Suspicious Adapters

Show Outline with Links

Introduction

We have already considered the great threat posed by the interactive impressiveness of AI, which will be the temptation to what I called AI-dolatry.

In our treatment of that danger, we noted that there are two kinds of idols. There are idols that must simply be destroyed, and that’s that, and then there are the idols which must merely be demoted. That statute of Chemosh on your mantelpiece is something that simply has to go. The savings account that you thought about far too fondly and frequently is an idol that needs to be demoted.

If AI is something that falls in the second category, as I believe it is, then we need to get some handles on it. How do we practically keep from promoting it wrongly, and how do we go about demoting it if we have already sinned in this way?

While it may be lawful to keep the bottle around, how do we get the genie back into it?

Define It First

One of the first challenges we face is that of defining it too narrowly, leaving some things out and therefore unguarded. We need to be careful so that we know what we are talking about exactly. Your rock collection may not be AI, but after that things can get blurry and confusing.

One of the things that makes this discussion so zesty is that definitions vary. I expect some letters on this, but work with me. I am using a stipulated definition, gathered from various sources.

So what is a broad definition of AI? That would include any digital machine that performs a task that would normally require human intelligence to perform—say a thermostat that is programmed to adjust to changing outside temperatures, or the navigation app on your phone. It would not include your antique grandfather clock, which is analog and mechanical. These would be examples of ANI (artificial narrow intelligence), but they still fit under the broad definition of AI. ANI applications do one specific thing really well.

So the broad definition of AI includes the narrow applications. When the Internet of things becomes a thing, or when the singularity arrives, this broad definition would include anything that could be picked up and incorporated into the Borg.

The thing that is attracting much of the anxious attention would be AGI (artificial general intelligence). Examples of this would be those tools that can render Psalm 24 into common meter, draw you a picture of Donald Duck smoking a cigar in the Oval Office, or write you a sermon with three alliterative points on Ephesians 2:8-10. So the broad definition of AI therefore includes the massive AI tools, like the Large Language Model applications, and it also includes very narrow and specific AI tools.

Now for Christians to address this challenge by saying that the narrow tools are okay while the general tools are not is going to turn out to be way too simplistic. While a navigation app can seem like a strict, narrowly circumscribed business, it is just a matter of time before it all gets more sophisticated and broader—predicting traffic delays, for example, which is not the same thing as spotting a traffic jam. The boundaries between ANI and AGI are going to become increasingly porous, in other words.

And so what are some of the principles we will need to navigate all of this?

What Used to Be Bad Is Still Bad

Some of the very real problems that AI has handed to us are actually ancient problems, not new at all. And a difficulty related to this is that many of those Christians who are reluctant or hesitant about the developments in the world of AI are opposing it as though these old problems are new problems, and this tends to discredit their concerns.

Many of the practical problems presented by AI—in academia, for example—are simply problems that are centuries old, only now they are moving around at a high rate of speed. Some young men used to drive their carriages too fast, and now some young men drive their pick-up trucks too fast. The speed is new, but the problem isn’t.  

Take, for example, plagiarism or other forms of cheating. People have been cribbing the work of others long before the first computer was developed. Grok has certainly made it easier not to do the work yourself, but we have always had people who did not do the work themselves. If AI writes your sermon for you, that’s a sin. But if you were a nineteenth century preacher preaching a Spurgeon sermon you found as your own, it is the same sin. You are taking credit for something you did not do. It is lying.

In the old days, if a ghost writer wrote eighty percent of the book, and his name was not on the cover, then the author was a fraud. That is how you can find yourself with authors who had written more books than they had read. If the ghost writer’s name was on there, there was no fraud . . . for the buyer had his fair warning.

Another old problem, disguised as a new one, has to do with AI and the work of creatives—graphics, movies, and so forth. The charge is that if the machine is doing all the creative work, this is destroying the very idea of creativity. But did the invention of the camera flatten representational art? We used to have good painters and poor painters, and now with high quality cameras, every photographer became a high quality realist. The camera does all the work. But then you look at photographs by Ansel Adams, and you realize that this is not true. Put two men in front of the same moon rise, with the same kind of camera, and if one of them is Adams, you will get two different pictures. In a similar way, are we likely to discover that certain creatives are geniuses when it comes to writing prompts? 

Here would be a third problem that is not a new problem. John Piper famously argued against someone who would attempt to pray a devotional prayer that had been composed by AI, arguing that it would not be true devotion at all. But wouldn’t the same thing be the case with someone using The Valley of Vision or the Book of Common Prayer? This debate about set prayers is an old debate. If someone gave a prompt asking Grok to compose a prayer in the manner of Richard Sibbes, and then gave the results to a trusting friend, telling him that it really was Sibbes, could the friend conscientiously pray that prayer? The same way he had prayed prayers composed by Cranmer? Sure. Some people use set prayers as a way of not having to do any actual praying themselves, but others have used set prayers with great spiritual profit. With the Valley of Vision prayers, done right, the prayers are lifted up twice—once by the man who first composed the prayer, and secondly by a man centuries later who offers it up intelligently and sincerely. And if it was a machined prayer, it only got prayed once—when the believer first prayed it.

