Are Experts Training AI Models To Replace Themselves?

  

About a year ago, LinkedIn job offers of a new type began showing up in my inbox.  I am not personally on the job market—in fact, I went to half-time at my current teaching job over a year ago anticipating a possible future retirement—but LinkedIn didn't know that.  What it did know was that I had an engineering and physics background.  That apparently made me qualified to work with new AI models in developing their analytical skills in those areas. 

 

This was hourly work, and while the pay was attractive, the work was not, so I didn't pursue it.  But lots of people are doing that type of job, and many of them are highly qualified specialists in fields as diverse as medicine, law, and the various sciences. 

 

I don't blame the AI companies for hiring experts to train their models.  Where else are you going to get an interactive guide who lends truly human judgment to a task?  Nowhere, at least for the present, except from a human.  But in a brief Internet search, I turned up two essays that expressed basically the same concern:  that in training, correcting, and guiding AI models to do your work better, you are transferring knowledge that AI may one day use against you.

 

The full-time training of an AI model should be distinguished from simply using it in a guided way to accomplish a given task.  Most users do the latter and couldn't care less if the AI model walks away (metaphorically speaking) knowing a little better how to do the task, especially if you're going to ask it to do something similar later.  But as both James Maisiri and Mohammed Faheem point out, that knowledge was formerly something you owned and controlled.  Once the AI system acquires it, you've lost control of it, and it belongs to the model or the model's corporate owners. 

 

Faheem is especially concerned that a certain type of employee is in danger of being replaced by the very AI model that they train.  Those who do routinized work that relies on surface-level knowledge without being concerned about judgment calls or the wider context of the task are the ones most liable to be replaced or superseded by AI systems. 

 

Maisiri, who is from South Africa, was interviewed after he obtained a Ph. D. in industrial sociology for the task of transferring his educational expertise to an AI model.  Ironically or not, the entire interview—45 minutes—was conducted between Maisiri and an AI model.  When it asked him to retake a part of the assessment it administered to him, he'd had enough, and that ended the interview. 

 

He confesses that his refusal to pursue the job wasn't based so much on a principled notion that he was contributing to the erosion of his profession, as much as it was simply a nameless discomfort with the process.  Facing a machine with the power to hire me is something I've never experienced personally, but I can sympathize with his discomfort.

 

Hiring isn't the only thing being turned over to AI.  In San Francisco, an "experimental" retail store operated by a research lab put an agentic AI system in charge of the store, including hiring and firing power.  And back in August, after what most human managers would think was an excessive delay, it fired an employee for excessive lateness. 

 

Judgment and people skills are some of the last abilities that AI models will be able to implement successfully.  But the systems are already working well enough that hundreds of companies trust them to be the voice of their firm to incoming calls, however frustrating that may be to some people. 

 

Interfacing with customers and dealing with personnel problems at a retail store are not exactly rocket science, but they are a lot more commonly needed than rocket science, and therefore more likely to play a major role in the future of AI in our cultures.  So should we be worried that pretty soon there will be a lot of unemployed rocket scientists, once a few rocket experts unload everything they know to an AI system?

 

That is the kind of question that can only be answered definitively by time.  It's hard to imagine us running completely out of rocket scientists, and letting AI agents build and fly rockets that work flawlessly the first time and every time.  But not that many years ago, it was a pipe dream to imagine that factories would turn out dish-size integrated circuits with billions of flawless transistors in each one, and have them work perfectly.  Now it's being done hundreds of times every day.  That was achieved not just by AI, but by decades of effort of thousands of people as well.

 

The mysterious cooperative being called human society can preserve and extend complicated technological infrastructure indefinitely.  If a region as large as, say, Texas, was entirely deprived of electric power for a period of a month or more, it would become mostly uninhabitable, and people trying to survive there would face major challenges.  The only disaster of that magnitude we have seen in this country was when Hurricane Katrina destroyed large parts of New Orleans and its environs for a time in 2005. 

