Conditioners and Millstones

What are the conditioners telling the children?

Elon Musk announced this week that he was relocating SpaceX and X, the company formerly known as Twitter, to Texas. The reason for this move is Gavin Newsom’s recent enactment of a law that, for all intents and purposes, makes children who attend any California public school the property of the state. In practice, the public school system is required to withhold, from the parents, information critical to the well-being of the child. The only possible reason for this is to blind the parents to their child’s difficulties and preclude parental influence from being brought to bear in a direction that is at odds with the agenda of the education establishment. This state of affairs takes de facto decision-making authority regarding child welfare out of the hands of the parents, and puts it into the hands of public educators.

I have raised questions in other posts about whether some sins are worse than others. Jesus of Nazareth made some rather pointed remarks, notably lacking in winsomeness, regarding people who cause spiritual harm to children. Those people, he said, would be better off being forcibly drowned, having had a large millstone hung around their neck, than suffering the punishment that actually awaits them. If this is not a comparative statement then it is nothing at all. From Jesus’ perspective, spiritually harming children is an act that marks someone as a target for an unusually severe form of divine retribution, one that is apparently above and beyond more run-of-the-mill punishments. Adopting the position that public educators, as a general matter, have a greater concern for the welfare of children than do those same children’s parents, is a really breathtaking level of usurpation - and this in service to facilitating the gender transitioning of children. The spiritual harm this will inevitably do to the children, to say nothing of harm to the parents, is not something that is going to go well for the officious bureaucrats of the state, when the time comes.

Newsom’s actions got my mind noodling on the increased prevalence of adversarial policies being adopted by the state and other institutions, directed against the prerogatives of parents. And when you observe the general insanity on college campuses, you have to ask yourself, “how did these kids get to this vaunted moment in their lives, having never become better thinkers than they are now showing themselves to be?”

C.S. Lewis writes, in The Abolition of Man, about an elite cadre of “conditioners”, a loose confederation of teachers and thinkers of deep thoughts, who advocate an approach to cultivating understanding, so called, that has the effect of crippling rather than ennobling. How is it that so many products of modern, intensely managed childhoods seem to have such a tenuous grasp of reality? C.S. Lewis might like a word.

Ultimately, observing the uniformity of dumb ideas being held by many of these students, I began to wonder whether popular children’s television programming might be one of the things conditioning and homogenizing the thinking of these young minds, along the lines of C.S. Lewis’ “conditioners”. What are the ideas contained within children’s television that are being quietly normalized, at the scale of entire cultures, in the minds of children who watch TV? And how is a parent or grandparent supposed to develop an informed point of view regarding the prevalence of, not merely the egregious and obvious things, but the more subtly pernicious ideas that are quietly slipstreamed into children’s storylines? The challenge of getting a grasp on these issues is compounded by the velocity with which new episodes arrive. The sheer volume of new shows necessarily overwhelms any parental defense. It’s like trying to fend off a drone swarm without having anything like the intellectual equivalent or capacity of an Iron Dome.

I was recently watching one of the later Star Wars movies with my 8-yr-old, and after some shoot-em-up battle or other, in which the bad guys were defeated (I confess I lost interest in Star Wars after The Return of the Jedi) all of the combatants returned to base. There was a massive celebration as everyone returned from battle, many much worse for the wear. The camera panned across the chaos and delirium, as those who survived the battle greeted one another, and momentarily passing through frame I observed a clip of two women embraced in a passionate kiss. Nothing was said about this in the dialog. It was just a fleeting image presented to all the viewers without comment. It wasn’t even part of the storyline. If you blinked you missed it. But it was there, and for some number of children it planted a seed of normalization.

As I was mulling over this state of affairs, I wanted to check myself against my own bias, because I really had to admit that, though I have the creeping intuition that children’s programming is conditioning children in subversive ways, I couldn’t really quantify the extent to which children are being targeted with, for example, sexual agitprop.

The challenge, as always, is how to validate one’s own intuition.

To solve this problem, I decided to create some software tools which can automate the characterization of the verbal content of children’s programs. The unstructured and varied expression of human dialog has historically constrained the ability to automate subject-matter characterization at fine grains. Beyond key-word searching, which is hardly comprehensive, ferreting out meaning from the subtleties of human expression has been exceedingly hard. This is especially true when fleeting comments are made that normalize some idea which may itself be entirely irrelevant to the plot itself. Word searching means you have to know in advance what it is you’re looking for. A more robust, fine-grained semantic characterization is needed. Like the momentary clip of the kissing women in Star Wars, minor comments in the dialog plant similar seeds of normalization, but they may fly mostly under the conscious radar of someone engaging with the plot.

