ai – The Ruminathans

How I use large language models in my writing process

Ruminathans are pretty rare. I publish one about every two months on average, with high variance. I could probably increase my output substantially if I used generative AI (gAI) to the fullest. The reasons I don’t are:

The output of gAI is persistently not up to my quality standards.
I want to exercise my writing and thinking ability, so outsourcing that work to gAI would be like asking a robot to go jogging on my behalf.
I genuinely enjoy writing and thinking, so outsourcing that work to gAI would be like asking a robot to make love on my behalf.

That said, here are some ways I use gAI in the writing process: What it does and doesn’t do for me, and what I think that reveals about the future of essay-writing as a human endeavor.

Research

As the quality of traditional search declines in the face of the SEO arms race and the explosion of (gAI-produced) slop, it’s easier to find empirical data by posing research questions in ChatGPT (I’m using a paid version). In my recent posts about macroeconomics it helped me find the right databases. You have to verify that the links exist, that they are reputable, and that they say what ChatGPT says they say.

It’s far too easy, though, to use this “research” to cherry-pick data. You can ask ChatGPT to “Find data that supports [INSERT PET THESIS HERE]!” and it will oblige. It won’t push back and warn you that the balance of the evidence undermines your pet thesis.

You can, of course, prompt it to evaluate the available data. But that reveals the limitation: You can prompt it to produce anything you want. Asking it to evaluate the available data implies that you are looking for a balanced assessment. It will oblige, leaving you wondering if it just mirrors the biases embedded in your own word choice.

An easily confused sparring partner

Exploring a single topic in a longer conversation with ChatGPT requires a lot of care. The fact that it does not get meaning becomes apparent when you whiplash it around. Ask it to defend a thesis, then defend the antithesis, and then ask it to revert to arguing the original thesis: After that back and forth, it tends to get very muddled about which argument supports which position. It neither notices inconsistency nor is it embarrassed by it.

Say what you will about humans and their ornery unwillingness to update their priors: You can count on them to stick to their guns, and that’s what makes a good sparring partner.

ChatGPT is not a real sparring partner: Dance around it for a minute or two, and it will start punching itself in its tattooed-grin face.

Flattery will get you everywhere

Speaking of its ever-chipper attitude: Once I write a finished draft of an essay, I ask ChatGPT to evaluate it for clarity, elegance, and originality. Even 5.0, which is supposed to have overcome the obsequiousness of prior versions, is still an inveterate flatterer. And hey, it’s nice to hear some positive feedback.

To its credit, if I push a half-written fragment through, it identifies that it’s flawed and provides some feedback about how to improve it. But its concrete suggestions are largely unusable: It will never sit you down and say: “I don’t understand your point” or “This is too clunky to be salvageable” or “Stop beating this dead horse.” And that’s feedback that you need to hear sometimes.

Bottom-line: This step adds very little value to the writing process. I still do it because, yes, the charitable feedback provides a little pick-me-up after I’ve reached a draft. And I’m still hoping that someday it sounds a big alarm… but I’m not holding my breath.

My sworn enemy

You can, of course, prompt it to take a critical view. My preferred approach is to ask it to write a 300-word review of the article from the perspective of the author’s sworn foe.

The result is a cold slap in the face. There’s little of substance in the critique: usually it does not identify anything – neither stylistic nor content-related – that I did not anticipate and consciously choose. I find this step helpful, though, insofar as it shows me how someone might react emotionally.

To some extent, I can anticipate the perspective. There are people who are, shall we say, uncharitably disposed towards me. I can – equally uncharitably – simulate their voices in my head without ChatGPT’s help. But after the initial flattery, my bot-enemy’s perspective is a salutary palette-cleanser.

After my foe has weighed in, I usually ask ChatGPT which of his critiques are fair. If I agree and can make some changes without too much effort, I do so, but usually not the changes ChatGPT suggests.

The “best available editor”

Ethan Mollick, one of gAI’s biggest boosters, introduced the concept of the “best available human” to argue in favor of widespread use of gAI. “Sure,” says Mollick, “the most skilled humans can still do many tasks better than AI. But the issue is not what the most skilled humans can do but what the best available human can do. And much of the time, the skilled humans aren’t available.”

