Hello, loves! Tout le monde déteste l’IA.

My moral principles were called into question yesterday. After cooling down, I think I should, well, think about them. Not sure I like the answer.

I had tooted this little cri de cœur a bit ago:

Damn. DAMN. I was researching FSMs, as one does. The cursed AI overview offered an example. Very trivial, nothing like what I need.

It has that question area. I told it to make a change in the direction I wanted. It provided a better example. Two steps later, it offered a style of solution that I would not have thought of, that is better than what I would do in that situation.

It’s not what I need, but as a step toward what I need but it’s useful.

I am fascinated, and I do not like that.

My brother Hill called me out, in a fashion that felt to me like an attack. We remained civil and he closed with “friends call each other out”. I replied “OK”, still somewhat bent out of shape.

Hill famously (private communication) says that he does and says things to bring about the changes he wants to see in the world. If I generously assume that he wasn’t just lashing out because he still can’t find that thing he lost, I owe it to him, and myself, to figure out what change he might like to see. This article tries to do that.

Hill also said “ya gotta own it”. So, maybe in what I tooted above, I wasn’t owning it.

Owning It

I dislike them intensely

You can click the LLM-AI link on this article, or in Site Categories, to find what I’ve written about the LLM-AI thing, but I’ll try to summarize here and bring my thoughts up to date.

I dislike the “AI” thing intensely, and mostly, I think, on moral grounds, including but not limited to:

The investment gold rush is funneling incredible amounts of money into the hands of super-capitalists, the oligarchy who already have many times their fair share of the world’s good things.

The emerging “AI” tools are being used to displace ordinary people from their jobs, increasing the wealth divide, and quite often literally ruining people’s lives.

The creation of an LLM consumes immense resources in terms of power and water, producing lots of heat that the world really doesn’t need.

The results produced by an LLM-AI are quite often rather good, in terms of aptness and correctness. And they are often subtly wrong. The mistakes they make are expressed just as confidently as the correct answers. Checking results, as any programmer knows, is generally much more difficult than creating correct results. LLM-AI results will almost certainly increase the defect content of software, which already has more than enough defects.

The LLM-AI does not understand anything. An individual who just checks, or blindly accepts LLM-AI results will understand far less than they would had they worked out the problem themselves. Software design quality is a function of understanding of the domain and the program. Use of LLM-AI will result in far less of the critical ingredient, understanding.

This article is one of my better expressions of objections to LLM-AI

I sometimes use the “AI Overview” in Google

In my defense, I tried Kagi and Duck Duck Go, with the “AI” turned off, and found them far less useful than even vanilla Google. I went back to Google. Therefore, I grant, I am somehow undermining my principles, insofar as somehow, using Google funnels more money to the rich. Mea culpa.

And I did not turn off the “AI Overview”, although I believe there is a way to do it. Therefore, I grant, I am consuming resources and generating entropy that I could avoid. Mea culpa.

And, as my aforementioned toot indicated, I sometimes read that overview, and find it useful. Even worse, in the case in hand, I typed follow-up queries into its little window, and actually got a useful answer that I am certain I would not have through of. Mea maxima culpa.

Morality

I am rather well-educated in conventional Christian morality, having attended parochial schools, a Catholic high school, and two Catholic colleges, the first one (even worse) a Jesuit institution. Also, somewhere about the middle of all that, I lost any faith in the standard Christian god. The story doesn’t hang together for me.

But faith in God isn’t required for morality. In my view, it might be harmful to morality, because not doing evil for fear that one will be seen and punished isn’t really being moral. Not doing evil because it’s evil, that’s being moral.

My sense of morality is well described in Wicca: “An it harm none, do as thou wilt”. To me, harming another is evil. (And, it probably follows, harming oneself is also evil.)

However, the Rede is too demanding. We cannot move in the world without harm to some other. We cannot even die without harm to some other. It seems to me, therefore, that we should move in the world so as to do more good than harm.

And there’s the rub:

Utilitarianism

The notion of doing more good than harm asks us to balance good and harm. That’s a dangerous notion, even to a naif like me. It can be pretty easy to rationalize that good things for me are much better than any corresponding harm to the other. And, whether I learned it from the Jebbies or not, I am really quite good at rationalizing.

But I’m stuck with it, because I can think of few if any potentially evil actions that I might take that couldn’t be justified in some situation, however unlikely.

So … our purpose here comes down to examining how I rationalize reading and prompting the “AI Overview”, in the light of my pretty clear list of evils of LLM_AI.

Harms of the “AI Overview”

The resource usage of an “AI Overview” seems to be estimated as about ten times the cost of a keyword search, but still tiny, about the cost of a few seconds of watching television or running one’s microwave or laptop, and literally just a few drops of water. Of course, when we multiply that by billions of queries we do get truly massive figures.

The dollar cost is estimated to be less than $0.00005. A minuscule fraction of one cent. Of course, multiplied by billions, you’re talking real money.

So. Slippery slope here, isn’t it? My queries about FSMs or whatever it was consumed about one milliliter of water, while I melted about 50 ml in my iced chai, and ultimately drank it.

It’s pretty darn easy to say that learning a new technique that I might use and write about is worth at least $0.00005 added onto the cost of my google queries. On the other hand, I could turn off the summary in normal queries and only use it intentionally. That would save the planet a bit and would focus my attention more on what’s going on.

Where I’m At

I’m trying, honestly, to feel guilty about this, and I’m just not getting there. I don’t do vibe coding, and intend not to. I don’t have the AI option turned on in PyCharm, and intend not to.

And I find the “AI Overview” to be useful enough that, despite the peer pressure so far, I’m not going to turn it off.

First medicinal wine from a teaspoon, then beer from a bottle, an’ the next thing you know, your son is stuffing tokens into ChapGPT and listening to some out-town jasper hearin him tell about vibe coding, not wholesome pair programming, no! But coding where they just set and gawk.

It’s a slippery slope, and I am on it. I am resolved not to slip lower, and believe (or tell myself I believe) that this far down is OK.

I’m owning it. I’m doing this. Is there harm? Yes. Is there good? Yes. Are they in proper balance? I’m saying yes, with my eyes open. But can I trust my eyes? Maybe not. Could I do even less harm? Yes. Would I lose more benefit than the cost? I’m telling myself so.

I could be wrong. I may well be wrong. I won’t stop thinking.

Thanks, Hill. Sorry to let you down. Sorry to let myself down.