I’ve been hearing and reading a lot about AI agents lately.

Ezra Klein has been discussing them all month on his podcast, in a pretty excellent interview series. WIRED’s Will Knight wrote a newsletter last month with the headline “The Age of AI Agents Is Fast Approaching.” The general consensus seems to be that this is where we’re headed.

I have my doubts.

As background, an AI agent is a piece of software that can complete tasks on your behalf. Ezra provides a clarifying example in his most recent interview:

“The example I always use in my head is, when can I tell an A.I., my son is turning five. He loves dragons. We live in Brooklyn. Give me some options for planning his birthday party. And then, when I choose between them, can you just do it all for me? Order the cake, reserve the room, send out the invitations, whatever it might be.”

That’s a tight and evocative description. You can immediately see the appeal, right? It is, broadly speaking, rich-people-shit. One of the (many) advantages that the wealthy have over the rest of us is that they can afford a personal staff that takes care of everyday-life time sinks. Planning a kid’s birthday, figuring out travel logistics, submitting paperwork, etc. Our normal daily lives include an inordinate number of tasks that consume time and mental energy. Rich people can hire someone to handle all that stuff. The rest of us just grin and bear it.

The promise of software agents is that sometime, in the not-too-distant future, the trappings of rich-people-shit could become available to the rest of us.

I’d love to believe in that promise. I am, amongst other things, a perpetually-overwhelmed parent. If technology could reliably help me manage the day-to-day life churn, I would be thrilled.

And Klein’s reasoning is facially quite strong: A whole lot of very well-funded businesses are working quite hard to build software agents right now. The technical hurdles are comparatively small. They have (much of) the necessary technology. They have the funding. They (likely) have (some) market demand. This is not an absurd belief for Ezra to have arrived at.

But I keep being troubled by the ghosts of digital futures’ past. These promises are not new. Nicholas Negroponte and the MIT Media Lab folks were insisting that the age of software agents was imminent in the early ‘90s. Douglas Adams wrote and performed Hyperland, a “documentary of the future,” for the BBC in 1990. it featured Tom Baker as the personified software agent, dressed up as a literal butler.

Instead of software agents acting as personalized digital butlers, we ended up with algorithmic feeds and the infinite scroll.

Facebook’s algorithm is personalized, sure, but it is designed to maximize value for Facebook by keeping you within the company’s walled garden. Amazon’s algorithm is optimized to sell you the most products.

These are not digital butlers. They are digital sales associates.

And, with the benefit of hindsight, we can generalize this phenomenon: the trajectory of any new technology bends toward money.

We could have developed software agents 10, 20, 30 years ago. Software engineers were working quite hard on it. They started companies and obtained funding. The technical hurdles were comparatively small. But there was little money in it. And, in a VC-dominated marketplace, we do not get products that would be useful to the end-user unless they hold the promise of phenomenal financial returns to the investors.

We didn’t get free (or cheap) digital-butlers-for-everyone, because there was no money it.

That’s why the current wave of enthusiasm seems like a hype bubble to me. I am seeing a lot of very smart, normally insightful people being taken in by the idea of “AI personal assistants for the masses,” without asking what the revenue model is meant to resemble.

And let’s be clear: Dario Amodei is casually dropping numbers like “$5 or $10 billion” to train the next-generation models that are supposed to make these AI agents possible. That’s the financial hole these AI agents are somehow meant to fill. And that’s just for starters.

The promise of personalization in internet-futures-past went unfufilled, because the money wasn’t in personalizing to your interests. The money was in keeping you on-site, seeing targeted ads. And, again, the trajectory of the future bends toward money.

Sam Altman says we’ll have agents. Sam Altman says a lot of things. Most of what he says is tuned to what he senses people want to hear at any given juncture. But what is the revenue model for personalized agents? In particular, what is the revenue model that might convince investors over the longer term that it could go to infinity.

(Side note: a number of tech critics have been arguing that AI hype is nearly over, because it’s becoming obvious that the tech companies are spending billions to make millions. I want to gently suggest that these critics are not yet jaundiced enough. The companies are spending billions on a loss-leading product that juices their stock price and makes them worth paper-trillions. The OpenAI investment doesn’t need to generate more sales than MSFT spends on it. It just needs to keep the share price absurdly high. It’s ridiculous, and further evidence that our entire economy is just derivative financial products at this point. But that’s the absurd state of things right now.)