Frankly, I’ve been underwhelmed by Mollick’s writing. In this case it seems like he’s unfamiliar with the concept of negative numbers and the value “zero” as a valid solution to many problems. If the best available human would do a billion dollars of damage, and an AI agent only does a million, the correct choice is to abandon whatever the f#*k you’re doing, not outsource the problem to AI.

Still, in Mollick’s spirit: I’ve tried to prompt ChatGPT to be an editor at a national magazine who is willing to read my article and ask guiding questions and then give me a thumbs up or down for publishing it. An actual national-league editor is not available, but her simulacrum is.

The editor succeeds at asking a set of interesting questions, and it can be prompted to stop asking and reach a verdict. As interesting as some of the questions are, the problem remains: It will not shoot me down. And sometimes, an idea has to be shot down.

Helpful, question-led suggestions – while perhaps leading to incremental improvements – serve only to lead me further down a rabbit-hole, when what I really need to do is cut my losses and move on to the next idea.

The value of blind spots

We all have our blind spots. They come naturally from being an embodied soul with a particular life history. What we think and what we write is just as much a product of what we don’t see as what we do see.

That’s what makes conversation so compelling – and reading and writing is a form of conversation. Through others’ eyes we can peer into our own blind spots. But it’s necessarily their own constrained perspectives that give them that vantage point. An outsider’s perspective is not valuable only for the unfamiliar ideas she brings; it’s equally valuable – sometimes more valuable – for being unburdened by the ideas the insiders take for granted.

In a good conversation – including one with an editor – your partner will point out your blind spots unprompted. I’ve not seen ChatGPT do this. I’ve tried. This essay went through the process I described above, intentionally without this section on blind spots, which I had mentally sketched out but not committed to pixels. The idea of “blind spots” lay in the incomplete essay’s blind spot. ChatGPT did not point it out (nor did it point out any other noteworthy blind spots).

You can, of course, prompt the bot to look for blind spots. An explicit prompt (“Look for blind spots”) does not work: You get no fresh insights. You can ask it to write a comment from a particular writer’s perspective (“Imagine you’re George Orwell”) and it does better, but mostly it delivers a superficial emulation of the style, not a fresh perspective. It’s nowhere near the experience of being confronted with someone who makes completely different assumptions from your own.

Large language models seem to encompass every perspective, and the perspective from everywhere is a perspective from nowhere.

I could easily use gAI to create more output, more quickly. But I struggle to see what service I’d be doing others or myself.

Making machines smarter or ourselves dumber? Why I’m betting on model collapse.

If the emergence of artificial intelligence is as profound a development as both its boosters and its doomsayers predict, then we all need to make some pretty important decisions: bets about where we invest our savings, our time, and the time of our children.

In decision theory, you choose a course of action from several options based on the expected value: You look at representative scenarios, value them in whatever terms you want (monetarily or in terms of your personal utility), and weight the scenarios by their respective probabilities of occurrence. Then you choose the option with the highest expected value.

I’m acutely aware of the limitations of this approach to making decisions. But let’s take it as a given for the moment: What are the representative scenarios for how the world evolves with AI? I’ve been thinking about three:

AI becomes smarter than us: We train models that outperform us humans at every conceivable task. The emperor really has clothes.

We become dumber than AI: We train ourselves to no longer appreciate the difference between “intelligence” and its parody. The emperor has no clothes, but nobody is capable of calling him out.

AI fizzles: Whatever mechanisms have made AI appear smarter from iteration to iteration cease to deliver even marginal improvements and even lead to deteriorating performance (aka “model collapse“). The emperor has no clothes, and we’re willing and able to call him naked.

Our next step would normally be to come up with probabilities of each of these scenarios occurring and then devise some strategies with respective payoffs for the three scenarios.

But this approach makes sense when there are positive outcomes. It doesn’t matter what the probabilities are, if all but one of the payouts are zero (or less) under any conceivable strategy, then those scenarios aren’t even worth planning for.

I’m struggling with imagining any positive payoffs in the first two scenarios.