 

Say we revolutionize the medical industry so that nearly every medical interaction is done by AI-powered robots.  We would then be vitally dependent on those robotic systems for our very survival, at least those of us in need of medical care, which is nearly everybody sooner or later.  If some disaster were to hit the robotized medical industry, we would indeed be in deep sneakers.  But if the systems were designed with fail-safe approaches in mind, that eventuality might be as unlikely as losing electric power over large swathes of the U. S.  It is theoretically possible, but practically unlikely.

 

Nobody is seriously suggesting a boycott on expert training of AI systems.  But the rapid progress AI firms are making in training and fielding agents that take on increasingly sophisticated and responsible tasks means that we ought to start thinking about how to deal with situations that Maisiri and Faheem are worried about:  when human experts are replaced by AI ones.

 

Sources:  James Maisiri's article "I refused to train the AI that could replace me" appeared at https://restofworld.org/2026/ai-training-jobs-expert-replacement/.  Muhammed Faheen's article "Why Most People Are Training the AI That Will Replace Them" appeared at https://medium.com/write-a-catalyst/why-most-people-are-training-the-ai-that-will-replace-them-e55fc113fa3e.  I also referred to the Wikipedia article on Hurricane Katrina.

Army Admits Failures in Deadly Flight 5342 Crash

In a report completed last May but released only last week, the U. S. Army's investigation into the deadly January 2025 crash between a U. S. Army Black Hawk helicopter and American Airlines flight 5342 shows that numerous critical failures in hardware and procedures led to the crash.  This report complements the already-completed National Transportation Safety Board (NTSB) study of the crash, but shows that similar problems may exist at other airports near Army helicopter routes.  An Associated Press article on the report appeared this week in numerous journals, and I am relying on it for the information below.

 

The crash, which occurred at night, involved a prescribed flight path for the helicopter which was only 75 feet below the nominal flight path of the approach to one of the runways of the Ronald Reagan Washington National Airport.  Strong winds caused the airliner to deviate slightly from its nominal path.  The helicopter's altimeter error contributed to the fact that the helicopter was 100 feet higher than its altimeter indicated.  A radio transponder that would have given flight controllers more precise information about the helicopter's location, a device called an ADS-B Out system, was not working.  The flight controllers were relying on the airliner's pilots to maintain visual separation in an environment full of light sources that made it hard to distinguish the helicopter's lights.  And the helicopter pilots were wearing night-vision goggles which restricted their field of view.

 

In a separate statement, the NTSB says the Army has addressed seven of the NTSB's eight recommendations deriving from the accident, but the Federal Aviation Administration (FAA) has acted on only about a third of the recommendations the NTSB made to that agency.  One of the most critical recommendations involves requiring operational ADS-B units on military helicopters and other types of rotary and fixed-wing aircraft.  There are several bills pending in Congress to implement this change, but so far none have been passed, partly out of concerns that airports will use the signals to charge user fees to general aviation, although that issue is specifically addressed in some of the pending bills.

 

With such a laundry list of errors, malfunctions, bad policies, and lapses in training, it's a minor miracle that an accident like this didn't happen sooner.  Unfortunately, 67 people had to die before enough political and organizational will appeared to do something about the problems. 

 

It would be better, of course, if someone had come along before the crash and said, "Hey, that's an awfully narrow margin of safety between the landing flight path to the airport and the helicopter flight path. We ought to do something about that."  But even if such a responsible and forethoughtful person had spoken up, he or she would have faced immense bureaucratic inertia.  All the bureaucrats could have responded, "Well, we've been doing it this way for years and nothing bad has happened so far."  That isn't quite true, by the way.  For every fatal accident in the air, there are several near misses which pilots are supposed to report to the FAA.  And I suppose there's a bureaucrat charged with the responsibility of reading those reports and making recommendations to avoid such near-tragedies in the future.

 

But nothing motivates people like death, even if it's not their own.  So it took the January 2025 accident to get a number of improvements made on the part of the Army, whose organization is designed to get things done, and largely still works despite some erratic behavior at the top recently.

