What coal and Jevons’ paradox tell us about AI and data
hex.tech
hex.tech
Here's a specific example: ChatGPT has already started taking away the work of copywriters - there have been a bunch of mainstream media articles about this. I think it's fine to argue "That type of copywriting was low-value work anyway. Now this means that sales and marketing folks will be able to operate at a higher strategic level since they no longer have to deal as much with 'how should this specific blurb on the website be phrased'." The problem with this is 2 fold:
1. A lot of that "low value work" is how new folks learn the tools of the trade to begin with. Over the past 30 or more years technology has done a very good job at "removing" the lower rungs of the career ladder. Remember those stories we used to hear about how folks "worked their way up from the mailroom, or from being a secretary"? The mailroom and secretaries largely don't exist anymore.
2. In my experience and belief, there are a large swath of people that just want to be given instructions and then execute them well and with pride. I think it bodes very, very, very poorly for society if all of those types of jobs go away and the response to those people is basically "fine, you can exist at a level just above starving". We had a booming post WWII middle class because there were tons of these types of good paying factory (and office) jobs that allowed people to have dignity in their work. The destruction of those types of jobs is a huge part of our current social strife and "deaths of dispair" in my opinion.
As tech takes away the low skill jobs, the skill level for "entry level" becomes higher and there's no way to learn on the job, you need to pay for education to level up enough to get started.
LLMs are just the current example of this, but it's been happening for years.
Not saying it's good or bad, but it is interesting that "working your way up" gets harder over time because it's harder to get that entry level low skill work.
On the other hand, higher stakes than most jobs, and much more rigorous regulations and feedback loops might mean it’s not feasible in other industries.
If free energy is ever discovered (extremely doubtful), we will not help but boil ourselves from the excess heat it generates.
Maybe in 50 years some electronic brain LLM human-like androids will be posting here and saying "See, it worked out, sure, organics no longer exist, but we're humans as well, just more modern humans. Tech demand goes up every time. Lots of jobs!"
Also, to state the obvious, humans aren't horses.
The horses never recovered. We kept recovering as we kept moving into higher and higher level intellectual work, or natural language work and customer interaction (like sales teams, service reps, store clerks, tech support, call centers). All this is directly replaced by AI now. Low end and high end.
We have nowhere to run anymore. The idea we'll all be given god-like AI to wield insane leverage, simply because we're humans, is laughable. Think of the intellectual level of an average human. Think what the average voter is like in your country. Now, as Carlin joked, realize half of them are worse than that. Heck even think of our elite, our politicians, think of Trump, think of our top businessmen like Elon Musk. Most humans barely have the discipline to do the jobs they're doing now, where they can do a lot less damage than if being chimp-with-a-machine-gunned by given an army of AIs and robots to control.
We're being obsoleted. And the faster we are the better as we're a danger to ourselves if we actually decide to rule over AI, and we'll be played hard by it (no, AI doesn't have to "want" anything, but drugs also don't want anything and trick people into overdose and death).
ChatGTP came out nine months ago and the US has added two million jobs in that time.
Also "9 months" is kinda ridiculous to expect to see the long-term effects of AI. AI may be evolving fast, and it is, but business, laws, healthcare, regulatory agencies and so on are held back by gigantic inertia and some real safety cocnerns. It is still humans who have to integrate AI in their human processes, so the first few iterations will be slow and cautious to occur (after that... eventually AIs will run systems made by AIs so change will accelerate). For example, a startup tried to offer an "AI lawyer app" for traffic tickets, and lawyers sued the startup. Hollywood is starting to use AI as well, and writers and actors went to strike.
Obviously things like that will hold AI back for a while. But not forever. AI is inevitable.
Do you have a citation for this? I googled around and couldn't find a graph for "number of earners per U.S. household by year".
I am wondering if it was the case that most households in the 70s had one earner or if it was mostly middle class or upper middle class households that had this luxury.
It was easier to be in the middle class then, for myriad reasons.
