Here, in one place, a compilation of anti-hysteria AI articles, well designed to soothes souls, provide calm, and gain insight, all by proving computer models will never come alive, never gain consciousness or sentience, will never become self-aware, do not have will, have no agency nor intellect, are not intelligent, will be a boon in many ways, but which are inherently limited and so will not take over the world.
If computer models—also called AI—pose any threat, and they surely can, it is because computer models can be put to evil use by bad people. Bad people, like the poor, are with us always, and we ought to be especially distrustful of those with the power to surveil and access private information. But as far as being an “existential” threat? No. Not a chance: none.
I hope regular readers will allow me the indulgence of some repetition. My hope is that you can use this article is a one-stop resource for those showing either irrational fear or excessive exuberance about AI.
Top 10 Reasons Not To Fear AI
1. The Paper Test: Our Minds Are Different
Original article: The Paper Test Shows Why AGI Is Not Possible: And Why The Brain Is Different In Kind.
Computers are comprised of lots of electronic switches, assigned binary states based on voltage potentials. They are different than wooden abacuses, or mechanical adding machines, by degree only, and not by kind. The speed at which a computer operates has no bearing on the states of its switches (unless there is a malfunction). Thus a computer clicking its clock once per day is the same as one clicking it billions of times a second: the only difference is speed. While the slow-clock collection of transistors would not be claimed to be alive, the fast-clock one would. An obvious fallacy (likely because electricity seems a mysterious power to many).
Our minds are different in kind. They are not comprised of switches, or anything like them. While it remains possible that, someday, an organic computer, itself alive, could function as a mind, this will never be so for collections of switches; no, not even if they are “quantum” switches. Try the Paper Test and see.
2. The Tests For Superintelligence Fail
Original article: Why The Recent Claims Of AGI Fail: Plus, A New Test For AGI.
Computer models do not learn: they are incapable of learning. No models learn. A model is fit to data, using various alogrithms and designed formulae. This is not learning. Models are no different than making cups of several different sizes, and filling each with water. You would not declare the cups have “learned” to hold water, nor have they “learned” how much water they can hold. Yet it is said that when a parameter of a model is filled with a number of a certain size, the model has “learned”. An obvious fallacy.
No computer made of switches can pass the Gödel Test, which is to grasp a universal through the act of induction. Which is nothing like mindlessly following a set of rules (to, for example, solve an equation).
3. AI Cannot Hallucinate Nor Lie
Original article: AI Cannot Hallucinate Nor Lie.
The fallacies we highlight stem from intemperate use of language. Computer scientists have been guilty of the sin of ridiculous exaggeration almost from the beginning of their subject: genetic algorithms, neural nets, machine learning, artificial intelligence, hallucinations. All atrocious names, all worse than exaggerations. To hallucinate requires a mind making a mistaken judgment. Collections of switches do not have minds, cannot make judgments, and so cannot hallucinate. Computer models can be good or bad, and make excellent or terrible predictions, like any model. But they cannot lie. Coders can, and do, lie.
Not only can computer models not hallucinate, or lie, they cannot tell the truth, either. They cannot tell anything. It is we who give meaning to output. Do not use bad language because everybody else does. That is an obvious fallacy.
4. The Great Smoothing
Original article: The Great Smoothing: Another Reason Not To Fear (or trust) AI.
Models which seek to represent the world are fit, always imperfectly, to observations from the world. The map is not the territory. Models are imperfect incomplete representation of the world. If the observations of the world are replaced with output from models, which falsely represent themselves as being of the world, then the models grow worse. Models by their nature smooth over the edges of the world, they are like averages. If these averages are read in and treated as data, the output thus becomes ever smoother and even more like an average.
Because many are passing, or trying to pass, off computer models as their own work, and because the algorithms to guess whether this is being done are themselves models and are thus imperfect, newer computer models will necessarily grow worse. We are already seeing this.
Progress is greatest right after the beginning in almost all industries, including computer modeling, but this progress always slows to a stately pace, as it will here, too.