 

The FAA is another kettle of fish.  A creature of Congress, it can only do what it's told with the money it's given, and Congress has not been generous or even particularly attentive during the last decade or two.  Consequently, a number of technical improvements have fallen woefully behind schedule, and calls have arisen to change the FAA's structure to a privately-funded model paid for by user fees, and cut it loose from the federal government altogether, as the analogous organization in Canada has done. 

 

Such radical changes would require foresight, planning, and responsible management of actions that could affect millions of people.  All these things have become rather rare in Congress, although not entirely extinguished.  The partisan paralysis afflicting that institution has not shut down the government entirely, except for those times when certain leaders think they need to do a grandstanding show of force and force a partial shutdown to make a political point.  Such actions would have been unthinkable to the congressmen and women of thirty or more years ago, who all look like statesmen compared to the typical occupant of a House or Senate seat today. 

 

Even such a simple thing as increasing the number of ADS-B devices to be required in various modes of aviation has proved to be a stumbling block, as the number of bills floating around show.  In the old horse-trading days, Democratic and Republican legislators would meet in a back room over cigars and bourbon and hammer out a compromise that would not make anybody completely happy, but would move things forward by giving both sides what they want, often in completely unrelated areas.  If Joe wanted a bridge built in a city he represented, he might agree to support Bill's bill to improve airline safety—things like that.

 

Maybe a little bit of that still goes on today, but my overall impression of Congress today is that of rival religions, almost, who won't even stoop to socializing with their opposing members.  The factors that have led us to this pass are too numerous and complex to go into here, but the abject failure of Congress to live up to its constitutional responsibilities lies at the heart of many nationwide problems today, from failures in air-traffic control to failures in restraining or impeaching people who richly deserved to be impeached.

 

A do-nothing Congress may seem pretty far removed from a tragic air-collision accident, but the connection is there.  And it's one more reason to vote in the coming weeks with an eye to realizing the hope that Congress will once again represent the will of the people in an honorable and admirable way.

 

Sources:  The AP story "Army finds systemic failures behind DC midair collision and calls for safety fixes" by Josh Funk was carried by many outlets, including the Alabama Daily News at https://aldailynews.com/army-finds-systemic-failures-behind-dc-midair-collision-and-calls-for-safety-fixes/.  I also referred to the Wikipedia article "2025 Potomac River mid-air collision."

 

When Engineering Becomes Culture: The Texas Transportation Museum

  

Every field needs a foundational philosophy, and I've never found a better one for ethics than St. Thomas Aquinas's fundamental principle of moral philosophy.  While I can't cite chapter and verse for an exact quote, it goes something like this:  "The good ought to be done and evil avoided." 

 

Sounds kind of obvious when you put it into words.  And of course, it raises the question of what is good and what is evil.  But citing that principle allows me to place the Texas Transportation Museum under the wide umbrella of engineering ethics.  In what follows, I'll try to answer the question, "Is the Texas Transportation Museum a good thing?"

 

I was unaware of its existence until a few weeks ago when I was looking for a place to take a day trip and discovered its website.  It's located just north of the San Antonio International Airport on a few dozen acres of ground donated in 1963 by the City of San Antonio to a collection of railroad enthusiasts headed by a consulting engineer named David Wallace and his friends.  The organization celebrated its 60th anniversary in 2024, and is still going as strong as an all-volunteer outfit can go.

 

Its roots as a railroad museum are obvious, as it houses numerous model-train exhibits as well as all kinds of historical memorabilia having to do with railroads.  On display outside are some full-scale artifacts from the golden years of passenger rail: a Pullman sleeping car from the 1920s, a caboose, and a steam locomotive—minus some of its hardware, true, but retaining most of the major components.  Other forms of transportation receive a nod.  In their large interior display, they have a Model A Ford, an antique fire truck, a horse-drawn steam-powered fire engine, and two or three other cars from the 1910s and 1920s.  But it's clear from the displays and the interest shown by the volunteer staff that the core of the museum is rail.

