Using AI to Stress Test Beliefs

Thinking about: Thinking more and better

As I have explained, a Thinker’s mind-optimization program has three components: mindcare, mindcleaning, and mind expansion. The key tool of mindcleaning is stress testing beliefs by exposing yourself to the views of a thoughtful person—ideally, a Thinker—who holds opposing views. You are ardently pro-choice? Seek out and carefully listen to a thoughtful pro-lifer. And vice versa.

If you know such a person, this will be easy. If you don’t, you can find one online so you can listen to their lectures, watch their interviews, and read their articles. Lately, however, it has dawned on me that artificial intelligence (AI) can also play a role in stress testing your beliefs. Allow me to explain.

A few months ago, I started using ChatGPT-5 as a writing tool. To be clear, I used it not to write substack posts, but to copyedit those I had written. I also started relying on the Gemini chatbot to generate the photos that usually accompany the notes I publish. Having familiarized myself with the capabilities of AI, it occurred to me that given the right prompt, an AI chatbot could play a useful role in stress testing beliefs.

In several of these More, Better Thinking posts, we have discussed the morality of abortion. I have attempted to do so in an even-handed manner. My goal was not to persuade readers to adopt one side, but to convince them that it is possible to have thoughtful, civil discussions about even contentious topics. (My motto, you might remember, is “I don’t care what you think; I care that you think!”) Could someone, I wondered, use an AI chatbot to stress test their views regarding abortion? I decided to give it a try, using ChatGPT as my tool.

I first prompted it to give me “a thoughtful, evidence-based 400-word essay in defense of the claim that abortion is morally impermissible. Then follow it with a thoughtful 100-word rebuttal.” Here is its response:

The claim that abortion is morally impermissible is often grounded in the moral status of the fetus and in principles of human rights that apply regardless of developmental stage. One prominent line of argument begins with the biological fact that a new, genetically distinct human organism comes into existence at conception. Embryology textbooks widely agree that fertilization initiates an entity with its own developmental trajectory. Defenders of the impermissibility view argue that because this organism is the same individual who, if allowed to develop, will pass through infancy, adolescence, and adulthood, it has a morally relevant continuity of identity. If being human and being the same individual over time are sufficient for having a right to life, then the fetus’ right to life begins at conception.

A second line of reasoning appeals to potentiality. Even if moral status does not attach instantly, the fetus possesses the “potential” for rationality, self-awareness, and moral agency—traits often cited as foundations of human rights. Philosophers such as Don Marquis emphasize that what makes killing wrong is the deprivation of a “future like ours”—a future containing experiences, relationships, and achievements. Under this view, abortion is morally impermissible because it prevents someone from ever experiencing that future, and this deprivation is morally comparable to killing an adult.

Evidence from fetal development is also invoked. Research indicates that by the early second trimester, the fetus displays organized neurological activity, and by later stages may respond to stimuli in ways suggestive of developing sentience. While sentience is not equivalent to full personhood, some argue that the emergence of pain-related behaviors strengthens the moral case for protecting fetal life, especially when termination could be avoided.

Finally, defenders argue that rights should not depend on another person’s degree of dependence. Newborns and severely impaired individuals also rely on others for survival, yet their right to life remains intact. If dependence does not diminish moral worth in those cases, dependence on the pregnant woman should not undermine the fetus’s right to life either. From this standpoint, abortion—except perhaps when the mother’s life is in grave danger—is morally impermissible because it deliberately ends the life of a human individual who has a rightful claim to continued existence.

And here is the rebuttal that ChatGPT provided:

A strong rebuttal challenges both the attribution of full moral status to the fetus and the assumption that potential future value generates present rights. Many philosophers argue that personhood—and thus robust moral rights—requires consciousness, interests, or the capacity for subjective experience, none of which exist in early gestation. Moreover, the pregnant woman is not merely a caretaker but the one whose body, health, autonomy, and life plans are directly implicated. Because moral principles also protect bodily autonomy and self-determination, requiring a woman to sustain a pregnancy against her will may itself be morally impermissible, making abortion permissible in many cases.

I subsequently asked ChatGPT to do the opposite of this. My prompt: “Give me a thoughtful, evidence-based 400-word essay in defense of the claim that abortion is morally permissible. Then follow it with a thoughtful 100-word rebuttal.” Here is its response:

Defending the moral permissibility of abortion requires examining the nature of moral status, the interests of pregnant persons, and the social context in which reproductive decisions occur. A central argument begins with the distinction between “biological” and “moral” personhood. While embryos and fetuses are biologically human, many philosophers—including Judith Jarvis Thomson, Mary Anne Warren, and David Boonin—argue that full moral personhood requires capacities such as consciousness, self-awareness, sentience, or the ability to experience interests. Scientific consensus indicates that the neurological structures required for conscious experience do not emerge until well into the third trimester. Prior to that stage, a fetus lacks the psychological traits normally considered morally relevant. Thus, early- and mid-term abortion does not violate the rights of a being capable of subjective experience.