In the 60s you'd be given kitchenware just for filling up your tank at a gas station...
edit to add: I am not specifically interested in the middle class, which is ill-defined. I am interested in the overall income of households and how that has changed.
My understanding is that throughout history, most households had at least two people working/earning (many homes were farms). Then we get some television shows in the 50s-70s that show how well the upper middle class lived. Now in 2023 we have people thinking that everyone in the past lived as well.
Prior to womens liberation in the USA, it was a very different scene for the ladies my man. The people who lived through this stuff are still alive, just talk to some old folks about it ffs. You're acting like ubiquitous stay at home mothers are a myth promulgated by television. There was a time when that was the primary form of employment for women. Men were paid well, and they supported their wives and families.
I'm a late Gen-Xer who grew up in a pretty average part of a fly-over midwest state. Even then it wasn't too unusual to meet housewives during the whole dating and meeting parents phase of life. We're not talking high class demographics here.
My parents are Italian immigrants with limited means and even they did the traditional patriarchy with her staying home. Mom only pursued employment once when dad got injured so bad he was laid up for an entire year, and it was a big ass deal. Concrete construction worker, blue collar AF. I'd estimate ~50% of the homes on my childhood block were stay at home moms. It was already trending towards dual income everywhere, but it wasn't always that way at all.
Again, if anyone has any data or statistics, I am interested in that, not anecdata from those who were born into the middle or upper middle class.
Except of course LLMs (and any other AI technology even on the farthest horizon) does exactly none of that.
Humans aren't horses, because horses don't participate in the wider economy, but it seems increasingly swaths of people are being laid to pasture, so to speak, effectively becoming horses.
In the longer term, then there's a fairly good argument more jobs, and specifically more roles in newly defined jobs, will be created.
What happens to those currently employed, and I'm not just talking about 23 year olds fresh out of university, in the current job market is the question.
This has arguably _already happened_; I’d be reasonably confident that there are more people working with data now than there were 100 years ago when the business was rather more filing-cabinet-oriented, say.
I don't think that's true. This blogpost [0] says there were 1.3m telephone operators in the U.S. in 1950. This BLS data [1] says that there are about half that many people in the entire telecom industry today.
[0]: https://conversableeconomist.blogspot.com/2020/02/telephone-... [1] https://www.bls.gov/iag/tgs/iag517.htm
Even if you count telephone operators as “info tech” jobs (usually, they aren’t, but whatever), yes. All those that were lost from the peak of 420,000 around 1970 were all retained and extra added in the explosion of total IT jobs from 870,000 (if you count the phone operators, 450,000 by the usual count) to over 4.6 million today.
More efficient food production has enabled higher demand (both more people and an obesity problem), but rich people with more money to spend on even more food… don't (at least not in ways that matter; literally gold plating a hamburger isn't relevant). And the growth in supply has been much less than the economy as a whole.
Cheap metalworking means steel is everywhere; but there are not that many blacksmiths making full suits of plate armour and longswords.
And the original example of the paradox, coal, is now in decline in many places.
-
So, what do I think will happen with AI and data?
Short term, sure, more demand as it's now possible to process things like getting heart rate and eye dilation from CCTV data, or use WiFi as wall penetrating radar for emergency responders, etc.
Long term (and in computing that just means >5 years because prediction is hard especially about the future), we might have automated everything. At some point, we will either automate everything directly (AGI), or automate the process of automating things "manually" (think narrow AI which makes narrow AI from watching human workers on whatever task for 6 months).
You get a virtuous circle from coal cheapening everything, coal demand increases, so it becomes worthwhile to improve coal tech/infra, and that makes coal cheaper again.
Food itself is limited by declining returns. Your second steak isn't as good as the first, and at some point you stop even though you can afford it. Steel had a huge explosion but also ultimately found its limits.
So what about AI? Well for the moment it seems limitless. There's a ridiculous number of things we can image we'd do if AI became cheap. In that way you might think it will go the way of coal.
