As with much of this author's content this is a strong opinion that lacks nuance, but I basically agree with the fundamental assertion: that the lasting impact of this AI bubble will be to further centralise power, taking it away from workers.
My hope is that a desire for authenticity prevents this from happening – whether that's a strong bias towards human content creators, towards speaking to a human on the phone for customer support (already something companies try to win customers on), or even winning customers on well-paid humans cooking their food for them (something that seems to be increasing).
Unfortunately, I suspect we will get a two-tiered system, where the "middle class" (whether that's disappearing is another question) can afford human content/human support/etc, and the working class are forced to endure poor experiences with AI generated content and so on. This may even get worse over time if, say, AI hits education and provides a worse quality education, but that's probably no different to what we already have with public school funding issues in the US/UK and many other countries.
> Unfortunately, I suspect we will get a two-tiered system, where the "middle class" (whether that's disappearing is another question) can afford human content/human support/etc, and the working class are forced to endure poor experiences with AI generated content and so on
I’m imagining world where companies like Netflix and Spotify introduce dirt-cheap subscription tiers that are populate with AI generated content, while they raise the prices on their existing offerings that have stuff made by humans.
If you’re poor you watch shitty, AI-generated movies on Netflix for $1.50/mo.
«The literature that the Ministry of Truth produced for the proles was of "a lower level," and consisted of "rubbishy newspapers containing almost nothing except sport, crime and astrology, sensational five-cent novelettes [and] films oozing with sex."» [George Orwell, "1984"]
In fairness I don't think we've needed any AI or even state help in achieving this but it stands to convert some recently loss making outlets into profitable entities again.
> If you’re poor you watch shitty, AI-generated movies on Netflix for $1.50/mo.
I suspect even worse scenario. If you're poor you watch shitty, ai-generated content for the current price. If you're persuaded to belive that you're middle class you cough up twice more to watch shitty, ai-generated content called premium becuase it will be "artisanely curated by humans (tm)". If you're really rich you will go to the theatre or opera.
> If you're really rich you will go to the theatre or opera.
Mm, Shakespeare.
'tis easy mimic'd in style and verse; and the many devices used, familiar. When anachronism strikes, few yet notice, and fewer yet call out, for the Bard is older to us than he to Chaucer.
What folly, that the rich dress and peacock themselves so, sitting quiet and polite; gathered at great expense to what was, in days yore, the entertainment of yeoman and serf who cheered and jeered as saint and villains pranced before?
Hark, though; I say not that Avon brought no talent, rather that the talent of The Globe was to his time as the talent of The Disk to ours, as the easy-read and prolific prose of Pratchett is the closer cousin, despite the esteem of the powerful going to those hardest to follow in modern vernacular — Ulysses, War and Peace, Crime and Punishment, and yea, also the Bard and King James.
I'm seeing a world where kids are put in a room with an AI that "educates" them, setting the lowest possible bar for personal development for your average kid and as quickly as possible expelling kids who are deemed to be a problem.
Why would the poor not be able to afford human generated content? Copying human generated context costs basically nothing, so the marginal cost of letting the poor view it is basically nothing.
I'm old enough to remember when operating systems cost money[0].
And compliers. And encyclopaedias. And maps.
I've got too much good zero-cost audio and video content to get through, even at double speed.
[0] MacOS 8, I don't remember what I spent on it (UK), but wikipedia says it cost $99 in the US when it came out in 1997. Inflation adjusted, $188.56 today.
> How do you explain why more recent movies are more expensive to buy or rent on streaming services, then?
Artificial price inflation and recouping costs combined. Fueled by FOMO, people tend to pay more to be able to access it and be up-to-date (TM) in their social circles.
They'll probably recoup the costs without inflated prices, but if they can exploit that title for more money, they'll do it.
Creating human-generated content will cost a lot more, so it will be paywalled. This pattern already exists, limitless machine crap will just make the differential greater.
Kinda interesting that facebook is switching from "real people" generated content (i.e. someone you more or less know) to "random crap" content (not even the tiktok algo). It is like shooting themself into a leg.
People just make content that get views and optimize what gets eyeballs. Take people out of the equation and add gradient descent and who knows what insanity will be unleashed.
Because it's upper limit is the best human created content. It doesn't create anything out of thin air. Just "emitting" things mixed from its training set.
Also, a machine cannot create equally complex or more complex than itself, so AI is always capped at human capacity, at most, asymptotically.
It can be faster. It can batch process, but it can't process "like a human".
