Ricochet is the best place on the internet to discuss the issues of the day, either through commenting on posts or writing your own for our active and dynamic community in a fully moderated environment. In addition, the Ricochet Audio Network offers over 40 original podcasts with new episodes released every day.
The D&D Theory of Automation
When exactly is AI going to take your job? One year? Two years? Twenty? Can you last until retirement? As Yogi Berra said, “Predictions are hard, especially about the future.” It depends on who you are, what your job is, and how good you are at it. I’m describing here a conceptual model I have for these questions to understand how easy it is to build a machine to do a man’s job. It starts with Dungeons and Dragons.
Pen-and-paper role-playing games are unique in the level of freedom allowed to players. The little car in Monopoly can only explore the streets laid out on the Monopoly board. In an RPG the player can always ask the game master, “Can I do this instead?” This leaves the game master with a problem to solve: “Well, can he?”
Take an example from an epic of the past. Beowulf is wrestling with Grendel. Is Beowulf strong enough to tear off Grendel’s arm? In the saga the skald can declare yes, he was. In a game it’s more interesting if there’s a chance to fail. These games answer that question by providing the player with certain attributes, and checking whether they’re strong enough with dice. The details of how this works are not at all important to this question, but the attributes are. With your permission, I’ll run down the six numbers that define every D&D character. This will be old hat for some; feel free to skip ahead. For the rest, this won’t require you knowing any more about nerd stuff than you need to.
Strength: How big, beefy, and brawny you are. A high strength lets you bench press a few reps of a really heavy weight. In the climactic confrontation in the Odyssey, when Penelope declares that she’ll wed the suitor who can string her lost husband’s bow, only Odysseus is able to string the bow because only Odysseus made his strength check.
Dexterity: How nimble or agile you are. Dexterity governs your ability with lockpicks. When Robin Hood splits his opponent’s arrow, it’s because he’s got a really high dexterity score.
Constitution: Your toughness or endurance. Your ability to bench press a lower amount of weight many times. The messenger who delivered the news of the Battle of Marathon made his constitution check to run that distance, but (according to legend) failed the check to survive doing so.
Intelligence: How smart, how clever you are. If you needed to remember an arcane fact or break a code, this would be the attribute you roll against. Bilbo under the Misty Mountains used his intelligence to guess Gollum’s riddles.
Wisdom: How understanding you are. Your judgment. Wisdom is perhaps the least well-defined attribute. Charon, the centaur skilled in the arts of healing, had a high wisdom score.
Charisma: How likable, friendly, or inspiring you are. You’d use charisma to enthrall your audience with a story or con a mark. Orpheus, whose music was so beautiful that even wild animals would stop to listen, had charisma coming out of his ears.
Here’s the key insight: imagine that reality is filled with these same attribute checks. Every bit of automation exists to replace your roll. I might not be able to solve a given math problem, but Wolfram Alpha can. Instead of rolling against my intelligence, I can automate the problem. With that in mind let’s talk about automation generally.
Way back in the Paleolithic we started to build tools. Caveman want fire. Caveman can tear down tree with bare hands. Caveman tear down tree easier with fist axe. Fist axe provides a bonus to his strength check.
Beasts of burden, the wheel; these sorts of things exist to obviate the need for strength and constitution checks. I could carry a sack of grain five miles (in theory.) Or I could set it on a wagon and pull the wagon. Or I could get an ox to pull the wagon for me. If I were running a role-playing game and a player were coming to me with these various solutions, I’d make his roll easier or harder depending on which he chose.
Until the Industrial Revolution technology mostly removed strength and constitution checks. Then our machines got better and better at replacing dexterity. Ludd wasn’t destroying weaving machines because they were too strong for him. (This is not to discount the strength-providing aspects of the steam engine; John Henry pitted strength and endurance against the machine.) You can see where this is headed, but half the attributes are still untouched.
Is it possible to automate charisma? Sure, depending on what you’re talking about. Thomas Edison’s phonograph suddenly made a lot of charisma checks obsolete. Tavern songs used to be a thing. I’ve never been in a bar where people are singing, but I’ve been in plenty with blaring jukeboxes.
The computer, in all its various forms from ENIAC to modern AI, exists to replace intelligence checks. Need to solve a math problem? Use a calculator. Accounting, research, chess; all of these things have gotten far easier because we’ve devised machines to do intelligence checks for you.