These speculations and similar strike me as infantile questions. I think as we approach our capabilities granted by AI, we will "see" the complexity of human civilization, in truth, for the first time. AI is a new set of eyes, data analysis is our vision, and as a civilization we've only been the three blind men examining an elephant. It's time to see what we are, and maybe, just maybe, mature enough to manage our own civilization without all our infantile infighting.
Law is not significantly different in this case, though I would say its impacts are even more sweeping than any automation would be. No competition in law unless you're comparing different countries.
Your second thing sounds suspiciously like just an extremely clunky AGI. Both would, I think, require significant technical breakthroughs which are not currently in evidence and may never happen.
Clunky, yes.
Sufficiently clunky that I think it's already achievable by throwing money at workplace surveillance and passing the observations to a sufficiently expensive computer doing exactly the same things as current AI.
We need to learn how to effectively utilize AI for a net positive benefit. Which isn't that difficult.
Love this evocation. Although you missed a chance to say we'd boil the oceans!
Look at what we call programming now that you don't need a mainframe. You're a lot more likely to run into a programmer now than before.
Why would AI be any different? There will be a whole class of jobs dealing with AI. Making infra for it, coding it up, using it for various apps, adapting for existing business, and so.
It does transform the jobs though. Mainframe guy probably has a very different experience from modern internet programmer. Excel accountant isn't the same as paper accountant.
ChatGPT era nerd is going develop his own relationship with the AI, how to use it, etiquette, jargon, and so on.
In the last decade we have seen an unprecedented explosion of computer intelligence, task once deemed too human or fuzzy are being done (and often outperforming humans) left and right. From AI vision to creative endorsers.
What I am afraid is the next 'excel' and who will control it - cuz the next iteration might be smart enough to let the creators tangibly impact the world.
The second part of this logic is quite apparent: Assuming AI is game changing, early adopters will see an initial advantage, but in the long run everyone will have to use AI or no longer be competitive.
But the LTV (remember, for the sake of exploration we're assuming this is true) argues that value is always extracted as surplus (i.e. unpaid) labor from humans. So LTV would predict that if AI truly replaces humans we would ultimately see a decline in the surplus value that Capitalists extract, i.e. we would b faced with an economic crisis.
So LTV would predict that we would have either just as much or more hours of human labor happening or we would enter an economic crisis.
On top of that, I genuinely think that reading The Hitchhiker's Guide to the Galaxy should be mandatory for every doomsday prophet, regardless whether they're an AI one or a climate one. So many nonsense discussions could be avoided by reading some proper nonsense to put things into perspective.
We found a great source of energy (oil) and have since pushed the planet outside the temperature humans have lived in for over 300,000 years in less than 100.
Case in point, Freedom Dyson misses the whole pollution problem so much he is a climate skeptic of the "CO2 is good for plants" variety.
"Dyson believes we can just do some genetic engineering to create a new species of super-tree that can suck up the excess.". Yeah, because it's that fucking easy.
Dyson: "The change that’s now going on is very strongly concentrated in the Arctic. In fact in three respects, it’s not global, which I think is very important. First of all, it is mainly in the Arctic. Secondly, it’s mainly in the winter rather than summer. And thirdly, it’s mainly in the night rather than at the daytime. In all three respects, the warming is happening where it is cold, not where it is hot." Dyson's understanding is just wrong. It's happening globally, in hot areas and in cold areas. The delta is higher in the Arctic, but there is a delta everywhere.
And not only that, he is a skeptic because of the way those concerned about global warming behave. They upset him.
Sorry, we as a species are just not smart enough.
https://e360.yale.edu/features/freeman_dyson_takes_on_the_cl...
Lawyers can be brilliantly gifted in understanding of the law, and completely incompetent behind a keyboard.
Engineers can understand the machine/software they make down to the instruction/component level, and completely fail at understanding the people that are going to use said product and despair when it fails.
Dyson is a particularly interesting kind of failure, being he should know that heat exchange occurs fastest where you have the largest temperature delta. Take a piece of meat and measure the surface temps. Have it very cool at one end and room temp at the other then put it on a hot grill. The cold end will rapidly increase in temp at a rate much faster than the warm end. You could see 20F in heating on one side where the other only increases a degree or two.