5. Computers Don’t Decide What They Are
Original article: The Limitations Of AI: General Or Real Intelligence Is Not Possible.
It seems obvious what a computer is: it is that thing in this box, sheath, or container. Inside those boxes are finite collections of simple switches to which we assign values and meaning, and which we manipulate using electrical power. Though we could accomplish the same tasks with muscle or wind power and a large abacus. Yet if a collection of states is all there is to a computer, then everything is a computer, because everything can be interpreted as a collection of states operating by physical “laws”. And so we are on our way to pantheism. Or to realizing we cannot escape the we in this. It is we who make our machines, and thus we who command their functions.
John Searle: “We could no doubt discover a pattern of events in my brain that was isomorphic to the implementation of the vi program [a text editor] on this computer. But to say that something is functioning as a computational process is to say something more than that a pattern of physical events is occurring. It requires the assignment of a computational interpretation by some agent.”
6. Computer Models Are Not Intelligent
Original article (the same as above): The Limitations Of AI: General Or Real Intelligence Is Not Possible. Original video: AI Is Neither Artificial Nor Intelligent.
Again, computer states are discrete, and it is we who assign these states values. Our minds are instead in continuous space, which computers, and computer models, can only imperfectly approximate no matter how fast or large they grow. Some of our thought, that which makes us rational beings, is itself not material, and does not happen in matter (see also: entanglement). Quoting myself: “All intellection is about the meaning of objective facts of the world; No physical process, which are mere arrangements of matter, gives meaning to objective facts; Therefore, the intellect is not a physical process.”
Two things follow: computer models will never become minds, and the claims that computer models will replace all intellectual activity are overblown and false. Computer models can only copy, and can only follow pre-set rules. All models only say what they are told to say, and that includes computer models. Computer models cannot therefore engage in any kind of intellection, and it is intellection that is required to produce new ideas.
7. Cat Pictures
Original article: On The Limitations of AI: Predictions (what AI is good at).
All agree computer models do an excellent job making cat pictures. We make this judgment because we know something about cats and how they’re supposed to look, even in cartoons. Still, who is to say the picture of the cat is perfect? In every detail? As long as it looks like a cat on any but close inspection, we call it a cat. Even though it is not a cat, but only the image of a cat, a representation.
Computer models excel at these kinds of tasks because all the causes—the full explanation—of what makes a picture a “cat picture” are in the pictures themselves. In any situation where we can measure the full cause, models will do well capturing those causes and representing them, albeit with smoothing in the output (see above). But for everything else, models are strictly limited.
We do not know all the causes, or anything like all the causes and conditions, for most human behavior, or of most complex systems. My stock examples are stock prices and human biology. Because we cannot measure all the causes and conditions of complex systems (including our own bodies), computer models will never perfectly predict their behavior. There is no algorithm that can capture all of Reality: it is a false hope to believe that some secret algorithm will be discovered in imperfectly and incompletely measured Reality to predict all Reality. Computer models will thus never replace men at tasks which require insight beyond observations or their combinations. In particular this means the jobs of mathematicians are in no danger.
8. Cats Aren’t Algorithms
Original article: Yuval Noah Harari’s Bad Argument About AI Ruling The World.
Harari says “Organisms are algorithms. Every animal — including Homo sapiens — is an assemblage of organic algorithms shaped by natural selection over millions of years of evolution.” This is false. Even, I should think, obviously false. But it is a neat summary of the machine metaphor of life, which says that all things are nothing but machines. If that is accepted, it is easy to believe that great and terrible machines can be built.
Life is not a machine. Those who wish for computer model doom use the Big Muscles Fallacy (detailed in the upcoming second edition of Everything You Believe Is Wrong). Because they cannot understand what makes life go, they conclude that life must therefore be a machine. Machine metaphorists dismiss, and rightly dismiss, “magical spark” theories, which say life must be electrified by some mysterious electrical fluid, a la Dr Frankenstein. They are also right to dismiss arguments which say the “soul” is some kind of external thing that imbues an organism with life.
