Overlapping

A rather full day - first session of "Varieties of Religious Experience," followed by a five hour "Faculty Retreat on AI and Higher Education." Since I was running out of the adrenaline that has somehow kept jetlag at bay all week, I smushed them together.

More specifically, so as not to squander one of our fifteen weekly sessions, "Varieties" started with William James' essay "The Will to Believe."  (Its shorter precursor did too.) It's a good introduction to Jamesian matters and (almost) short enough to cover in one class. And, I realized rereading it, it's the text of a lecture given to college students, the philosophy clubs at Yale and Brown, and begins with James' challenging these student audiences to hear him out on subjects his own students won't follow him on. James' argument is a "justification of faith," a defense of the value of believing rather than suspending belief even where evidence is insufficient, at least in the cases where the decision is "momentous" and "forced," and notably with regard to moral and religious issues.

I have long defended to my own students the lawfulness of voluntarily adopted faith; but as soon as they have got well imbued with the logical spirit, they have as a rule refused to admit my contention to be lawful, philosophically, even though in point of fact they were personally all the time chock-full of of some faith or other themselves. (2)

James' students, inspired by the apparently high-minded idea that "It is wrong always, everywhere, and for every one, to believe anything upon insufficient evidence" (8), are naive about the ways the human mind works (it's always driven by its passions) as well as the reality of the universe. Deciding not to believe is itself a decision, James argues, and there are worse things than being duped, since one can correct. Sometimes believing is good, even existentially necessary, for the believer. And there may even be cases where faith in a fact can help create the fact (25)!

A stranger set of arguments than many suspect, ranging from epistemology through psychology to something like a theory of religion. 

Religions differ so much in their accidents that in discussing the religious question we must make it very generic and broad. ... [R]eligion says essentially two things. ¶ First, she says that the best things are the more eternal things, the overlapping things, the things in the universe that throw the last stone, so to speak, and say the final word ... ¶ The second affirmation of religion is that we are better off even now if we believe her first affirmation to be true. (24-25)

The Varieties of Religious Experience will comes at religion in a different way, as we'll be seeing, more "scientific" and distanced. But it's animated by the same sense that the religious question is not merely academic but deeply personal. Ultimately its terms are as evocatively vague as James' here, but that's part of the Jamesian thing. It might seem, I said, "squishy" to refer not to "eternal" but "more eternal things" (so things can be more or less eternal?!) and what on earth does "overlapping" mean? But I confessed, too, that this squishiness, which can frustrate the more analytical, is part of what I love about James.

In any case, I had to run to our faculty retreat, where the university president was just wrapping up some introductory reflections which apparently centered on the question whether artificial intelligence is really artificial. Our dean added that it seemed an open question whether or not it was intelligence, too.

There followed several presentations by local faculty, a zoom presentation by one of the authors of Teaching with AI, and some invitations to use AI ourselves. (For some of my colleagues, perhaps amazingly, this was their very first time!) A philosopher colleague invited us to feed a scannable article to Mistral's Le Chat (one of six main forms of generative AI, of which ChatGPT is the most famous but not necessarily best), first asking for a summary and assessing it, and then asking it questions until we succeeded in getting it to report something false. 

I tried with "The Will to Believe":

Break down the main claims in William James' "The Will to Believe" and suggest more accessible language for these claims.

It did very well with the first half of the prompt. For the second, well...

I guess I succeeded at the prompt I'd been given. James thinks the "will to believe" pertains only in limited cases (although "forced" and "momentous" are pretty squishy categories too), offering nothing like this license to believe whatever makes you feel good. But I suppose this is how James' pragmatism sounds to a lot of people who don't get the existential undercurrents. In this bowdlerized version at least, it makes me want to agree with James' students that such an argument is more than a little reckless. 

But I wasn't with students. I was with AI, as a colleague was lecturing about the different ways large language models synthesize data. "Hallucinations," where AI seamlessly segues from plausible to bizarre claims, arise when they're asked questions about things which have not been addressed in the materials they're trained on. As they train farther (and as more people feed them their own prompts and responses), there are fewer and fewer of them. They're driven by the probabilities of proximities of words and concepts. 

How can William James' "The Will to Believe" explain the mechanisms and ethics of AI?

It gave me an answer, of course. 

Was this spot-on or hallucinatory? Suddenly my jet lag hit.