as usual. Because this is not usual. It is life.
©Elizabeth Anker 2024