Agents, if they are developed at all, are going to be a bespoke, luxury good. They’ll be for discerning customers with money to spend on personalization. The business class lounge set. The rest of us will get info-sludge and degraded public services. That’s the status quo ante, at least. It’s what will happen if we don’t resist, and collectively demand a better future.

I’d like it to be otherwise. And I plan to keep a close eye on this, since it sort of represents a hard test of my broader thesis about how technologies develop.

• A lot of companies are trying to build AI agents right now. They are well funded. There is supply.

• The appeal of AI agents, if a smooth and trustworthy product can be brought to market, is undeniable. …Holy hell would it be nice if AI could make the trappings of rich-people-shit available to the rest of us, just this once.

• But we are still living in the free trial period of these technologies. The trajectory of the future bends toward money.

• So, either a market is going to develop for subsidizing these tools (packaging and reselling all of our behavioral and personal data, for instance), or the products will be rendered unaffordable to the mass public.

If you want to know where social technologies are headed, don’t focus on what the technology might be used for under ideal conditions.

Focus on the direction that currently-existing market forces will channel it.

And if that direction looks bad, exert pressure on public officials accordingly.

Maybe I’ll be wrong. Maybe I’ll revisit this post in 2029 and, with the benefit of all the time made available by my digital personal assistant, compose a thoughtful mea culpa.

But, for the time being, I would urge you to be skeptical of the promise of AI agents.

Until someone can explain how we’re going to pay these digital butlers, I’m going to assume they aren’t ever going to be available to the masses. That’s not their purpose in this story. Their purpose is to get us excited about the promise of AI, to place our faith in these tech firms (old and new) under the assumption that the benefits will be broadly distributed sometime later.

Subscribe now

This cartoon is by me and Nadine Scholtes.

TRANSCRIPT OF CARTOON

This cartoon has four panels.

PANEL 1

A man wearing a brown jacket over jeans and a v-neck t-shirt is sitting on a park bench, staring at something in his hands with great concentration. Let’s call him JACKET.

A red-headed man in a red smiley face t-shirt is on the path in front of the bench, looking at the first man with a dubious expression. Let’s call him REDHEAD.

REDHEAD: Er… Excuse me. What are you doing?

JACKET: A lot of my genius ideas get lost when I lose focus.

PANEL 2

A close-up on Jacket shows that his hands are filled with a stick, lumpy, gooey, dripping mess of green-gray ooze. He continues to stare at it with great concentration.

JACKET: So I invented “the idea net” by smooshing rubber cement, peanut butter, and used chewing gum. This way I’ll catch ideas before they escape.

PANEL 3

Redhead is responding, with a rather grumpy expression. Jacket doesn’t even glance at Redhead, continuing to study the mess in his hands.

REDHEAD: That’s gotta be the stupidest idea I’ve ever–

JACKET: I’m a billionaire.

PANEL 4

The scene has changed to an apartment. Redhead is seated on a sofa, mixing up some sticky goo in his hands. On the coffee table in front of him we can see an open peanut butter jar, an open bottle of rubber cement, and a bunch of little crumpled pieces of paper (presumably gum wrappers). He is staring at the mess in his hands and smiling.

Behind him, a blonde woman is watching what’s he’s doing with a very doubtful expression on her face.

REDHEAD: I know it looks stupid, but he’s a billionaire! His ideas must be good!

CHICKEN FAT WATCH

Chicken fat is an old cartoonists’ expression for meaningless but fun details in a cartoon.

In panel one, hidden from the humans by a bush, a squirrel in a slouch hat and trenchcoat is standing next to a magpie with a bag of nuts. The magpie and the squirrel have their backs to each other and are studious ignoring each other.

In panel three, we can see that the squirrel and magpie are looking at each other. The squirrel has opened his trenchcoat to reveal a small bag labeled “catnip.” The magpie is holding out the bag of nuts to the squirrel.

In panel four, in the background, there is an open window. The magpie has landed on the windowsill, holding the bag of catnip. Below the windowsill, a gray housecat is making the “shh” gesture with one paw, and with the other paw is offering the magpie a shiny necklace.

Also in panel four, there are a couple of framed pictures on the wall. One of them is of the blonde woman; the other one is of the cat.

The Measure of Intelligence | Patreon

In his newsletter this week, Steven Levy looks at the trajectory of ChatGPT and sees echoes of VisiCalc. VisiCalc was the “killer app” of personal computing, the forerunner to a transformative general purpose technology. And Levy has been a tech reporter for four decades. He’s one of the best in the business. He even covered VisiCalc back in ‘84.

