On May 1, the website Ask.com shut down for good after more than 30 years. Originally known as “Ask Jeeves,” the site was a search engine that invited users to enter search prompts in the form of a question. Long before ChatGPT, Ask.com understood that people liked the experience of typing questions and immediately getting answers.

In reality, though, Ask.com did not provide answers, so much as possible answers. It was a search engine, after all, and it responded to questions with lists of seemingly relevant hyperlinks. It’s perhaps no surprise that such a site could not compete in the era of large language models like ChatGPT that provide concise, confident, and conversational responses to even the most complex questions, all without the need to scroll through a long list of search results.

I was reminded of this website recently when I was reading about the limits of large language models in healthcare settings. In theory, it could be incredibly helpful if physicians, nurses, or even patients could simply type symptoms into something like ChatGPT and immediately be given a correct diagnosis.

A recent study in JAMA Network Open found that commercial large language models are pretty good at doing just that. But there was a problem. A big one. While large language models could generate a final diagnosis with an impressive degree of accuracy, they did a decidedly poor job at differential diagnosis. As anyone who watches The Pitt knows, differential diagnosis is the process of coming up with all of the possible causes of a patient’s ailments and then systematically narrowing down the list as new information becomes available. Large language models aren’t good at this, the study found, because they tend to “collapse prematurely onto single answers.”

This is bad news because, as the authors noted, a crucial part of being a good physician is being able to “preserve uncertainty.” While first impressions might be right much of the time, the discipline of differential diagnosis ensures doctors don’t latch onto the wrong answer simply because it’s the most straightforward.

The incompatibility—for now, at least—between large language models and mindful diagnostic procedures is an example of an increasingly prevalent epistemological challenge. In a world saturated with information, answers are decreasing in value. Finding answers is incredibly fast and easy. What’s harder—and more valuable—is understanding whether and why something is truly right and truly true. 

A similar phenomenon is occurring in the realm of religion. As early as 2023, tech companies were marketing the opportunity to chat with AI “versions” of God and Jesus and Buddha and Apollo. Part of me wants to dismiss such ventures as stupid and vapid and not worth spilling any words. However, if we are being honest, much of cultural Christianity—and many of our churches—offer a model of the Christian life that functions as little more than a self-assured answer-bot. 

• What do I do about this crisis in my life? Pray.
• What if things don’t work out? Don’t worry! God’s got this!
• Will I go to Heaven? Jesus died for your sins! So… probably?

All of those responses are perfectly defensible so far as they go, but the older I get, the less sufficient such replies feel. I find myself unsatisfied with simple spiritual answers, even if they are true. I’m much more interested in learning from people who have spent long nights of the soul contemplating and wrestling with such questions. Weighty, hard-won wisdom is more valuable to me, even if such wisdom never lands on a clear answer. 

It’s a lot like differential diagnosis. Who among us would feel satisfied if we went to the specialist seeking help with a long, painful illness only to have the doctor jump to the most obvious diagnosis without taking the time to ask any questions? The doctor might well be correct. Still, many of us would spend the next several hours or days doing our own research in an attempt to convince ourselves that the doctor was right. Instead, we want our doctors to ask a lot of questions, take the time to process our responses, and explain why the most obvious answer is or is not correct. 

The same is true of the church. Easy answers might suffice for a while; they might even feel comforting. But if that is all the church has to offer, Christians will eventually start to lose faith in the church. They will second-guess the church and begin to look elsewhere. 

The challenge also presents an opportunity, though. If the church can be a place of depth, a place where easy answers are beside the point and where soul-searching questions are met with thoughtful discernment, that church will increasingly stand out.

The world doesn’t need ChurchGPT. It needs something closer to Ask.com, a place that invites people to ask complicated questions, and then has the patience to stand by them as they slog through a long list of potential answers.