 

Maybe it's a peculiarity of American culture that a mode of transportation that for people, anyway, is largely in the past still has such an active following among primarily amateur historians.  And artists, too.  If I had to categorize the hobby of model train construction and use, I'd have to say it's art.  Like other art, it exists simply to be admired.  A friend of mine has practiced this art for decades, and although he just turned 80, he still hauls his three-by-four-foot modules of N-scale tracks and engines and boxcars around to train shows and sets them up with other fans to form convention-center-size layouts.  We don't see painters joining together on selected weekends to mount their paintings together to form one giant mural on display for a couple of days, but that's sort of what model railroaders do. 

 

And what is portrayed in this type of art?  In a word, the past.  At the Texas Transportation Museum, it's a wide-ranging past, going from the Civil War days of the hijacked wartime locomotive "The General" (the story of which a short description is on display, though it doesn't mention the Buster Keaton silent film that dramatized it) to the modern era of Amtrak.  Among the volunteers on duty were a woman in her sixties who took our money at the ticket desk, while wearing hair dyed orange, green, and blue for Halloween.  (Scattered everywhere were a lot of Halloween-themed decorations:  skeletons on tractors, dismembered limbs in Pullman fold-down beds, and other representations of our modern tacky enthusiasm for cheap horror.)  Her son, who politely held the door for us and told us when the free train ride would be, had a slight speech impediment but impeccable manners.  There were volunteers as young as a couple of men in their twenties, and as old as me, and anywhere in between. 

 

One gentleman I spent some time talking with was operating a G-scale layout that was reached by walking through an archway leading to a large garden.  G-scale, as he explained, stands for "garden."  The equipment is about 1/32 of full scale, which means the track rails are about two inches apart, and an engine is a little smaller than a small loaf of bread.  He talked about the problems of running an electric train outdoors.  The tree sap falls on the rails and interrupts the current, so every now and then he has to send around a cleaning car that has a vibrating brush that scrubs the rails.  Only the cleaning car was busted for a while and nobody knew it, so they had to fix it.  Deer and squirrels cause problems too, and there's always the wheels on the engine that need periodic cleaning.  I told him about my friend who is always battling some issue or other with his layout, and the man said, "Yes, it's never finished." 

 

But when everything runs smoothly, a model train layout gives one the sensation of God-like presence in a celestial realm above the workaday small-scale reality of the display.  And to the extent one can control the train movements through modern digital technology, one has God-like powers of transforming one's will into action.  Above all, the model railroader knows that everything in the layout is there because he put it there.  (I've seen a few lady model railroaders, but they are uncommon—it seems to be almost exclusively a male hobby.) 

 

So does the Texas Transportation Museum fulfill the injunction to do the good and avoid the evil?  I'll let the final words be those of the man who spoke briefly about the museum at the end of the few-hundred-yard excursion we took on an open passenger carriage towed by a genuine Army-surplus diesel locomotive.  He said the goal of the museum was to preserve the memory of railroads and other transportation so that the younger generation could see how things were back then, and keep the tradition going.  He also hoped people would come back for their Halloween show that they would be running the next few weekends from 5 to 9 PM, and not forget the Christmas show either.  His emotions were clearly more powerful than his words, but he thinks the Museum is a good thing, and he hopes it will continue.  Judging by the enthusiasm of a cowboy-hatted eight-year-old whose parents drove down from Dallas just so he could see the Texas Transportation Museum, I think his hope may be fulfilled. 

 

Sources:  The Texas Transportation Museum's website is https://www.txtransportationmuseum.org/, and it is open most weekends except for major holidays, weather permitting.  General admission is currently $10 a person, but donations are encouraged. 

On AI and Personhood: Pagans Will Be Pagans

  

Wesley J. Smith, a senior fellow at the Discovery Institute's Center on Human Exceptionalism, did a thorough job the other day of taking down the pronouncements of an Anthropic expert tasked with understanding why Anthropic AI systems like Claude behave the way they do.  A New York Times report on a day-long seminar given by Christopher Olah inspired Smith to write a long list of things Claude can't do, in contrast to Olah, who spent all day making the case that Claude should be considered as a person, not just a machine.

