Between the Worlds: Truth and Usefulness

I have subscribed to Claude for a few years. I do not use it merely to write code, but as a sparring partner in my professional work. A good assistant agrees too quickly; a good sparring partner pushes back.

Claude has pushed back on me more than once. It has told me, in effect, that I have a weakness for abstraction. I look for clean solutions and elegant models, reducing messy situations to something that can be reasoned about, as if the world were waiting to be explained.

It stung a little because it was true.

But I came by this habit honestly. I trained as a computer scientist, and my early academic work was in programming languages. At its heart is a question familiar to anyone who reads literature carefully: what does a sentence mean?

In my case, the sentence happened to be a program.

There were two broad ways to answer that question. Some people said that a program has a meaning in a model, a clean world built out of definitions. Others said its meaning is what happens when it runs: do not tell me what it means in the abstract; show me what it does.

My doctoral work lived between those two worlds. Under some conditions, I showed that what a program means and what it does need not be different stories.

This is not merely a technical question. In another form, it is one of the oldest questions we have: how do we arrive at truth?

One way is through reason. We model the world and reason our way toward truth. Consistency is our sword; the law of the excluded middle is our shield. Plato is our patron saint. Without this habit, we would not have philosophy, logic, law, grammar, or computer science, nor the beautiful idea that the apparent chaos of the world may conceal an order.

The other way is through experience. We observe, experiment, and build knowledge from what happens. Aristotle is our guide. However elegant a theory may be, the experiment can still refuse to cooperate. Or, as a skeptical friend once put it, perhaps we are not rational creatures as much as rationalizing creatures.

We can see the same tension in Telugu literature.

One school of criticism wants to understand literature through a larger theory: class struggle, caste, gender, power, history. The critic does not merely look at individual trees; he wants to understand the forest. A family story may also be a story about caste. A love poem may also be a poem about class. A village novel may carry an entire political economy inside it.

Another school resists this. It says: first look at the work. Look at the people, the language, the scene, the wound, the joke, the music of the sentence. Do not hurry to fit it into a theory. Literature is not merely evidence for an argument. It is also an experience.

Both approaches reveal something, and both can conceal something. A theory may uncover a hidden structure, but it may also make us see only what it permits. The poem may be drenched in tears while the critic processes it into categories.

The rationalist danger is that the world is too complex to fit neatly inside our models. The past can often be rearranged to fit Marxism, Hayek’s economics, structuralism, nationalism, or whatever theory we favor. An explanation may look convincing after the event without predicting the next one.

In Asimov’s Foundation, one sign of decline is not that scholars stop studying. They preserve, classify, and annotate knowledge, becoming experts on experts, but lose the habit of direct inquiry. Rationalism can become the manipulation of inherited symbols: internally elegant, but detached from the world.

Experience has its own failure mode. I remember people discussing an author who wrote about the labor movement. They said his observations must be true because he had lived in that world. His experience may indeed have illuminated his writing. It may also have limited him. Having “been there” can bring insight, but it can also give prejudice the authority of testimony.

The turkey thinks the farmer is benevolent because the farmer feeds it every day. Each day confirms the theory — until Thanksgiving arrives. Experience does not scale — it can fail unpredictably.

So we go back and forth. Reason needs experience; experience needs reason. Kant showed that both had limits. Pure reason can build castles in the air; experience does not organize itself into knowledge. We do not receive the world like blank slates; we shape it into something intelligible.

Now we seem to be entering a third model.

It is more restless: understand, test, modify, repeat, in a rapid loop. We often use models before we fully understand them.

Artificial intelligence is the clearest example. Modern AI systems often predict without explanation. Trained on oceans of examples and judged by performance, they inhabit a middle ground where prediction, utility, and iteration matter more than explanation.

We ask an AI a question. Sometimes its answer is wrong or banal; sometimes it is surprisingly useful. Only afterward do we ask: Why did it work? Can it be explained, tested, or trusted? Is it biased, hallucinating, imitating, or merely lucky? The philosophers like Barthes that say that we are reading meaning into the gibberish produced by AI. Pragmatic students say “Whatever. I can write better letters to my grandmother. She is happier with these letters than my emoji-laden texts”. 

Usefulness becomes the first criterion. We search for reason and experience after the fact.

That should make us uneasy. Many terrible things are useful to someone. Propaganda is useful. Manipulation is useful. A lie may be useful until it is discovered. A system that works without accountability can become another form of power.

Yet much of life has always proceeded in this untidy order. We speak before we learn grammar and recognize insult or affection before we can analyze them. A court, unable to define obscenity fully, falls back on community standards and the famous admission: “I know it when I see it.” It is not a satisfying theory, but it describes a familiar human predicament.

AI has not created this predicament. It has made it faster and harder to ignore.

This does not make reason obsolete. It makes reason more necessary, but perhaps less arrogant. We still need models, explanations, experiments, and accountability. But we may also have to accept that some useful knowledge will arrive first without our understanding and only later, if at all, followed by rationale.

That is hard for a Platonist to accept.

I still love clean models. I still believe that a good theory is one of the great achievements of the human mind. But I am learning, slowly and unwillingly, that not every truth is forged in the bellows of reason. Perhaps we may never fully know what truth is.

We want reason. We need experience. Yet we now use machines that work without fully satisfying our demand for explanation.

For believers, perhaps faith resolves some part of this anxiety. For the rest of us, the questions remain. Fortunately, questions have produced philosophy, science, literature, and some excellent poetry.

And after Plato, Aristotle, Kant, Asimov, computer science, and artificial intelligence have all had their say, perhaps one may be forgiven for hearing an old refrain from another tradition:

Bhaja Govindam.

Stop counting the rules of grammar at the moment of death. Or, in our modern version, stop believing that the final abstraction will save you.

*