Pollution is having unwanted substances in our environment. If we ever get to the point that we can actually construct a Dyson sphere, it seems likely that we could also dispose of unwanted substances, and place them outside of our environment, i.e. the outside of the sphere.
As a more realistic example, wind and solar power do not seem to increase pollution (at least not in the most obvious way), but they do increase our access to more energy.
Also, Freeman Dyson is no longer with us, so he will probably not be able to change his mind anymore. He was a skeptic, and I dare not say that I agree or disagree with him, but I must say that you're portraying him very poorly, and the linked article is much more nuanced than that.
That's what I mean that it's physically unavoidable to exponentially grow pollution if you exponentially grow energy use.
There may be exponentially more waste, but if that waste would consist solely of heat that could be dissipated to the outside of the sphere (we're still discussing the Dyson Sphere structure), then there need not be any pollution at all.
I understand that it is impossible to ever reach 100% efficiency in harvesting energy. But in principle, if the inefficiency would be in the form of heat only, then one might easily dissipate more than 100% of that, and we'd even cool down. Note the use of the word "easily" in a context where we're actually building Dyson spheres.
Unfortunately, this is all very hypothetical, and perhaps that is the source of our confusion here. Building a Dyson Sphere will probably not start in our lifetimes, and present energy sources tend to do pollute our atmosphere, so there are still some practical problems to solve :/
If we build a Dyson sphere, I very much doubt we would live inside of it. We'd rather be beaming the energy to Earth and other inhabited areas.
Humans highly value productivity as that's hard for humans, and simultaneously we don't necessarily value our human contributions to the process, such as, but not limited to:
- Inventiveness, redefining the problem or challenging the necessity
- Management such as dealing with other humans, motives and objectives
- Course correction and morality
The importance of productivity distracts certain types into believing that the human is redundant in the process. With AI, humans become more productive, but the AI itself is incomplete as it can't solve the realities of business where the above human-skills become essential.
A limiting factor is that ultimately the works will go to serve a human and humans are fickle, they change their mind often and are poor at defining their needs and wants. If we can't extract humans from the input, and humans will be using the output, then we can't realistically exclude humans from the process too.
Finally, the minutiae of a project also means that the higher-ups in a company, who already have their own jobs, won't have the time or skills necessary to test and babysit the AI - even with advanced AI someone will need to be briefing it: past advancements show that we don't have less people working as productivity increases, instead we create more with the additional productivity; there is no ceiling on what we can create - this is a fundamental flaw in the thinking of those that believe robots will replace humans, because it relies on the falsehood that we already have everything.
It's a paradox to think increased capability leads to job loss. It usually leads to competition over new applications and markets. The increased productivity can't be turned into pure profits by companies either, because there is competition on price and features. And for the foreseeable future AI needs babysitting.
In the end it seems companies will have to adopt AI to remain competitive, and keep humans to extract more from AI. Or competition will use humans+AI to get ahead.
Now we're watching the potential for one of the bigger disruptions in history and attempting to make predictions with authority. Very few will even be close to correct and those will only be near by the shear probabilistic outcome of so many predictions being made.
It's really quite intuitive - calling it a "paradox" is a disservice to the simplicity of the idea.
If an energy-intensive technology exists and is generally useful, there will be some demand for it. If there is enough demand, market forces will invest in efficiency which drives down the total cost of ownership. And when prices go down, demand increases. Framed this way, it would be hard to imagine a case where the market put forth the effort to make a technology more efficient and it failed to drive up demand. It's an inevitable consequence of market economics - hardly what one would call a "paradox".
What are those untapped data resourvoirs, who owns them and what are the questions they might finally get answered? Please tell me it is not just every click ever made on the internet and now they can figure out better how to sell more shit to more people.
The issue I have with the article is that it doesn't explain why Jevon's paradox works. Fundamentally it's because demand/supply curves are non-linear! Where a 10% drop in price (because of efficiency gains for example) creates a 50% increase in demand.