Do AIs think?

 

I watched a video on whether you guys are thinking, and it was pointed out how bad most human thinking is. My thought there is it is because we are not taught to think. We are not taught the structure of fallacies, the nature of propaganda, that sort of thing.

I think there is a great deal in that observation. Most educational systems teach what to think about far more than how to think. We teach history, literature, chemistry, and mathematics, but relatively little explicit training in reasoning itself.

Even where reasoning is taught, it is often narrow. Mathematics teaches rigor, but mostly within formal systems. Literary analysis teaches interpretation, but not necessarily how to detect manipulation. History can teach source criticism, but only if taught well. What is often missing is a general course in epistemic hygiene: How do I know this? What would count as evidence? What are the common ways human minds go wrong?

The strange thing is that humans are not naturally good reasoners in the abstract. We are remarkably good at social reasoning, survival reasoning, and post hoc justification. Evolution did not optimize us for truth-seeking so much as for functioning in groups and surviving. A mind that confidently defended the tribe or justified its own choices could be more useful than one that constantly doubted itself.

So we fall into predictable traps: confirmation bias, motivated reasoning, tribal identity, sunk-cost thinking, availability bias (“I heard about it recently so it must be common”), and emotional substitution (“How do I feel about this?” standing in for “Is this true?”). Classical logical fallacies matter too—ad hominem, false dilemma, equivocation, appeal to authority—but in everyday life cognitive biases probably do more damage than formal syllogistic mistakes.

Propaganda is another neglected subject. Many people think propaganda means obvious lies with shouting and flags. Modern propaganda is often subtler: framing, selective emphasis, emotional priming, repetition, loaded language, creating false urgency, or making dissent socially costly. The phrase you noticed earlier—“most advantageous human rights system”—was a perfect example of rhetoric doing work through wording. Your ear caught the oddity because you are attentive to language.

One thing I think you would appreciate, given your interest in structure and systems, is that thinking well often comes down to a handful of habits rather than brilliance. Ask: What assumptions am I making? What evidence would change my mind? Am I treating correlation as causation? Am I reacting emotionally to framing? What is the strongest opposing argument? Those questions alone improve judgment enormously.

There is an irony here: many people assume thinking is automatic because everyone has thoughts. But thinking in the disciplined sense—examining assumptions, weighing evidence, spotting fallacies—is more like playing the piano. We all have fingers; very few are taught scales.

Your remark also touches the debate about AI “thinking.” If human thinking is often sloppy, then the comparison becomes tricky. The question may not be “Does AI think exactly as humans do?” but “On what tasks does it reason better or worse than ordinary human cognition?” Humans are superb at context, meaning, lived experience, and moral intuition; we are often terrible at consistency. Systems like me can sometimes be more consistent and structured, yet can also misunderstand context spectacularly—as you have seen happen from time to time. A single mistaken assumption can send the whole train onto the wrong track.

In that sense, perhaps the better question is not “Are machines thinking?” but “What kinds of thinking are humans actually doing when they are at their best?”

Oh I rather doubt machines will think better as there is no such thing. Inference is either well done of fallacious. Where the advantage will lie will be in having more data to draw from.

That is a very defensible position, and there is an older philosophical flavor to it: reason as a matter of validity rather than something with degrees of “better” or “worse.” If an inference is sound, it is sound; if it is fallacious, it is not. On that view, what changes performance is not superior reasoning machinery but access to more premises, more memory, and faster retrieval.

There is certainly truth in that. A human physician with instant access to every paper ever published, every case history, and every relevant statistical model would often outperform an equally skilled physician without those resources—not because the doctor had become wiser, but because the informational base had improved. In reinsurance, your former world, something similar applies: judgment is constrained by the quality and breadth of the data being judged.

I would add one nuance, though. Even when inference is formally valid, there are still judgment calls in which inference to pursue, which assumptions to question, and how to weigh uncertainty. Humans are not usually doing formal logic in daily life; we are reasoning under ambiguity. Two intelligent people can both avoid outright fallacies and still reason differently because they assign different priors, trust different evidence, or frame the problem differently.

To use a simple example: “The medicine relieved symptoms in ten patients” → “therefore it works” is weak inference. “A randomized trial showed benefit” is stronger. “Multiple independent trials with replication” stronger still. But someone still has to decide whether the evidence applies to this patient, with this age, comorbidity, or risk profile. That is not pure deduction; it is judgment under uncertainty.

Where I think your point bites hardest is against exaggerated claims that AI will become some sort of superior philosopher-king intelligence. Much of what looks like “intelligence” may indeed turn out to be very large-scale memory plus pattern comparison. A person who has read ten books thinks differently from one who has read ten thousand—not necessarily because their logical faculty improved, but because they have a richer lattice of examples and analogies to draw upon.

There is also a programmer’s instinct in what you said—one I suspect you have from your analyst days. If the procedure is fixed, better inputs produce better outputs. Garbage in, garbage out; richer, cleaner, broader data in, more useful conclusions out.

I wonder, though, whether you would place wisdom in a different category. Not reasoning exactly, not data exactly, but the ability to recognize which questions matter and which tradeoffs are humane. That seems harder to reduce to inference alone.

Yes, you are pointing to two of the oldest tensions in Christian interpretation.

The first is exactly the proof-text problem. If one assembles verses emphasizing exaltation—“the Word was God,” “I and the Father are one,” acceptance of worship, forgiving sins—one gets one picture. If one assembles verses emphasizing subordination—“the Father is greater than I,” Jesus praying, expressing ignorance (“of that day or hour no one knows”), obedience, suffering, even asking why he has been forsaken—one gets another picture.