So I’ve been wondering whether advances in artificial intelligence, and large language models in particular, might provide an opportunity to elevate the signal above the noise in this area. Can they do a more complete job of identifying these subtle normalizations that are embedded in these shows? If so, it might be possible to build tools that can do this kind of analysis at scale. This might change the balance of power, at least where arming parents with information is concerned.

There are two essential technical challenges in doing this. One of those challenges is in acquiring the TV show scripts in the first place. The other is in how the AI models are exploited by the tool(s). I should emphasize that the work I’ve done so far on this is confined to the dialog - I am doing no image analysis that might, say, quantify the prevalence of media like Star Wars that silently employs the use of fleeting images to implant subversive ideas.

I have built a set of tools to partially solve the problem of acquiring the dialog from children’s television shows. I am not going to go into the techniques I’m employing to do that. Suffice to say that for some number of the top ten children’s shows, I am able to automate the process of hoovering up the scripts for subsequent analysis. For purposes of development and testing, I have collected the scripts of around 6000 episodes of television programs, split almost evenly between two of the top-ten most watched children’s TV shows.

With that problem partially solved, I turned my attention to automating the analysis of the scripts using a number of different AI large language models. All of this is working and I have run the analysis on all the 6k episodes I have collected so far. I have also done some random auditing of the resulting characterizations. I would say the results are mostly accurate though not perfect.

As an aside, this kind of analysis is computationally expensive. It costs roughly $.10 per episode for the computational costs involved in doing this analysis. So all things being equal, I could have done all of the analysis for around $600. But, inevitably, you end up spending on invalid runs. I’ve done a couple of runs where I only realized after the fact that my prompt had an error. So I may have spent around $1000 to acquire this data. I mention the economic costs primarily just to give the reader a sense for some of the hurdles. Still, at $.10 per episode, it is far more economically feasible than trying to do this with rooms full of human readers. It’s also pretty amazing that it takes only about 3-4 seconds to semantically characterize the script of a one-hour TV show.

The real key to effectiveness may turn out to revolve around clever prompt engineering, and that is an area I believe I still have room to exploit in more creative ways. I have barely scratched the surface where this opportunity is concerned.

The early returns on the data, for the two series that I have been able to evaluate, indicate that there is definitely a noteworthy presence in both series of normalizing what most Christian families would consider to be disordered relationships. Specifically, that usually takes the form of a noticeable recurrence of passing references to family structures reflecting same-sex relationships (i.e. a family comprised of children with two mothers or two fathers.)

Once I have more thoroughly audited the results of the automated analysis, I will do a post with the detailed data on each TV show and name them. But for now, I can share that roughly 4% of the episodes of one of these series presents children with family structures containing same-sex parents. The other TV series normalizes the idea of same-sex attraction in just over 10% of the episodes. Sometimes the AI models have flagged episodes for very subtle interactions, like good-natured teasing of one of the male characters about his possibly marrying another one of the male characters some day.

As an aside, some readers of this post may have no objection to homosexuality as a general matter. But the question of whether homosexuality is moral is separate from whether any TV show producer should be conditioning children in a particular direction without explicitly declaring what they are up to. In many areas of life - the food we eat, for example - we recognize the importance of truth in labeling. Whatever one believes about homosexuality, TV shows that quietly condition children, via the subterfuge of soft normalization, are sinister forms of manipulation. My research into the content of these shows is leading me toward the opinion that producers of media directed at children should be required to openly label anything they produce in regard to what it normalizes. If there are already laws that require this, then we have either an enforcement problem, or insufficiently detailed disclosure requirements.

My interests are not confined only to how these shows are conditioning children in regard to human sexuality and gender. I hope also to do evaluations at scale regarding how these shows characterize environmental concerns, human preeminence in the world, human freedom, civil rights, and even economics. One could also easily imagine expanding the range of media sources, from musical lyrics to children’s textbooks.

It does appear, based on experiments with multiple generations of language models, that these models are just now getting to the point where they can be reliably used for this kind of analysis. Models older than nine months or so seem to yield far less reliable results than do the more recent models.

I plan on gathering scripts from as many of the top-ten children’s TV shows as I can. Some of those will take longer to gather and may entail actually streaming some of the episodes to capture the audio for automated transcription. Happily, automated transcription is another technology that is coming into its own in terms of accuracy.

At any rate, this is one of the things I have been working on around the edges of my schedule. I realize there is significant and even understandable angst among many Substackers about the implications of AI generally. I have my own reticence about man-made things that talk. But I am not in the camp that anthropomorphizes AI models, nor do I venerate them, or hold them in awe. My concerns about all of this are much more directed toward untrustworthy humans than toward inanimate models. But these models’ distillation of numeric weights, from massive bodies of human text, may make them suitable resources for unmasking what seems to be one source of a trickling, insidious conditioning of children that is adversarial to the wishes and worldviews of many parents.