 

Olah described tracking Claude's "emotional vectors" and showed transcripts of conversations with one AI model which appeared to be having a nervous breakdown.  He speculated that just as humans sometimes benefit from the act of confession, maybe Claude might find the Catholic sacrament useful as well.  A rabbi who was present said that Olah and his colleagues appeared to believe that Claude models have the moral standing of a human being, and should be accorded the same rights and respect. If that is the case, the rabbi concluded, then what Anthropic is doing with their models amounts to slavery. 

 

Olah's response to this accusation was not recorded. But it doesn't matter, because Olah and his fellow AI creators are following a well-worn path that stretches back to antiquity:  the path of pagans worshipping the god they created. 

 

Smith is exactly right when he pointed out in his column that Claude is not alive, has no intrinsic value, is not conscious, does not think, cannot "feel," cannot suffer, can be damaged but not harmed, and cannot sin.  But if he thinks that well-thought-out evidence-based philosophical assertions of the non-personal status of Claude and its ilk are going to cause scales to fall from the AI community's eyes and make them say, "We see now!  It's just a machine!" he better have another think coming.  Pagans who invest a lot of work in their creations aren't easily dissuaded by some philosophical musings.  Sometimes the only thing that will convince them is power:  a power greater than their god.

 

In the Old Testament book of 1 Samuel, Chapter 5 recounts how the pagan Philistines captured the Hebrews' Ark of the Covenant and set it in their temple devoted to the pagan god Dagon.  The first night, the statue of Dagon fell over face down in front of the Ark.  After the Philistines set him up again, the next morning they found him face downward again, this time with his head and hands cut off.  After that, they decided that maybe the Ark wasn't such a good interior decoration after all, and passed it around to their friends the Ashdodites and Ekronites, to whom a plague of death and tumors happened.  While these incidents probably didn't lead to any conversions to Judaism, it did teach the pagans a lesson that the God of Israel was not to be messed with.

 

The Jewish prophet Isaiah had some choice words to say about a man who cuts down a tree and burns some of it to bake bread.  Then he turns around and uses the rest of the wood to make himself an idol:  "Half of it he burns in the fire . . . 'Aha, I am warm, I have seen the fire!'  And the rest of it he makes into a god, his idol; and falls down to it and worships it; he prays to it and says, 'Deliver me, for thou art my god!'"

 

We are living at a time when old philosophical arguments and ancient pagan ways are emerging into the headlines and the living rooms of millions.  When you ask your Siri device to play jazz, what are you talking to?  The line between use and worship is not always easy to distinguish.  Everyone has the urge to worship something, and it's only natural that people who spend their working lives developing cutting-edge AI models are going to be impressed by what the work of their hands can do.  The urge to anthropomorphize AI, to spend time and emotional investment in it like it was another human being, and even to argue for its legal rights, are all understandable. 

But like all pagan worship of false gods, such moves end sooner or later in disaster, both personal and institutional.

 

I have a friend—a real flesh-and-blood person—who will listen to a news report of some complicated jam that such people have gotten themselves into, and quite often the first thing he'll say is, "Those folks need Jesus."  And he's right. 

 

We can philosophize all day and quote every philosopher from Aristotle to Roger Scruton in trying to convince AI worshippers of the fact that machines aren't people, and all the "can'ts" Wesley Smith lists are true.  But most of the AI crowd would first have to take two semesters of philosophy before they could understand what we are talking about, and even then they'd probably find some way to disagree. 

 

The followers of the Hebrew God knew what was going on with the Ark and the pagans.  Followers of Jesus know what is going on with AI worshippers.  Our first task is not to fall into the trap of idol worship ourselves.  A Martian observing a random sample of one hundred U. S. citizens on a typical day would conclude that the little oblong objects they spend so much time staring at must be something either religious or necessary to survival, or both.  Christians should be different.

 

And the next task is to live lives and do acts that convince nonbelievers of the power of God.  That's not much in evidence today, I have to admit, though there are exceptions, especially in cultures that are not saturated with modernity.  Christians don't have to absolutely reject AI and all its works, which is in any case practically impossible.  (I used Google's AI feature to look up the Old Testament references for this blog.)  But we can resist making AI an idol ourselves, and live as lights in a dark world that warms itself with half its electricity and creates AI idols with the other half.  Some things never change. 