So below is an example of a prayer composed by Grok, responding to the prompt “please compose a collect prayer, in the manner of the Book of Common Prayer, thanking God for rain, sunshine, and a bountiful harvest.” Now I confess that this gives me the creeps, just like it does John Piper, but I am curious as to why. Do we want to say that set prayers are unlawful unless they come to us pre-prayed?

However we solve this problem . . . it remains an old problem.

If you get the creeps, listen to your creeps. But the creeps need to be supplemented with arguments. My father was born in 1927, and a few years ago I was getting him set up on his iPad. My father was a Nebraskan, and so I said something like “show me Omaha,” and a map of Omaha popped up on the screen. He had a one-word response. “Demons,” he said.

And that reminds me of the time that Tolkien wouldn’t use a reel-to-reel tape recorder until after he had recited the Lord’s Prayer in Gaelic in order to deal with the demons first. The creeps can be healthy, but they will need reinforcements.

Another old problem will be the issue of employment, but it will be an old problem with a new wrinkle. In the old days, new technology would throw practitioners of the old technology out of work, and they would then have to find a new occupation in order to provide for themselves. The buggy whip makers had to learn how to code. But the promises being made for the economic impact of AI present a far more serious situation. The old jobs will become obsolete, and so will take away the living of millions of workers. But the promises also say that this will result in a massive increase of wealth, such that those who are thrown out of work will not need to find another occupation. They will be the beneficiaries of the new universal income policies. So they will not have had their living taken away . . . just their reason for living. When we are promised the contented life of a satiated larva in the Hive, for some of us the allure is missing.

Helpful Uses of AI

So let me outline some of the uses of AI that I think are fully appropriate, along with a few cautions.

AI is a very slick search engine. You need to be careful with this one though. When you use google to search, it gives you a column of links, and the clear implication is that you need to evaluate the claims made on your own. “These are places that talk about your question, and now you are on your own.” But AI responds to you with an authoritative voice, telling you that this is the way it is. But it got that info by scrubbing the entire web, and there is a lot of bum dope out there. In addition, AI doesn’t just search for things, it just flat makes things up. More than one attorney has gotten busted for submitting a brief with non-existent court cases in it. AI is not a truth machine, but rather a consensus machine, so take care.

There are two ways to deal with this. One is to search for things you already kind of know, but you need the specifics. You remember that C.S. Lewis wrote an essay that had a liberal lady judge in it. What was that essay? When the answer comes up you recognize it, and can say “yes, that’s the one.” A recent example was when I remembered there was French phrase out there for “staircase wit,” but didn’t know the exact phrase. When AI gave me the phrase, I recognized that it was the one I was after. The confirmation is built in on the basis of your current bank of knowledge.

Another way is this—my Logos Bible Software has a feature called Study Assistant, but it doesn’t scrub the web for answers. Rather, it scrubs the library inside Logos, a library I already trust. The answer comes up, and the specific claims are footnoted so I can go check. This kind of thing is just great.

A friend of mine is a Wycliffe Bible translator, and over the course of decades he completed a translation of the New Testament. That translation work serves as the resource for an AI program developed by Wycliffe—one that is enabling him to get the rough draft of Old Testament books far more quickly. All the checking and cross-checking still has to happen, and the finished product will be the work of translators, not a machine. But still.

A songwriter and composer can use AI to produce demo tapes. Nobody wants to go to a concert with a computer in the spotlight, but a songwriter can have a far better chance to showcase his songwriting abilities than he did before.

Blinking Yellow

The fact that using a technology is lawful does not mean that we understand all the ramifications of using it. Inventions transform human society—the automobile and telephone to take two early examples. We are on the threshold of something far more massive than that, and the two great challenges on the human level will be the enthusiastic early adapters (“what could go wrong?”) and the Luddites (“nothing about this can possibly be good”)

Aside from the issue of false worship (which I have already argued is the greatest threat), there are four major practical threats—let me call them social, cultural or political threats. I mention them here, not to give a complete treatment of them, but simply to acknowledge the reality of them. And I hope to be able to get to these and other issues in future installments.

There is the threat of a refusal to shut down . . . if any of the anticipated problems start to get out of hand, where is the breaker we could throw? If there are unanticipated problems, where is the breaker we could throw?

There is the threat to meaningful employment for humans . . . new technologies have always introduced employment dislocations, but if that happens at 100X, the word dislocation doesn’t seem nearly strong enough.

There is the threat of human expertise atrophying, such that meaningful review of engineering plans, say, becomes impossible . . .

There is the threat of old-fashioned tyranny, with tyrants wielding their power through the instrumentality of AI. Think of it as a secularist interdict—a region of the country is getting restive and uppity, and so the president-for-life orders everybody’s cars and stoves and computers turned off.

Your card doesn’t work at the ATM because of that meme you shared. You renegade.