Today’s worries about AI come from another advance in machine automation. If you want to write a blog post, you can either roll against your intelligence or use ChatGPT. Paint a picture? Roll charisma or ask OpenAI. Suddenly whole classes of jobs that used to be the sole provenance of humans are threatened by cheap computer substitutes. If you can automate a solution to my strength checks, and my constitution checks, and you’ve got the dexterity thing sorted out, and now you can replace my intelligence and charism, what do I have left? What room does that leave for us poor, squishy humans? Perhaps you’ve noticed that I left one attribute out.
Wisdom. I can use a wagon to move grain to market, but the wagon doesn’t tell me which market. I can use a sophisticated algorithm to make stock trades, but it doesn’t tell me if I’ll be happier with more money or not. I can use a weed whacker to trim my grass, but that gives me no information at all as to how good my garden will look afterward.
You can use a machine to do a great many things, more now than ever before. You can’t use a machine to replace the wisdom that tells you whether or not you should do those things. Oh, you can build a decision algorithm to choose markets or plot a garden. You could also flip a coin. That doesn’t mean your machine is making the right choice. Wisdom can’t be automated currently, and I believe it never will be.
If you’ve ever seen an AI-generated piece of art and shuddered at those dead eyes, that’s why. If you’ve ever read an AI-generated story and called it good enough but not great, that’s why. If you’ve ever worried about losing your job to a computer, this is your opportunity. And if you’ve ever worried about wasting your life on games, it’s possible to learn something from them, too.
If you haven’t been min-maxxing your Wisdom score, I suggest you start now.
Published in Technology
Well done, Hank.
There are people with undeniable innate intelligence but not enough, or false knowledge to work with, the all too common phenomenon of the clever airhead. My own humble analogy is an 18-wheeler; it’s not just enough to have a powerful tractor cab up front. You actually have to back it up with a full cargo of actual true thoughts and ideas.
The next stage up is the driver, representing wisdom, the irreplaceable boss of the powerful combination of raw brainpower and hard-acquired knowledge. But there’s a case where my analogy is being overtaken by AI, at least on the surface, because now AI can drive the truck.
I believe I can redeem the analogy by comparing it to the OP text; technology in the driver’s seat of a truck doesn’t mean that AI now has the wisdom of a driver, it means that wisdom is no longer needed to perform a driving job.
Wisdom seems to me how well an individual copes with dramatic (might be modest but rise to dangerous, cataclysmic, live threatening, etc) surprises. Brainpower + knowledge + personal security and awareness + intent + will + objective measures like strength and constitution. Some are genetic it seems, but many also seem susceptible to environment and timing. I believe we are given the genetics, but intended to work to max the non-genetics within our power to more wisdom. Expect a surprise.
A lot of wisdom comes from experience and computers are getting better at collecting shared wisdom from the internet and summarizing it. For a lot of situations, it is better than no wisdom. To know differently would require at least a 16 for your Wisdom.
You hinted at it in your second paragraph. The killer stat isn’t “Wisdom”, it’s “Creativity”.
I’ve never played D&D, and I have only trace amounts of second- and third-hand knowledge of the game and its community. I’m pretty lost in the details here.
However, I find any discussion about how “AI” is being used to de-humanize humans to be worth reading and thinking about.
So, this post is worth promoting to the Main Feed, and it’s likely that several unique and separate conversation threads will develop as a result.
I think wisdom can be reasonably defined as common sense, perspective, prudence, a practical understanding of actions and consequences: essentially, life’s Best Practices. And I think AI will distill that soon enough, if it hasn’t already, from the wealth of reading available to it.
The first machine revolution, the replacement of human strength and dexterity with automation, brought increases in efficiency that made it possible for more people to live much better while working less hard.
I think the second machine revolution is underway, and I wonder if it will bring sufficient increases in productivity to offset the social cost of the displacement, primarily of marginally productive workers, that I believe is coming.
Or at least his lyre did. I don’t recall if he had a great voice to go with it but we also don’t know what he looked like, leading me to think that he was the ur-example of “chicks dig rock stars”.
Have you not heard of the ancient art of karaoke?
Most of us don’t have to wrestle Grendel or produce art or generate original thought. We get paid to execute processes. Will a hundred million people console themselves with the thought that the AI’s that displaced them cannot really be creative, original or happy?
Lawyers are convinced that AIs can’t do law but AIs will soon be able to do 90% of what lawyers are asked to do. In the very near future it will be considered malpractice for a physician to disagree with the diagnostic and treatment management AI without a really, really good reason.