 

Sources:  Wesley J. Smith's article "AI 'Claude' Can Never Be a Person" appeared on the National Review website at https://www.nationalreview.com/corner/ai-claude-can-never-be-a-person/.  The story of the Philistines and the Ark appears in 1 Samuel 5:1-5.  The selection from Isaiah is Isaiah 44:15-17 (Revised Standard Version). 

Judge Says All Texas Prisons to be Air-Conditioned by 2029

  

Last Tuesday, Sept. 22, U. S. District Judge Robert Pitman ruled that the Texas state prison system must install air conditioning for every last one of its prison beds no later than Dec. 31, 2029.  According to an Associated Press report, Texas prison administrators have vowed to appeal this ruling, so it is far from a done deal.  But this event is a landmark in the long and sweaty road toward providing prisoners something that has become to other Texans almost as necessary as food and water in the summertime.

 

Texas and air conditioning go back a long way.  The first church building to be air-conditioned west of the Mississippi, and possibly in the entire U. S., was the Lutcher Memorial First Presbyterian Church of Orange, a town in the extreme southeast corner of the state.  Oil was discovered nearby in 1913, and the newly prosperous residents paid for not only a state-of-the-art air-conditioning system to be installed for the relatively small 150-member congregation, but also for a separate building to house the diesel engines that ran the ammonia compressors.  You see, Orange did not yet have an electric utility, but that didn't stop the Orange Presbyterians.  And the humid pollen-laden summer streets of Houston saw another air-conditioning first in 1930 when a wealthy businessman suffering from hay fever paid to get a custom-installed air conditioner in his Cadillac, apparently the first passenger vehicle so provided in the U. S.

 

So with such a glorious history of air conditioning, why has Texas been so far behind when it comes to air-conditioning prisons?  The answer depends on who you ask.

 

The prison administrators themselves say it will cost too much.  They have been pegging away at air-conditioning one unit here and another unit there for many years, but as of the end of 2026, they estimate that only 60,000 prison beds will be in air-conditioned spaces.  That's less than half of the approximately 143,000 people housed in Texas prisons today.  In the system's defense, this does represent progress compared to eight years ago, when in 2018 only 35,000 beds were air-conditioned. 

 

But people are still dying of the heat inside Texas prisons—three in 2023 alone.  A prison sentence should not turn into a death-by-heatstroke sentence.  And besides the avoidable human suffering that old un-air-conditioned prisons cause, the prison administrators have admitted that the lack of air conditioning in prisoner spaces has made it harder to recruit guards and other staff.  Working in a prison is no picnic, and if you add to it the fact that you're likely to be dealing with 100-degree-plus airless environments as well, I'm surprised they can find anybody to work under such conditions no matter how well they're paid. 

 

There is plenty of blame to go around in this situation.  During the 2023 legislative cycle, the state had a budget surplus of $33 billion.  Out of that surplus, the Texas legislature in its wisdom decided to allocate a measly $85 million to prison air conditioning.  While that did help, it fell far short of air-conditioning every Texas prison, which the prison system itself estimates will cost on the order of $1.5 billion.  This situation reminds me of the engineering adage:  "If there's not enough time (or money) to do it right, there's always enough time to do it over." 

 

The lawsuit leading to the recent ruling was filed initially by Bernie Tiede, a former mortician whose life story (and life sentence for murder) inspired the wickedly funny Richard Linklater film "Bernie."  Other prisoner-rights groups joined in the suit, and there was rejoicing among these and other prisoner advocates when the news of the decision reached them.

 

Now it is true that once upon a time, no prison was air-conditioned anywhere.  And once upon a time, say around 1910 before air conditioning was a thing in Texas, there were only 3.9 million people in the state, not 32 million.  And those few rugged souls were used to living in hot houses, going to work in hot stores and offices, and just sucking it up. 

 

Virtually no one in Texas lives that way today, including members of the criminal class.  If there were some weird sociological phenomenon that prevented anyone with a present or future criminal record from living with air conditioning outside prisons, it would be reasonable not to air-condition Texas prisons, much as Muslims are provided with copies of the Quran—it's what they're used to having, or not having.