Society works because we have a stake in each other’s existence. If we become unnecessary to each other – if we no longer need to tie ourselves into a web of reciprocity – then we will become charity cases either for whoever controls the “Artificial General Intelligence” or for the AGI itself. Being interdependent, as we are now, is risky; being just plain dependent is a blind alley. I cannot see this as a positive outcome.

I don’t believe this scenario is very probable. But it doesn’t matter. No matter what its probability, it has 0 (or even negative) outcome value. Unless all scenarios have non-positive outcomes, there’s no sense in planning for it.

If you put a gun to my head and asked which of the three scenarios was the most probable, I’d have to go for scenario two. If the overwhelming majority of people do not see a difference between real intelligence and its parody, they cannot value that difference. And so in that scenario, I will also become a charity case because I cannot sell anything anyone wants to buy. Unfortunately, that feels like the direction we’re heading in. Nor am I alone in thinking that.

That is why I don’t think the elaborate game of decision theory makes sense here. There is only one scenario that has positive outcomes. I’m placing my bets on AI fizzling and the strategies that optimize for that scenario: Maintaining my skill set without AI assistance. Investing in a broad range of asset classes. And helping my children learn critical thinking and, hopefully, the courage to call emperors naked. And maybe if we all take that approach, we’ll make Scenario 2 less likely.

We cannot afford what we do not produce: social security, demographics, and the hidden cost of innovation

Across the developed world, hands are wringing vigorously about the sustainability of public pension schemes. Due to changing demographics, fewer and fewer workers will have to support more and more retirees.

What gets lost in the debates about the solvency of Social Security or similar programs is that the problem is not caused by the public nature of the scheme. Any conceivable private alternative would face the same underlying challenges. And those underlying challenges lie at a deeper level than demographic change.

Misfortune, senescence, and our long childhood

More than any other creature, we humans face a unique set of survival challenges:

Misfortune: Our own productivity is subject to variation for reasons beyond our control, whether from illness, climate variation, or monetary policy
Senescence: We are likely to live beyond our ability to provide for ourselves
Long childhood: We arrive in this world unable to provide for ourselves, and we have to learn survival skills through a long phase of cultural transmission.

To address these challenges, working adults have to produce surpluses. We have to produce more goods and services than we can consume, and we have to find ways to store the surplus against both rainy days and old age. But no matter how productive we are, and no matter how ascetically we live, we cannot individually store enough food, water, shelter, medical supplies, etc. to get by for more than a few months, maybe a year at best.

At its simplest, we solve the problem with the intergenerational compact in which working adults provision their elderly parents while raising their own children. But we’ve always mutualized that compact beyond our immediate bloodline. Whether as hunter-gatherers, pastoralists, agriculturalists, industrial workers, or knowledge workers: We have always had to throw ourselves at each other’s mercy, paying forward some of what we don’t consume today. In turn, we look to others to provision us in our times of need. “What goes around comes around” is our great hope.

Project Civilization has been about creating institutions to buttress that hope. Central distribution hubs to store food (aka cities). Credit contracts and written records thereof. Private property. Insurance. The thing we call “money.” All those institutions are grounded in our collective will to allow a state to enforce these arrangements. With rule-bound violence if necessary.

Because it’s ultimately the state’s enforcement capacity that keeps our hopes credible, we’ve even short-circuited these institutions and have asked the state itself to collect and redistribute surpluses. That’s what public pensions like Social Security do. And that’s equally the premise of publicly funded education for our children and income security programs such as unemployment insurance to handle misfortune.

It doesn’t matter which mechanism we use: Whether collected by the state, saved in money, or invested in private enterprise, without surpluses generated by others there is nothing to store and nothing to distribute.

Demographic change: cause or effect?

The problem with our pension systems is not in how they are financed. The problem is that the working age population may not produce enough surpluses to simultaneously provision the retired and provision and invest in the next generation. It’s not about surpluses of money. It’s about surpluses of goods and services.

Economists speak of the age dependency ratio. There are different formulations thereof, but in its simplest form, it’s the ratio of the population under 15 or over 65 to the working age population between 15 and 65.