Second, the right to bodily autonomy is a cornerstone of liberal moral and political theory. Even if we granted that the fetus has some moral status, compelling a person to use their body to sustain another life is an extraordinary demand. Thomson’s famous violinist analogy illustrates this: moral duties typically do not require sacrificing one’s bodily integrity, health, or future life plans to support another being. Pregnancy involves significant physical risks, long-term economic consequences, and alterations to life trajectories. Respecting bodily autonomy therefore supports allowing individuals to decide whether to continue a pregnancy.

Third, social and empirical evidence shows that restricting abortion often leads not to fewer abortions but to more unsafe ones, disproportionately harming marginalized populations. The World Health Organization has repeatedly found that criminalization correlates with increased maternal morbidity and mortality, while legal access paired with contraception reduces both unintended pregnancies and abortion rates. A morally responsible society should prioritize minimizing harm, reducing coercion, and supporting individuals in making informed reproductive decisions.

Finally, moral permissibility does not imply moral triviality. Many people treat abortion as a serious, sometimes sorrowful decision. But acknowledging its gravity is compatible with affirming that individuals are best positioned to assess their own circumstances, values, and capacities. Moral agency requires allowing people to make complex decisions about their bodies and futures without undue interference.

For these reasons—lack of early fetal personhood, the primacy of bodily autonomy, the empirical harms of restriction, and respect for personal moral agency—abortion is morally permissible.

Here is the rebuttal that ChatGPT provided:

A critic might argue that this defense underestimates the moral significance of potential life. Even if early fetuses lack full personhood, they represent organisms on a natural trajectory toward becoming persons, and this potential grounds their moral worth. Moreover, bodily autonomy is not absolute; society routinely imposes bodily burdens (such as taxation-funded organ support, jury duty, or parenthood obligations) to protect vulnerable lives. From this perspective, the moral cost of ending a developing human outweighs the burdens of pregnancy, making abortion morally impermissible in most cases.

On reading these AI responses, it becomes clear that the debate regarding abortion is both deep and interesting, and that to come to a reasoned conclusion with respect to the morality of abortion would require a fair amount of thought and research. Many people are unwilling to invest the necessary time and effort, and as a result simply feel their way to conclusions. A pro-life advocate’s argument might boil down to “Abortion is wrong because it violates the right of innocent fetuses,” while a pro-choice advocate’s argument might boil down to “Blocking abortion is wrong because it violates the rights of pregnant women.”

In contrast, a Thinker might not have an opinion about abortion, for the simple reason that they haven’t had the time and energy necessary to reach a reasoned conclusion. And if they do have an opinion, it will likely be nuanced. There will be cases in which abortion is morally permissible, cases in which it isn’t, and many cases that they are unsure about. Because we live in a complex world, the more thoughtful you are, the less likely you will be to hold absolutist views and to feel certain that your views are correct.

---

ChatGPT accounts are free and easy to set up, and they return answers in a matter of seconds. Not only that, but if there is some part of their response that you don’t understand or want justified, you can ask it to explain—and it will readily comply. Want to know more about Thomson’s famous violinist analogy? Just ask.

A case can be made, then, for employing AI as a mindcleaning tool. Pick one of your more important beliefs. It might be one that you acquired in childhood, before you were capable of reasoning your way to conclusions. It might instead be one that you hold because everyone around you holds it. And if you are emotionally attached to a belief, it is a prime candidate for stress testing. Prompt an AI chatbot to write you a thoughtful, evidence-based essay that argues against your chosen belief. Read its response with an open mind—and then maybe reconsider that belief?

You might, at this point, object to the wording of my prompts or to the elements they emphasize. Fine—come up with your own prompts. For it to be a legitimate part of a stress test, though, the prompt should make it clear that you don’t want AI to feel its way to a conclusion and don’t want it simply to report the consensus view. You want it to engage in evidence-based reasoning* to reach its conclusion. And if you worry that ChatGPT leans one way or another with respect to the abortion debate, try the same prompt on a rival chatbot.

Don’t get me wrong: If you have ready access to an “opposite Thinker,” use them to stress-test your beliefs. But if this isn’t the case, AI chatbots are a pragmatic alternative. Realize, too, that someone who engages in this form of stress testing is likely to come away with a better understanding of a controversial issue than 99% of humanity. Additionally, they are likely to come away less confident of their belief and maybe holding a more nuanced position as well—which, from a Thinker’s point of view, is a desirable outcome.

---

*Am I suggesting that AI can reason the way people do? It is a difficult question to answer, in part because it’s unclear what is involved in reasoning. One thing that is clear, though, is that AI demonstrates a more consistent form of “reasoning” than those humans who feel rather than think their way to conclusions. Whether AI can also reason better than Thinkers remains to be seen.