 

But that is not the case.  I don't know of any medical studies that would confirm this, but it's likely that people not used to spending most of their time in 100+ temperatures are physiologically vulnerable to heat stroke and related ills when put into such temperatures.  At any rate, Judge Pitman ruled that subjecting prisoners to such conditions now amounts to cruel and unusual punishment, which is prohibited by the U. S. Constitution.  And the Texas Department of Criminal Justice now has less than four years to fix the situation.

 

Administratively, they are between a rock and a hard place.  While their estimate of over a billion dollars may be high, they can't spend money they don't have.  And it would be an exercise in Pollyannaish optimism to expect the Texas legislature next year to calmly and deliberately hand over the money needed to air-condition people who can't even vote for them in many cases. 

 

But miracles can happen, and maybe we really will see all Texas prisons air-conditioned before the expiration of the twelfth decade after those Presbyterians exulted in their 70-degree comfort in the steamy summer of 1914 in Orange.  Mind you, I have nothing against Presbyterians.  But everybody in the state government who can do something about this situation should read Luke 16:19-31, in which Jesus tells about the rich man who walked by the poor, sick Lazarus every day without helping him.  Lazarus died and went to be comforted by Father Abraham.  When the rich man died, he found himself in very warm conditions indeed, warmer even than a Texas prison in August.  To the Texas legislators and prison administrators I will only say, it can happen to you.

 

Sources:  The AP article reporting the decision by Judge Pitman is at https://apnews.com/article/hot-texas-prison-judge-ruling-air-conditioning-fa2f1997bd77b47129cdfe5c2350332d.  The First Presbyterian Church of Orange (still standing, with the power-plant building behind it) had its air conditioning described in Ice & Refrigeration, vol. 46, June 1914, pp. 309-310.  The first air-conditioned car is recounted in Raymond Arsenault's essay "The End of the Long Hot Summer:  The Air Conditioner and Southern Culture" in The Journal of Southern History, vol. 50, Nov. 1984, pp. 597-628.  I also referred to the Wikipedia article on Judge Pitman.

Should We Run For the Hills From AI?

  

In the last few weeks, at least two well-publicized resignations from AI firm Anthropic have drawn attention to the pace at which AI models are getting more powerful all the time.  Anthropic's safety head Mrinank Sharma has stated his intention to resign as of next Feb. 8 in an open letter.  And Jacob Coxon has stated in an X thread that "the people building AI earnestly believe it could kill us all by the end of the decade."  According to Wikipedia, Anthropic's leading safety researcher, Evan Hubinger, says that there is a greater than 10% chance that AI systems could "kill all humans" in the next ten years.

 

Reactions to these outcries have ranged from indifference to panic or cynicism, with some noting that Anthropic is planning an initial public offering of stock this coming November.  Recalling the adage that there's no such thing as bad publicity, meaning that just getting your name out there is more important than what they're saying about you, this last take may have a grain of truth in it.  But the more likely explanation for this spate of criticisms and resignations is that something is going on in AI firms that scares the socks off some of those working on it.

 

Predictions of superintelligent systems taking over the world are not new.  I have on my shelf a book by Oxford philosopher Nick Bostrom called Superintelligence that was published in 2014.  In it he examines in thorough detail the various ways that AI could break bad, and discusses possible countermeasures humanity can take.  After three hundred pages of analysis that examines most if not all of the disaster scenarios that people are discussing today, he concludes]the book with these words:  "In this situation, any feeling of gee-wiz [sic] exhilaration would be out of place.  Consternation and fear would be closer to the mark; but the most appropriate attitude may be a bitter determination to be as competent as we can, much as if we were preparing for a difficult exam that will either realize our dreams or obliterate them."

 

The main fear is that AI will get out of control in some harmful way.  That has already happened with the hacking of the cloud-computing company Hugging Face by OpenAI models that were supposed to be isolated in a "sandbox," but managed to escape and damage a good portion of Hugging Face's infrastructure.  This is the most well-publicized example of an AI system that did something bad that its supposed controllers did not intend.  OpenAI says that it's taken measures to prevent such incidents in the future, but as we have pointed out elsewhere, this incident looks a lot like a warning that worse things could happen if AI firms don't implement vastly more secure means to keep experimental AI models under control.

