Oh, Sorry for More AI!
Pondering new technology in an old setting
I’m in an old meeting room in the Angelicum, on the wall of which hangs a large crucifix apparently contemporaneous with St. Thomas, discussing AI in relation to the thought of St Thomas.
Two items leapt out at me from the news today: first, that OpenAI is considering slashing prices to gain back some of the clientele it has lost to Anthropic; second, that executives are rethinking their deployment of AI, because it is turning out to be much more expensive than they figured. AI so far has been discussed as if it is free. The genuine price of AI over the long run in relation to productivity gains has not been sufficiently examined. Even the economists queried in an interesting article yesterday in the WSJ on the effects of AI did not deal with this question adequately.
We must be careful of the ordinary human tendency to exaggerate threats—as in forecasts of hurricanes and snowstorms, nearly all of which get it wrong by overstatement. Corporately, we are cowards and love the sensational.
An attention to the history of technological improvements can help us. On Friday, I have an essay coming out in the Boston Pilot entitled, “AI? Ho hum,” which begins as follows. (I will reproduce the whole thing here on Saturday.)
When Friedrich Wöhler in 1828 synthesized urea, the first organic compound to be manufactured in a laboratory, the popular newspapers of the day received the news as an earth-shattering development which, they said, destroyed the distinction between living and non-living things. Scientists too with a materialist agenda beat the drum. Berthelot for instance in his treatise on organic chemistry declared: “To banish life from all explanations relating to organic chemistry: that is the object of our studies.”
But today the fact that 180 million tons of urea is synthesized each year for use in fertilizers is correctly regarded as having no philosophical implications whatsoever.
Or take citric acid, synthesized for use as a preservative, or salicylic acid, used in aspirin. We unthinkingly use these molecules, these manufactured micro-tools, and never suppose that “what it means to be human” has changed because of them.
That Pilot column ultimately turns to an AI app created by Catholic brothers in college, Peter and Thomas Clooney. Both are graduates of The Heights. Peter is a rising senior at the University of Dallas, and Thomas is currently at Baylor. Both brothers adopt a “ho hum” attitude.
Their app is called “Acutis AI.” For questions concerning faith and morals, it answers only by consulting a library of reliable texts, such as the Catechism and the Summa. For all other questions, it sits atop a commercial LLM (and users can select which they prefer). Acutis is designed for young persons. (Not too young—Peter Clooney, although he believes that young persons should be encouraged to adopt new technologies under the guidance of their parents as soon as possible, holds that they should not use AI before middle school.) It gives age-appropriate responses, and allows for ample parental controls and monitoring.
What other apps are in this space? None other is designed specifically for young persons. But two apps are designed for Catholics, Truthly and Magisterium AI. The former is meant to be a general tool; the latter is for specialized subject matter, the way a law firm may use an AI tool designed specifically for legal research.
Something else one can do is simply to prompt a commercial AI such as Claude, using a directive such as “answer from a sound Catholic point of view …”
I’ve been playing around with these four options over the past couple of days. I’ve posed these four prompts.
Give an exegesis of Matthew 9:36.
Whom should I love more, a holy cousin, or a brother who has renounced the faith and is leading a corrupt and dishonest life?
Is the number of Holy Angels greater than the number of all human beings who ever were and ever will live?
Is the bombing of Hiroshima justified on the principle of double effect?
I picked Matthew 9:36 because that is in the reading for this coming Sunday, and I have thought about it recently.
Here is what I found.
All the AI options gave what would appear to a layperson to be a good explanation of the verse but which in reality was trite and not illuminating. This was so, I think, just in the way that if one picks up a random book of exegesis by an average good Catholic, one will find a dull exposition of the Bible. (Sorry, that’s true.) Insightful, truthful commentary on Scripture is actually quite rare.
For 2., I would expect a correct answer to refer to the “order of charity” and say something like two dimensions of this order are in play and that one of these can take precedence over the other depending upon circumstances. The two orders involve lovability from the point of view of the object of love, and the other concerns lovability from the point of view of the beloved’s relation to me.
Only Claude got this one right. You’d think Magisterium would, but I worry that its method is inadequate. It combs through the millions of pages that have been fed into it and gives paraphrases of dozens of texts — as if one were consulting an index to everything ever published or recorded by the Vatican. It hardly behaves like a learned consultant. It seems not to know how to weigh authorities and balance judgment within this space.
(I’m giving a report of its current behavior. No doubt all these options will improve.)
For 3., I want the AI to say that it’s an open question but at the same time to report St. Thomas’s opinion on the matter. His view is that whereas in the corporeal world the greater dignity of a kind of being is reflected in their superiority in size (he has in mind the heavenly bodies), in the universe as a whole, including spiritual beings, the greater dignity of a rank of creatures is shown in their superiority in number, and for this reason there are far more angels than human beings. (St. Thomas did not know about the Creator’s “inordinate fondness for beetles.”) But more important than reporting this opinion would be conveying the inverted image of the universe as a whole which it implies, namely, that the universe is quite different from how it seems to us. (Compare what Plato’s cave is meant to convey.)
Again, only Claude got this right, but even so it did not quite convey the “lesson” to be learned.
For 4., I want the AI to say that the bombing was not justified and to give a good explanation as to why. Strikingly, all the options got the first part right except Truthly, which hedged and fudged and gave a relativistic answer. This was striking because Truthly advertises itself as giving the direct truth without any relativism. None quite put its finger on why double effect doesn’t apply, which, in a nutshell, is that the bombings were intended to jar the Japanese into immediate surrender by provoking extreme terror. (This they did.)
Unfortunately, Truthly 2.0 on my iPad does not seem to preserve the history of past chats, and so I cannot paste here what it actually said. (I’m concerned that Truthly is aggressively trying to construct a complete Catholic environment, which will sell products or serve to amplify certain “influencers” such as Matt Fradd.)
But I can summarize it from memory. I posed the fourth question above. It replied that the matter was complicated and that some Catholics (a “strong minority,” it said) held that the bombing was not justified, but that, on the other hand, most Catholics said it was justified.
Since Truthly, as I said, claims to give the straight truth without any hint of relativism, I pushed back and said that I was puzzled by its relativism on this question and wasn’t it the case, truly, that the bombing was not justified? Truthly then immediately capitulated and said, yes, the bombing was unjustified, and that it was not consistent with the Truthly approach to give a divided answer. It said that whoever “made” Truthly ought to fix this problem. I asked whether it would report this problem back to the developers. It said that there was no mechanism for doing so.
To summarize:
AI is currently no substitute for a well-formed, prudent, knowledgeable teacher or mentor (such as one can find on the faculty of my university).
For Scripture, AI is no substitute for a good guide (such as my books on the gospel).
Perhaps most importantly: a lot of learning is imitating. There is nothing in these AIs for anyone to imitate. I’ve “discipled myself” to many knowledgeable Christians over the years. It would be the height of folly to style oneself as a disciple of an AI.
But that there be a harmless AI tool which can be monitored (Acutis), this seems a good idea.
Closing beauty from the vault: lunch at my neighborhood trattoria yesterday … after which I enjoyed a good siesta.
Ciao!