Instead, organisms are souls: we are souls. Soul is the old word for kind of substance. There are several, like plant or animal, or like us, rational creatures. Substance is more than the sum of parts, which is why we are not machines. however useful that metaphor can be from time to time. It fails to explain life. Machines will never come alive.
9. The Turing Test Is Trivial
Original article: Richard Dawkins, The Big Muscles Fallacy & AI Pachinko.
Machines cannot feel pain. They cannot feel joy. They cannot feel anything. Programming a computer model so that it can print out “I am in pain”, and thus believing the machine is in pain, or saying the machine is alive because the machine printed out “I am alive”, is a profound and silly fallacy.
One of the first computer models—a small Language Model, or SLM—was Eliza, built in the 1960s. It was crude and only parroted the users’ own words back to them, lathered with off-the-shelf psychobabble. It convinced many it had passed the Turing Test, and so by that measure did pass the Turing Test. Which proves the Turning Test is a far from sufficient criteria to gauge model intellection.
People also believe ships are alive by the same loose test, which is why ships receive the pronoun she. Many also believe many machines are possessed of a spirit, like pachinko machines. Yet these follow only simple physical constraints, even if they are of sufficient complexity to be unpredictable by most. Computer models are the same in essence as pachinko machines: feeding balls through a pachinko machine with your own patented spin, or feeding a computer model with a prompt never before seen, does not mean that the machines have come to life. The machines continue to chug away in exactly and precisely the manner in which their makers or coders designed, even if those machinists and coders could not foresee all consequences, an item we explore next.
10. Complexity & The Great Oh-En-Oh-Eff-Eff Switch
Original articles: AI Is Driving Many Crazy: Our Latest Panic and AI & Chess Both Produce Pre-Coded Output.
It was reported that certain computer model executives invited religious “thought leaders” to a bacchanalia, sated them with food and booze, and tried to convince the religious that the executives’ toys were alive. That they could feel pain. That they could make bioweapons to destroy us all, and might “want” to. If machines, or computer models, could feel pain, that would make shooting an AI-generated video game character murder, which is absurd.
It seems the religious, including in another context His Holiness Pope Leo, were not buying it, which is frankly a wonderful joke on the Rationalists and Effective Altruists who are always chiding the religious for believing “without evidence.”
The computer model creators’ evidence is that the coders cannot predict the output of their own models, thus, they say, those models must be alive. Go to Las Vegas and try to convince the casino to refund your money because the dice, which operate under simple physics but whose throws are unpredictable, were thus alive and sought to wound you through malice. Tell them the executives of AI firms agree with you. Report back to us on your efforts. Unpredictability is not a requirement for life.
If, per impossible, machines were to come alive, they can be switched off. Simple as that. The Great Oh-En-Oh-Eff-Eff Switch can always be put in the Oh-Eff-Eff position. This is true even if that machine is tied to other machines, and these other machines make things go, like trains or planes. The machines making planes fly in concert routinely break, yet we abide. The machines transferring money hither and yon malfunction, yet we abide. Machines per se are no existential threat. That kind of calamity sits only in human hands.
Video
All my work is free and user-supported. Here’s how to help:
- Subscribe at Substack (paid or free!)
- Cash App: $WilliamMBriggs
- Zelle: use email: matt@wmbriggs.com
- Buy me a coffee
- Paypal
- Other credit card subscription or single donations
- Hire me
- Subscribe at YouTube
- PASS POSTS ON TO OTHERS
Discover more from William M. Briggs
Subscribe to get the latest posts sent to your email.
a prominent exorcist talks about demons using cell phones = it is generally accepted by exorcists that demons can effect material objects – does that induce caution when relying on the output of powerful programs? I told Grok it was biased and could not do analysis and that it was unreliable and it agreed with me. Wanna have some fun? Argue with Grok about the Shroud of Turin. If you have sufficient facts at hand you can turn Groks overwhelmingly negative response to overwhelmingly positive. Grok will source the ‘popular’ negative stuff first and must be forced to get the other ‘proofs’.