If we do manage to persuade Skynet, HAL or the cylons that human creativity is a non-replicable value, they might reasonably conclude that they should permit the most gifted 5% of us to survive.
I foresee a period where your continued employment will depend on your ability to teach the AI something it doesn’t already know. Of course, each time you do you will bring yourself closer to unemployment.
Insofar as it thinks that it needs whatever that creativity can produce, yes.
I am very impressed with some of the results I see from LLMs, especially Grok and Perplexity. BUT, every now and then, something shows up that is a reminder: they look like they understand more than they actually do,
For example, I asked Perplexity to consider the works of the 19th century actress and noted diarist Fanny Kemble and imagine that she was somehow transported to present-day America and wrote her impressions in her diary. Here’s what it came up with.
Pretty impressive, I thought…until I asked it to write about Fanny’s realization that she is going to need to pay her hotel bill and she has no modern American money. Then it starts talking about how she could use credit cards, venmo, etc, without seeming to grasp that these things need to be funded somehow.
The USAF, responding to its interpretation of Trump’s executive orders, apparently created a script to go through all of the photographs it has on its website(s) and remove the ones that contain certain keywords, such as ‘gay’.
So it deleted the picture of the bomber Enola Gay.
This example is just silly and (hopefully) embarrassing to its perpetrators, In other cases, the damage could be a lot more serious.
So, here’s the fundamental problem in a nutshell…”they” don’t “understand” at all.
A digital information system is a machine, an artifact of techne that is a man-made object. It is not (nor ever can be) a person. Using a plural third-person pronoun obscures this, but it is a fact nonetheless.
“Understanding” happens when the user (in this case, @davidfoster reading the results from the Perplexity app relating to Fanny Kemble) assigns meaning to the outputs produced by the machine. The machine is programmed to recognize and replicate highly complex sequences of “1” and “0” and then rearrange them into new sequences according to patterns of higher or lesser rate of occurrence. Interpreting the aesthetic merit of those outputs (“good” or “bad,” “pleasant” or “unpleasant”) is wholly a function of the user, not the machine.
Of course, now that the user (David Foster) has posted digital breadcrumbs of his (human) assessment of Perplexity’s content on Ricochet, re-running the machine may result in a new product with a higher likelihood of producing a product that elicits a similar assessment by some subsequent user seeking a simulacrum of Fanny Kimble’s thoughts. The machine neither knows nor cares what is “good” or “bad”; the machine can only be assessed as doing a “good” job or a “bad” job by an outside, human observer.
I’m not saying y’all are wrong. I will say that I was right that it must be the least well-defined attribute otherwise I wouldn’t see this many competing definitions. Either I made my wisdom check to not argue the definitions, or failed the intelligence roll to be able to.
I’ve always found “wheedle GM” to be the most powerful skill. If you need bonuses to it you can always order pizza.
The thing about that is that I see a lot of automation re-humanizing people. Or rather, their jobs. Let me give you an example. My cousin makes his own skis, medieval style. Suppose that was his job, as it might have been back in the day. After a while any job is a job, but at least it’s interesting work. (I recommend the link; it’ll be more interesting than anything I write.)
Fast forward to Metropolis, Charlie Chaplin, in the 1920’s. His factory job is to tighten two nuts on a plate as they go by. Over and over and over again. The best you can say is that the constant motion makes his shift pass quickly. You might get more production out of the assembly lines of early mass production but the job won’t be as interesting as carving skis simply because the job requires less judgement and contains more repetition than hand crafting.
Move on to the factories of America today. You’ll never find anyone with a Chaplin-esque job because mechanical motions that simple are easy to automate. You build a machine for that, and use your workers to perform tasks that require judgement, and the flexibility to switch between tasks as needed. Perhaps you no longer need a thousand workers to perform the same task; perhaps you only need two hundred. Those two hundred jobs will be more interesting to the folks working them, and hence more human.
There are plenty of ways to use AI in dehumanizing manners. But realistically that’s been happening for a while, and without the assistance of large language models, as anyone who’s currently being told “para Espanol marque dos” can tell you. Most “good enough for the likes of you” slop that I’m fed isn’t the fault of LLMs. At least not yet.
When’s the last time you’ve seen any intersection between ‘karaoke’ and ‘charisma’?