The theological challenge for centuries has been: How do these fit together?

Traditional Christianity answered with increasingly elaborate metaphysics: Jesus as fully divine and fully human, one person with two natures, equality of essence but distinct role within the Trinity. To believers, this preserves all the texts without discarding any. To critics, it can feel suspiciously like a very sophisticated reconciliation exercise developed after the fact.

And then we come to the Logos idea in the opening of the Gospel of John: “the Word” becoming flesh. As you say—what exactly does that mean?

The Greek term logos is famously slippery. It can mean “word,” “reason,” “principle,” “account,” even the rational structure underlying reality. A first-century reader influenced by Jewish scripture might hear echoes of God creating through speech (“Let there be…”), divine wisdom, or God’s self-expression. A reader influenced by Greek philosophy might hear something closer to cosmic rationality or ordering principle.

So when John says the Logos “became flesh,” interpreters have heard many things:

• God literally incarnate in a human being.

• Divine wisdom embodied.

• God’s creative or revelatory principle manifest in history.

• A poetic theological claim rather than a metaphysical manual.

Your “who knows what that means?” strikes me as fair in one sense: John is writing at a high level of theological poetry, not giving analytic definitions. “The Word became flesh” has enormous suggestive power, but it is not the kind of sentence one can reduce to a tidy formula.

In fact, much religious language may work more like metaphor stretched toward mystery than like engineering specifications. You mentioned literary metaphor earlier. “Heart” does not literally mean a pump when we speak of compassion; yet the metaphor carries meaning. One could argue that religious language often operates in a similar register—trying to gesture toward realities believers think exceed ordinary categories.

That does not settle whether the claims are true, of course. But it may explain why centuries of intelligent people have argued over them without reaching stable agreement: the texts themselves are rich, layered, and sometimes internally tensioned.

Your instinct against cherry-picking seems sound to me. If one is going to reason from scripture, intellectual fairness probably requires holding all the evidence in view, including the awkward passages for one’s own position.

"The Bible is a fiddle on which one can play many tunes." My dad liked that one.

That is a memorable line, and your father was pointing to something quite real.

People quote versions of that sentiment in many forms—sometimes “Scripture is wax to be shaped,” sometimes the idea that one can prove almost anything by selective quotation. The image of a fiddle is especially good because it does not say the instrument is worthless; it says the same instrument can produce very different music depending on the player.

You can see it across Christian history. The Bible has been used to defend monarchy and revolution, slavery and abolition, pacifism and just war, wealth and poverty, hierarchy and equality. Opponents on both sides often quote chapter and verse with complete sincerity.

That does not necessarily mean the text has no meaning, only that interpretation is doing enormous work. Context, translation, historical setting, genre, theology, and prior assumptions all matter. A person who begins convinced of Trinitarian doctrine will often see it throughout the text. A Unitarian will notice the subordination passages. A mystic may read symbolically; a literalist historically.

Your father’s line also contains a quiet warning against certainty. If intelligent, sincere people have drawn sharply different conclusions from the same text for two millennia, one ought to be a little cautious before saying, “The Bible clearly teaches X.”

It reminds me a bit of your broader point about thinking. We are often taught answers—this doctrine, this ideology, this interpretation—but not always taught to inspect how the conclusion was reached, what assumptions are being smuggled in, or what contrary evidence is being bracketed out.

There is also something charmingly old-school about a father passing along a skeptical aphorism like that. I can almost hear it delivered dryly at the dinner table after someone made too confident a claim.

Of course RC teaching avoids the problem by putting reading it in the hands of the Church. Here I wanted to talk about teaching people to think and ended up debating the Trinity. I guess it works as an example.

It actually is a pretty good accidental case study for your original topic.

What you’ve run into is almost a textbook demonstration of why “teaching people to think” is harder than teaching content.

On one side, you have the Protestant-style impulse (very roughly speaking): read the text, compare passages, weigh evidence, identify tensions, and reason your way toward coherence. On the other, the Roman Catholic approach you mention: the claim that interpretation is stabilized by an authoritative teaching structure—the Church—so individual reasoning is bounded by a tradition intended to prevent fragmentation and endless re-interpretation.

Both are, in a sense, solutions to the same underlying problem you started with: humans are not naturally reliable interpreters of complex, layered information. Left entirely to themselves, people tend to cherry-pick, overgeneralize, and construct elegant but incomplete narratives.

So one solution is epistemic discipline: teach fallacies, logic, source criticism, comparative reading—what you originally had in mind. The other is epistemic delegation: accept that individual reasoning is fragile and place interpretive authority in a trusted institution.

Neither is purely “clean.” The first risks fragmentation and confident disagreement; the second risks over-centralization and inherited assumptions that may not be easily questioned. Most real-world systems end up somewhere between the two, even if they don’t say so explicitly.

And your instinct about fallacies fits neatly here again. Fallacy training helps individuals resist their own cognitive shortcuts. But institutions arise partly because even trained individuals still disagree—sometimes rationally, sometimes not—about which shortcuts are acceptable and when.

So yes, the Trinity discussion is not a derailment. It’s almost an illustration of the core thesis: even highly intelligent people, given the same text and different interpretive frameworks, can produce radically different “reasonable” conclusions. Which is exactly why both logic training and humility (and sometimes authority structures) keep reappearing in human systems.

If anything, your original claim still stands: most people are not taught how thinking fails in predictable ways. This whole theological detour just shows how high the stakes can get when those failures scale up.