The naïve narrative about the looming crises in public social security systems is about demographic change. On the one hand, extended longevity has meant that people spend longer in the “retirement” phase where they cease to be net surplus generators. On the other hand, we are having fewer children, which means that there are fewer surplus-generating adults.

In the naïve telling, when we set up our public pension systems, we assumed our population would stay pyramid-shaped forever, with a few surviving retirees supported by an ever-widening base of new workers. In the US, the ratio of worker to beneficiary went from 3.3:1 in 1985 to 2.8:1 in 2021. I’m using 1985 as the baseline because that is 45 years after the first Social Security payment, so roughly when the system should have hit a steady state.

According to the mid-point projections, the US is headed towards a ratio of 2.1:1 in the next decades. In countries like Germany, with lower birth rates and lower immigration, the ratio is already 2:1, with projections taking the number of workers per beneficiary even lower.

The naïve story comes with naïve policy recommendations: People should work longer (US, Germany). We need “pro-natalist” policies incentivizing women to bear more children (US, UK). We should privatize social security (US, Germany).

And the ultimate policy recommendation is always: We have to increase productivity with our tried-and-true approach of greater labor specialization enabled by more technology.

The fact that the explanation of the problem and the recommended solutions are naïve (at best) or proposed in bad faith (at worst) is illustrated with two simple observations:

The problem exists whether the mutualized surplus-storage solution is organized privately or publicly.
Any system – public or private – built on indefinite exponential growth of the number of human beings on the planet is absurd from the get-go.

I would like to take the analysis of the underlying problem a step deeper than demographic change and the dependency ratio. I will show that causes and effects are intertwined, the policy measures miss the point, and “more technology” cannot always be the answer because it’s often the problem.

The intensification of child-rearing

Let’s start with another naïve observation. Fewer children should also mean fewer “unproductive” mouths to feed and more time available to generate surpluses. So on the face of it, having fewer children could also alleviate the problems facing pension schemes. In fact, economists speak of a demographic dividend that accrues when societies reduce their birth rates. This is a point we’ll return to a little later.

But having fewer children doesn’t necessarily mean we need fewer surpluses if the amount of resources dedicated to child-rearing intensifies. It’s one thing to have seven children when you live on a farm in a low-density settlement and can count on them to start contributing to farm production from a very young age. Or to have five children in a cramped one-room apartment in a high-density city if you can send them to shovel coal into a factory’s furnace, and that’s all the job that will ever be available to them.

If, however, becoming productive in a hyper-specialized, knowledge-based economy entails 16-30 years of dependency, 10-24 years of which are devoted to resource-intensive education, then you may not have the resources to afford two children, let alone five or seven. The US Department of Agriculture estimates the cost of raising a child born in 2015 at $233,610. Crucially, that is the cost for raising children to age 17 and does not include college education and job training, nor the provisioning of young adults until they become net surplus generators. Nor does it include the opportunity cost – borne mostly by women – of parents who take off time during pregnancy and earliest infancy.

The high cost per child in places like the US might partly reflect overly generous consumption. There’s some truth to that. But the fact remains that when labor becomes hyper-specialized and knowledge-based it requires more resources to get to the point at which you become productive. It’s not just the basic needs and education. The more specialized and the more knowledge-based the economy, the harder it becomes to objectively measure whether someone has acquired knowledge and skills. Effort and resources go into signaling: prestigious degrees, resumé-padding activities, exclusive networks, fine clothes, all the things that might be subsumed under that horrid but real concept of a “personal brand.”

Raising the next farm hand or coal miner simply did not take the same amount of resources as raising the next oncologist, robot-plant operator, or social media influencer.

Pro-natalist economic policies like tax breaks and subsidies have been tried in many countries. They rarely and barely move the needle. And it’s not hard to see why. Having one to two kids on average is as much as we can afford on average.

The root cause is ultimately not that people don’t want to have children. The root cause is the amount of investment it takes to get children to be productive in the high-technology, hyper-specialized economy.

Insofar as “more technology” increases specialization and increases the need for learning, it may not contribute to the solution of inadequate surpluses. It cannot be a default answer to the problem.

Age or skill obsolescence?