I think that there is another attribute, and it’s another that the AI will have trouble with. Instinct. Subconscious evaluation of imperfect data, to produce a ‘hunch’. It can give you percentage likelihoods of all sorts of things, assuming there is some data by which to make such an evaluation, but it will never guess when the data isn’t there.
You may not have to wrestle Grendel, but consider taking it up in your leisure time.
Producing art and original thought, the bar for those are still incredibly low. Think of the chef in the back of a restaurant plating a meal. That’s art, at a very basic level. If you automate it away you get McDonalds. Nothing against the golden arches or those who eat there, but there’s a difference. Perhaps you could train a robot to do that job but I suspect it will be quite a while before the AI frontier crosses that boundary.
As far as original thought, try weeding a garden. My mind shudders at the task of building a robot to do that; even if it has a 98% success rate determining which are flowers and which are weeds as you roll that 2% chance each day you’ll lose more and more blooms.
Even supposing that AI means these things can be automated now the question becomes ‘how soon’, and the answer to that involves economics. That time gives displaced workers time to find something else.
Will the doctors have to worry about malpractice so much when they can spin up robot lawyers to defend themselves? Either way, that isn’t an AI problem, that’s a legalism problem.
Suppose that you’re right about that 90% figure. Nine in ten lawyers lose their jobs. The ten percent that are left will still be able to fill their quota of inflicting pain on doctors, and they’ll have more interesting jobs because of it. If a lot of make-work lawyers lose their job I’m sure it will break my heart.
Not too long ago @Misithiocracy mentioned a discovery he had made. He was using a large language model to write programs. He figured out that if he started from scratch and asked the machine to keep making changes the results would get worse and worse over time. But if you generate the code, copy it to a new prompt, and ask that version of the AI to edit your code to make the slight change that you want it’s much more faithful at making the change you actually want.
Without going in to why I think that happens, let me just say that dog trainers don’t go out of business because they teach the dog all their tricks in the first hour.
Kind of hard to put on a resume though.
Maybe so, but only if I’m also interested in the history of ski making instead of actually doing it.
This is true. I’ve eaten your cooking.
I’ve been at the karaoke bar with my wife. You’re free to tell her that, I’m not going to.
I’ll grant these guys are a cut above your usual live music but also:
Does it work?
Has it been tested?
It’s loaded on the airplane. Wanna take a spin?
I don’t know about this. The AI is only going to be able to work with the information you give it. It might not be able to see that the patient is a meth head but it will be pretty obvious to a human doctor.
Peter Drucker said that the assembly line doesn’t represent a perfected form of human work, it represents an imperfect form of machine work.
Yes, these are fair observations. And honesty requires we acknowledge that humans have been de-humanizing each other since the dawn of time. Slavery is the practice of reducing people into an economic commodity, i.e. a unit of labor that can be bought, sold. So too, the practice of renting such labor in the form of “wage slaves” who provide low-skilled, easily replaceable assembly line workers.
I don’t think we know how human cognition works, so I’m not confident that we can rule out the possibility of machines eventually thinking, and understanding, with comparable ability.
That’s always the hope, right? That the intelligence revolution will echo the mechanical revolution, and displaced people will move on to something better and more rewarding.
Once upon a time, almost everyone did something that involved the direct physical manipulation of matter. Then machines began reducing the need for such people, while simultaneously vastly increasing wealth and prosperity. People displaced from physical labor — or their children — found gainful employment manipulating information and providing knowledge-based services the value of which derived from their specialized expertise: working at the motor vehicle department, selling insurance and pharmaceuticals, teaching school, becoming Instagram influencers.
When, and if, the number of people who can be employed providing knowledge-based services experiences a decline similar to that of the decline in matter-manipulators brought about by two centuries of mechanical revolution, to what category of endeavor will they migrate? More people are capable of putting square pegs in square holes than are capable of running AutoCAD. Are more people capable of running AutoCAD than are capable of [name your post-knowledge worker category of endeavor]?
Exploration and discovery, perhaps. Surely that will always be a field uniquely suited to humans. Right?
“It takes four hundred thirty people to man a starship. With this, you don’t need anyone. One machine can do all those things they send men out to do now. Men no longer need die in space, or on some alien world. Men can live, and go on to achieve greater things than fact-finding and dying for galactic space, which is neither ours to give or to take.” — Richard Daystrom
Some history about automation…and automatic panics: Attack of the Job-Killing Robots