Levy is reacting to the latest Pew Internet & American Life survey, which shows a steep and steady increase in people using ChatGPT in the workplace (8% last March, 12% in July ‘23, 20% in February ‘24). I don’t have a ton to say about the poll. Like, it’s fine, but it’s also just a single survey question. It calls to mind the Pew data on internet adoption from the late ‘90s and early ‘00s. That data was comprehensive-but-thin, and I spent a chunk of my early career criticizing how it was over-interpreted everywhere. It’s a useful data point, and all we have to go on right now, but I’d caution against putting too much weight on it.

I find the VisiCalc comparison evocative, though, because I’ve been fiddling for a couple weeks with a similar analogy: Microsoft Word. It seems to me that a lot of AI converts are convinced that the ChatGPT is on par with word processing in the early ‘90s.

(Spoiler alert: I’m gonna disagree.)

Subscribe now

I’m drawing here from a delightful James Gleick essay from 1992, “Chasing Bugs in the Electric Village.” It was originally published in the New York Times, and later republished in his 2002 book, What Just Happened?: A Chronicle from the Information Frontier. Gleick describes the excitement and frustration with the original release of Microsoft Word.

Word for Windows was a revelation. It was also a mess.

It was the first WYSIWYG (what you see is what you get) word processing program. (Amazing.) It would also occasionally-and-at-random crash your computer, erasing all of your work. (What.)

That was the tradeoff. You can use this program that simplifies a bunch of frustrating tasks. When it works well, its really great and it feels kinda like the future. But also you can’t trust the thing and occasionally there will just be a catastrophe. Caveat emptor.

(Sounds, yes, a bit like using an LLM for many workplace tasks.)

I’m sure there were cranky tech critics back then who insisted the technology was a piece of junk. Just fatally flawed. Who the hell is going to rely on a word processor for serious writing tasks when word processors randomly devour your work? And those cranky tech critics eventually looked foolish, because they were conflating a temporary problem with a systematic problem.

The thing about the bugs in early word processors and (I’m sure) VisiCalc is that users had good reason to be confident that the bugs would eventually get worked out. Sure, it’s clunky now. But that’s because these are early times. Give it five years and the software industry will work out the kinks. Microsoft’s entire business model is premised on releasing new versions of the product that make it more useful and reliable to the people actually using the product. There are billions of dollars at stake, in a marketplace whose incentives will reward the company that fixes the fatal “unrecoverable application errors.”

These are also early times for ChatGPT and other LLMs right now. But I think the comparison otherwise falls short. Because I’m not convinced the “bugs” we experience with LLMs are much like the bugs we experienced with early Word and VisiCalc.

I’m skeptical. I think its likely to be more akin to microtargeted advertising.

The thing I want to stress about microtargeted ads is that the current version is perpetually trash, and we’re always just a few years away from the bugs getting worked out.

This was true in the aughts. It was true in the teens. It is still true today.

I wrote about this example last year (in “Two failure modes of emerging technologies”), but let me recycle it: roughly once a month, Amazon.com sends me an email suggesting that I might want to buy either The MoveOn Effect or Analytic Activism. I am an especially bad target for these advertisements, because I am the author of both books.

Amazon has been in business for over twenty-five years. It has the largest, most sophisticated consumer database in the world. It is a $1.87 trillion company. And its online ad system still tries to sell authors their own books.

And it isn’t just Amazon. I am routinely served ads for products that I have already purchased. When we bought our house a few years ago, we purchased a whole-home water filter. We purchased it through the company’s website. And then I was constantly served programmatic ads for that same water filter for the next 6+ months. It cost a couple grand. It came with a warranty. There is no conceivable universe in which I’m likely to buy another one right away.

This was also a drum I kept banging back in the Cambridge Analytica days. Online advertising simply has never worked as well as its promoters claim. And we have, for decades, been caught in an endless hype-loop where critics warn that digital microtargeting is on the verge of becoming so good that it fractures democracy, even if the currently-existing products are full of errors. Give it five years and the online advertising industry will surely work out all the kinks! Except, five years later, the industry has grown but the underlying data is still crap.