The adult worker gets squeezed at both ends. With a bit of poetic license: She has to work and scrimp for 50 years to raise her child through 25 years of education to become the geriatrician she then has to pay to keep her parents alive during 25 years of retirement.

When our pension schemes were devised, they assumed people would be net surplus providers until around 65, then live off the next generations’ surpluses before shuffling off their mortal coils at around the biblical three-score and ten years.

Yes, people now live longer. And not just that: They haven’t magically started to live longer. They live longer because we’ve unlocked ways to keep people alive longer. Those medical technologies are miraculous. But they also consume massive amounts of resources and labor. My father had a medical procedure that could have extended his lifespan by a decade or two but that cost somewhere close to the median after-tax annual salary. In his case, it failed, sadly.

Rather than retirement being a phase of reduced consumption, it often involves an intensification of consumption. Peering beyond the veil of money to the underlying real economy of goods services: We train and provision geriatricians and oncologists who might otherwise have become schoolteachers and carpenters.

Historically, however, old age did not mean you stopped contributing. You shifted your contributions from, say, hunting and gathering, to childcare and cooking. That’s what freed others to do the surplus-generating hunting and gathering. And childcare meant imparting the skills and knowledge required for children to become net surplus producers. Or knowledge about how to survive once-in-a-lifetime ecological crises.

Age is not the problem. The problem is living beyond the obsolescence of your skills. As technology changes, the skills and knowledge you acquired when you were young – the ones that enabled you to generate net surpluses – are now likely to become obsolete in the span of your lifetime. Sometimes more than once.

In some professions, it is possible to retain a high enough level of productivity to be a net surplus generator past age sixty-something. But probably not in the majority of professions. Whether you “retire-in-place” and still pull a salary, or spend time retraining to become productive in another domain, you are de facto benefiting from others’ surpluses in that moment. Through no fault of your own.

In a world in which technological obsolescence and global labor specialization can destroy your community’s economic foundation overnight, working adults have to be mobile. So the option of shifting from direct surplus generation to indirect support by providing childcare is not open to every retiree. And even if it is, many of the life skills he has to impart might as well be from the Stone Age.

Demanding that people work past 67 is likely to be as ineffective as tax breaks for having more children. As for lower fertility rates, the real root cause is the rapid pace of technology-enabled labor specialization.

Misled by nostalgia

We face interlocking constraints that are both causes and effects.

The resources and time required to become and stay productive are increasing.
People live longer past the obsolescence of their skills.
The horizon over which our skills are valuable is decreasing.

The dependency ratio – even in its elaborations – does not capture the reality of our predicament if it is not weighted by the intensification of child-rearing and eldercare, and the risk of obsolescence.

It’s a largely unchallenged dogma that technological and organizational innovation solve more problems than they create. However, we may have passed a point where that’s true.

Much of what we believe is rooted in nostalgia about the period known as the post-war boom known in France as the Trentes Glorieuses (“The Thirty Glorious [Years]”) and the Wirtschaftswunder (“Economic Miracle”) in Germany. Whether we experienced it directly or not, that period has shaped our collective consciousness. It birthed youth culture and rock’n’roll, and all the values, beliefs, and expectations we hold onto, no matter where we fall on the political spectra.

But the Trentes Glorieuses were an anomalous happy optimum where:

The investment in skills required to become productive was not too onerous, meaning children became net surplus generators more quickly.
Because children became net surplus generators more quickly, people could also afford to have more of them.
You could count on your stock of skills to see you through 40-50 good years of surplus generation without becoming obsolete.
People did not live for decades beyond the obsolescence or deterioration of their skills.

In contrast, now we face a world in which technological innovation extends our lives at great material cost, while rapidly rendering obsolete the skills we so heavily invested in. Both factors impose a set of constraints such that we can invest only in one or two children, if any at all.

We face this predicament globally, though its urgency will hit different countries at different speeds for historically contingent reasons.

“More technology!” is the answer for many: Make the fewer and fewer people more and more productive thanks to machines and AI. But if AI actually undermines skill development faster or more deeply than it augments human ability, the calculation will not work out.

But that is another essay and will be expounded another time.