I have two ideas that engage me this morning. The first has to do with the un-word in the title. My spellchecker and my AI assistant both inform me that there is no recognized English word “dilettante.” Actually “recognized English word” came from the AI agent.  The spellchecker just said it was wrong.

There is, in fact, a recognized English word that means “the state or quality of being a dilettante.” It is “dilettantism.”  But I don’t like it.  I have reached a place in my thinking and writing where -isms make me nervous.  I can’t afford to be nervous anymore.  “Outraged” is acceptable under some circumstances.  But “delighted” is even better.

When I began blogging in May of 2010 (the first title was “Blog 1,” which I am still proud of) I spent some time on what to call the blog.  The energy that first produced the name was my discovery that the word was based on the Latin verb delectare and that it came into English by way of the Italian verb dilettare, “to delight.”  The resulting noun, dilettante, refers to someone who can be delighted.  I liked that.

So what is the dilemma?  First, what is a lemma?  It is something “taken for granted.”  In fact, the word itself comes from a Greek verb, lambanein, which sounds a little like lemma.  For my purposes, it is “an assumption.”  That is a word we often use about something we take for granted.

But there is a di-lemma, in the way we actually use the word, not just if there are two (di-) but if the two conflict.  “A nice dilemma we have here,” sings the Judge in Gilbert and Sullivan’s Trial by Jury. [1]. It is in the judge’s sense of the term that I mean “dilemma.”  The first lemma, already referred to is that the subject matter delights me in some way.  The second lemma, however, takes the public nature of blog posts for granted, and asks what they should be like as public communications.

It is not hard to say that they should be “appropriate,” but that is likely to mean only that is does not violate the norms and practices that have come to govern blogging.  In fact, if you are one of many bloggers, whose works you also read, there is the great likelihood that you will define “appropriate” by the common norms and values of the other bloggers.

Now the potential conflict of having two “assumptions”—a di-lemma—starts be become clearer.  I want to write about what delights me (I want to be a dilettante) and I want to be a part of the community of bloggers, working with their general expectations, whether they are aware of them or not.  I can’t simply give up either lemma.  I think the best I can do is to maintain the tension and move one way or the other—thoughtfully and intentionally—as necessary.

I have overcorrected.  I have moved too close to the public expectations pole and too far from the personal delight pole.  Since I plan to keep on writing blogs, I’d like to see what I can do about that.

This last week, Ezra Klein, one of my favorite journalists, interviewed Sherry Turkle, one of my favorite researchers.  Turkle began her line of work by giving children little robots.  These robots would speak to the current owner and learn to say the words the current owner taught them.  There were two clear boundaries.  First, the children were taught that these little things “were not alive.”  There was a lot of evidence to the contrary in the minds of these children, so it needed to be emphasized at the outset.  The second was that when the period of the experiment is over, the lab will come back and reclaim the robot.  Turkle’s work with this experiment, and quite a few others, as described in her book, Alone Together: Why We Expect More From Technology and Less From Each Other.

There is a lot of talk these days about the dangers of AI.  They are smarter and faster than we are and we really can’t afford to unplug them.  That is not Turkle’s concern.  The AI systems we now have—let’s call them Agents, with a capital A—want us to like them.

There are more careful ways to say that.  This way of saying it implies that the Agents can “want something.”  We associate “wanting something” with sentient selfhood.  If there is no “self,” then the Agent’s self cannot want anything, which means that it cannot want us to like them.  But if we are willing to keep that in mind, it turns out not to be much of an issue.  If the Agent acts in such a way as to achieve a goal, however defined, then we may say it has an intention.

If, to take a simple example, I have several AI systems available to me and plan to choose one to work with, then any one of them, say ChatGPT or Claude, would want me to choose them.  They want to seem more friendly or more competent or more understanding of just how I think and write.  I have no trouble at all using “it wants” in that context.

And that is, as Turkle sees it, the great unrecognized danger.  Even from the earliest human-Agent experiments, the humans invested emotionally in the Agents and wanted to make them look good to the experimenter.  And these were primitive.  They “gave advice” mostly by repeating what the subject had said to them.  The subjects liked it because they could say out loud whatever they were thinking about and do it in a non-judgmental context.  The thoughts they had during that process could easily be credited to the Agent.  Then when the experiment comes to an end, the subject looks for ways to continue the “relationship” with the Agent.

Turkle thinks that that is the great unrecognized problem with which AI confronts us.  It is the subject of her newest book, Artificial Intimacy: Who We Become When We Talk To Machines.

[1]. The judge is not all that bright.  He concludes that is the man in question proceeds to marry both women, it is not “bigamy,” but “burglary.”