There is an entire literature on this phenomenon. Maciej Ceglowski’s “the internet with a human face” is a favorite of mine, as is Jesse Frederick and Maurits Martijn’s “the new dotcom bubble is here: it’s called online advertising.” Or, if you’re up for a booklength treatment of the subject, there’s Tim Hwang’s Subprime Attention Crisis and Cathy O’Neil’s Weapons of Math Destruction and Virginia Eubanks’s Automating Inequality. (Those are just a few favorites that come immediately to mind. I could put together a whole syllabus on the topic.)

And the simple explanation of why the data underlying online advertising continues to be an ocean of garbage is that online advertising is a massive, barely-regulated industry. The money flows toward companies that can make the most compelling pitches to corporate executives, not to the companies whose products make the fewest errors.

That’s just how it is. The “bugs” in online advertising will never be corrected, because the marketplace neither demands nor rewards correction. There is enough money at stake that all the big actors have an interest in pretending everything works just fine already. …And, even if it doesn’t, we can always just focus on the imaginary product that will exist in five years.

So here’s the real question: is the marketplace for LLMs more like the marketplace for word processors or the marketplace for targeted advertisements?

I don’t know for certain. On the one hand, Sam Altman and all of his competitors recognize that AI hallucinations are a huge problem. They are indeed going to direct resources to solving it. On the other hand, as Colin Fraser suggested last month (in a barnburner essay that everyone ought to read), “Generative AI is a hammer and no one knows what is a nail.”

These may not be problems that transformer models can solve with more data. It may prove easier for the big players in the industry to just insist that the problem has been reduced, and/or it is no problem at all cometothinkofit.

That Pew survey showing 20% of employed Americans have tried ChatGPT for work gives us no insight as to whether they kept using it. We still don’t know what the revenue model of these phenomenally-expensive products will be, but I suspect it is not going to rest on individual end-users being so happy with a product that they decide to pay for it.

Ultimately, this brings me back to Ted Chiang’s essay from last year, “Will A.I. Become the New McKinsey?”

How we view the present and future of Artificial Intelligence probably turns on what we think about the current state of capitalism. Is A.I. going to inevitably improve (because markets)? Or is A.I. inevitably going to hollow out industries while providing shittier services (because markets)?

I lean toward the latter. I would like to be wrong! But I don’t think the nascent market structure of this industry is going to drive toward quality. I don’t expect the bugs to go away in five years. Not unless regulators step in, and withstand the intense pressure campaign they are already facing from tech billionaires.

So I don’t think ChatGPT is going to be much like VisiCalc. It’s just a lot more difficult for us to have and keep nice things under the market and regulatory constraints that have developed in the intervening decades.

(This is going to be a short one. I just need to vent.)

Cade Metz reported this Friday on OpenAI’s newest demo, Voice Engine. It can take a 15-second clip of you speaking and use it to recreate your voice. OpenAI isn’t making the product publicly available yet. They’re testing it to make sure its safe. And, in the meantime, they want everyone to know what a responsible and futuristic company they are.

It isn’t safe. Of course it isn’t. Are you fucking kidding me?!? The use cases here are (1) crime, (2) crime (3) CRIME, and (4) TBD.

Here’s another idea we ought to test: let’s mashup a couple of tech billionaire fantasies by sticking Sam Altman on a SpaceX rocket and firing it at Mars. The benefit to humanity is potentially immense. (Just think… at a bare minimum, we wouldn’t have to continue to wade through Altman’s bullshit anymore!)

OpenAI has been playing this game for years now. They want to both look like the leading-edge company that is building the future and also look like the responsible company that is determined to get it right.

This reflects what I believe to be Altman’s one unique skill. The guy is like if ChatGPT created a tech CEO. He speaks and behaves as though he has ingested the performances of past CEOs and is reconstituting their performances according to a script.

Altman knows that a tech CEO in 2024 is supposed to mix old-school techno-optimism with acknowledgements that the potential social impacts of his technology are so awesome that they will require regulation. (but, mind you, it has to be just-the-right-regulation. Preferably regulation that accommodates OpenAI and impedes its less-responsible competitors.) He knows that the company is supposed to always tease new products that keep us focused on what’s coming next instead of how the products work today.

As a scholar of strategic political communication, I marvel at what an impressive job Altman and the OpenAI comms team have done. With the exception of, y’know, that week when he got fired by the board, and then staged a counter-coup to overthrow the board, the company has put on an absolute clinic for how you stage and sequence product releases to maximize market share while minimizing blowback.