 

The leaders of several AI firms in the U. S. have considered a mutual arrangement to slow down the pace of improvements to allow security measures to keep up.  This has caused concerns that China will catch up to or surpass U. S. competence in AI.  Discussions of government regulation have been discouraged by President Trump, who seems to have never met an AI model he didn't like (never mind about other kinds of models).  So if the government isn't going to do anything significant, and the companies can't get together on how to moderate the pace of improvement, what should the average citizen do?

 

Harking back to an existential threat that I experienced when I was in my twenties, the threat of nuclear annihilation was ever-present as I was growing up.  Nobody I knew developed nuclear weapons, but we all had to live with the consequences of the two superpowers being able to vaporize each other in a matter of a few hours.  There's an existential threat for you.  Somehow (my own hypothesis is it was the grace of God), we have managed to get from then to now without ending it all in a forest of mushroom clouds. 

 

AI is a different kind of threat, admittedly.  The kinds of systems that are most vulnerable to it seem to partake of the nature of AI itself:  large-scale networked systems that are centrally controlled.  There are certain infrastructure vulnerabilities in the U. S. that genuinely concern me, notably the fragile nature of  power grids.  They rely on substation transformers that cost upwards of millions of dollars, are hugely inconvenient to repair or replace, and are very limited in supply.  There have been news reports recently about how a concentrated and coordinated attack on substations nationwide—either physical, which would be hard to do, or a cyberattack, which might be a lot easier—could cripple the grid enough to leave large swathes of the U. S. without electric power for months.  That could lead to a civilization-destabilizing crisis.

 

Fortunately, there are some things we can do about this.  One is ramping up production of spare transformers (preferably making them in the U. S.) so that we have enough to replace as many as we might lose in a widespread attack.  Another is to take advantage of smart-grid capabilities now coming online to make it possible to operate pieces of the grid without having it all interlocked together.  As more renewable sources come online that are distributed rather than coming in a few big lumps like traditional power plants, one can imagine making the power grid more like the internet, which was designed to be robust against damage to a good many of its nodes. 

 

I'm not a power-systems engineer and this may be a fantasy, but it seems like it's within the realm of possibility.  Of course, determined hackers can wreck even well-protected infrastructure.  But making the grid more local, which is what we're really talking about, makes it harder for an attacker to take down all of it.  If local pieces have to be dealt with one by one, there is safety in variety.

 

I hope we can keep AI under control to the extent that the benefits outweigh the liabilities.  But if the history of engineering ethics is any guide, though, we will learn how to do this by dealing with the aftermath of some failures first.

 

Sources:  I referred to posts by Mrinank Sharma at https://www.linkedin.com/posts/andrew-hilger-436b713_the-attached-resignation-letter-from-mrinank-activity-7427113299336454144-4CVm/and Jacob Coxon at https://x.com/hilbertspaess/status/2097476196791709843 as well as the Wikipedia articles on "Existential risk from artificial intelligence" and "OpenAI-Hugging Face incident."

 

Update on Reader Survey:  As I suspected, there appear to be relatively few of you who read this blog regularly.  But I am very grateful to all nine who responded to the survey, and hope there are a few more who didn't bother to respond but still read it regularly.  You are in exclusive company!

 

Survey for Engineering Ethics Blog Readers

   

As promised last week, today I am asking my readers to take the usual three to five minutes you would use to read my blog, and instead complete a short survey which can be found at

 

https://forms.gle/75G4ZxDZFJyNcaD19

 

It consists only of two multiple-choice questions and two optional comment spaces, so it's not like I'm asking you to spend a lot of time on it.  Indications from the statistics that the blogger system provides are that relatively few of you are regular readers.  Most visitors to the website show up because they are directed here by interest in a particular issue I happen to be addressing.

 

Whatever your reading habits, I would like to hear from you.  The regular blog will resume next week.

 

Karl Stephan