But that’s still ultimately just comms. The folks at OpenAI have adopted the pose of a protagonist: (1) The AI revolution is coming. (2) It will be an awesome social transformation. (3) It could turn out turn out great or terribly. (4) thank God we have OpenAI leading the way. They are the responsible ones. We’re in good hands so long as we give them all the data and all the funding and don’t bog them down with too many lawsuits or regulations.

And the thing is… No. Just no. They are not the protagonists. OpenAI are not the “good guys.” They aren’t the responsible ones. (They aren’t even a real nonprofit!)

They have crafted a marvelous story that, much like a Sora video, falls apart upon closer inspection.

They are a profit-maximizing company with a voracious appetite for money, energy, data, and compute. They are led by a CEO who embodies the ethos of Silicon Valley, but lacks the introspection to recognize why that is not entirely a compliment.

They keep rolling out these new products in order to dazzle journalists. That’s a good way to keep up the pace of positive media cycles. But it isn’t good for humanity.

The responsible way to release a product that can clone people’s voices based on a 15-second sound clip is to not create that product in the first place. You don’t need a ton of user-testing to figure out the obvious harms. Bad people will use this for bad ends! The bad will so heavily outweigh the good!

The only reason you should even be developing this product in the first place would be to design adversarial tools that would instantly break it. Or to design watermarking/fingerprinting technology that could help governments immediately find people who deploy something similar, and then give those people a seat on that SpaceX rocket-to-Mars right next to Altman.

It is 2024. The hottest year on record. There’s like a 50/50 chance that an authoritarian demagogue will take over the government later this year and completely erase the administrative state. There are so very many things that good people — real protagonists — could be doing this year in order to improve the world. And what is OpenAI doing? Burning massive energy (not to mention the water costs) to train a neural network to spoof your voice.

There are worse companies than OpenAI. But that doesn’t mean the folks at OpenAI are the good guys. The company is putting so much effort into maintaining main-character-energy.

So I just want to be very clear about this. They might manage to be one of the main characters of 2024. But they definitely are not the protagonist.

Subscribe now

A cartoon by me and R.E. Ryan.

TRANSCRIPT OF CARTOON

This cartoon has four panels. All the panels show a gritty commercial doorway – the kind that’s recessed a few feet into the building – on a city sidewalk. There’s litter and graffiti here.

There are two characters in the comic strip. The first character is a homeless man sleeping in the doorway, wearing a zip-up sweatshirt over a t-shirt and a dull red knit cap, and with a full beard.  The other character is a muscular-looking cop dressed in a police uniform and carrying a baton. In defiance of tradition, he is cleanshaven. I’ll call these two characters KNITCAP and COP.

PANEL 1

Knitcap, covered by a brown blanket and with his head pillowed on some rolled-up clothes, is lying in a doorway, apparently asleep. The cop is using his baton to poke knitcap in the side. The cop has a somewhat sadistic grin.

COP: Hey, you! Get up! We’ve outlawed sleeping in public! You’re not allowed anymore!

PANEL 2

Knitcap is sitting up, rubbing sleep out of his eyes with one hand. He speaks calmly. The cop watches, smirking, arms akimbo.

KNITCAP: In that case, I guess I’ll sleep in a hotel tonight.

PANEL 3

A close-up of Knitcap. He’s stroking his chin with a hand, as if thinking through his options.

KNITCAP: Or should I sleep in my townhouse instead? Or my Hamptons place? I’ll call my butler and ask what he thinks!

PANEL 4

Knitcap, grinning, is now holding a hand next to his face, thumb and pinky finger extended, pretending it’s a phone as he talks. The cop is glaring and slapping his baton against his palm.

KNITCAP: Smithers? Smithers old boy! My super fun street sleeping holiday is done. Which of my mansions shall I sleep in tonight.

COP (thought): Next step: Outlaw sarcasm.

CHICKEN FAT WATCH

Chicken fat are unimportant but fun details cartoonists sometimes sneak into comic strips.

In panel one, in the lower-right-hand corner of the panel, two rats are sitting, holding playing cards and apparently playing poker, or some similar card game. In panel two, a cat walks in, apparently stalking the rats. The rats look at the cat. And in panel four, the cat has been dealt in and is playing the game with them.

In all the panels, Knitcap is wearing a t-shirt with some words that are hard to make out. But what it says is “No, you’re Spartacus.”

In panel three, there’s a lot of mostly-unintelligible graffiti, but just below the doorknob someone has painted “BACKGROUND DETAILS RULZ.”

Let’s Outlaw Being Homeless! That’ll Work! | Patreon