Well, nature's laws are pretty resilient as far as I can see.
Laws about thermodynamics, and entropy still hold. Also, as far as I can see, no living creature, incl The Nature itself succeeded to create something more complex and sophisticated than itself.
>> Unfortunately, I suspect we will get a two-tiered system, where the "middle class" (whether that's disappearing is another question) can afford human content/human support/etc, and the working class are forced to endure poor experiences with AI generated content and so on.
Due to the economics of information, this is unlikely for contents. The cost to watch Avatar is often quite close to a local film with <1% its budget. The most interesting clips made by content creators (Ted Talks, Veritasium, MrBeast, etc.) are accessible to everyone with internet access. There will be some price differences based on access points (theater vs TV vs phone) or resolution but not necessarily the content itself.
I also suspect the best contents in the future will be co-created by humans and AI.
This is a fair point, a different way of looking at things.
The way I was thinking about it was, for example, newspapers going online behind paywalls, YouTubers/Podcasters creating paid content on Patreon in order to fund their activities.
On the other side you get Buzzfeed creating AI generated quizzes (rather than news), and YouTube/TikTok content farms with AI generated scripts. Both of these are ad supported, so free to the end consumer, and therefore more accessible than a Patreon/NYT/etc subscription.
Articles from The Economist, The Wall Street Journal and The New York Times are better and also more expensive to produce than those from Buzzfeed, which justify their subscription fees. What if editors of such caliber could leverage AI-assisted tools to scale their efforts and create high-quality content at a lower cost?
While it's true that, even with advancements in AI, top-tier content uniquely shaped by the individuals or teams behind it will still be more costly to produce, these are likely to get cheaper over time as AI improves. As a result, high-quality content will probably become more accessible.
AI is the only way for small countries to compete with Hollywood. I'm sick and tired of all those glorified stage plays that my country produces simply because we have no access to Hollywood money.
Meanwhile Hollywood has so much money that they don't know what to do with it - they give it to directors like Michael Bay to create all those pointless cgi-fests.
When you say "to compete with Hollywood" you either mean one of two things: on quality or on box office.
The latter is nothing to do with CGI or production budget: it's about marketing budget, in which AI has less to offer.
The former, in my view, will not be aided by AI. The "glorified stage plays" are (not always but on aggregate) of significantly higher quality than quite a lot of Hollywood output. AI won't change that in any positive way.
Say you want to create a historical drama that doesn't look like crap. You have two options. Either create massive physical sets or (as is increasingly popular in Hollywood) CGI ones. Both cost an enormous amount of money. It isn't just tedious popcorn movies about superheroes or aliens that require massive budgets for effects.
This sounds a bit like District 9 (albeit having Peter Jackson as producer might exclude it) - the director then "graduated" to Hollywood, resulting in a decline in critical reception (though I've heard Gran Turismo is pretty good by video game movie standards).
If it comes to pass that AI dramatically reduces the cost of making a movie that works for Hollywood as well as small countries and probably doesn't result in the small countries being able to compete as a result?
In the same sense that the revolution that occurred in the digital distribution of games and general improvements in the accessibility of tooling meant the market for games expanded into more niches rather than producing more AAA games from smaller developers.
If a local director could make a movie with comparable quality to Hollywood movies - I would watch his movie. Hollywood only dominates, because their movies have way higher production quality.
There is only so much you can do with shoestring budgets that our directors have.
My question to you is: why do people spend money on those glorified stage plays and not just spend all their money on Hollywood movies? It's likely that Hollywood movies just don't align perfectly with the local culture due to limits/costs on localization. Now imagine if Hollywood movies had perfect dubbing (including changing scenes and lip movements), local references and jokes, removed cultural mismatches, etc. Would that make people less or more likely to spend their money on them?
We watch mostly Hollywood movies. Local output has a pretty low market share. Both due to low budgets and the fact that Hollywood can make a lot more movies per year.
> Unfortunately, I suspect we will get a two-tiered system, where the "middle class" (whether that's disappearing is another question) can afford human content/human support/etc, and the working class are forced to endure poor experiences with AI generated content and so on
Have you dealt with human support agents recently? It's just an exercise in gaslighting. I can't wait for AI to take over.
No. That is precisely the wrong attitude. It is the 'corporate' drive to push service to be cheaper (for them) that lead to outsourced and IP poor user services, which AI will take to a whole new level of bad.
What should be desired, is a return to quality and respecting the user (or at least taking the time to understand them), which unfortunately seems unnecessary now with a global reach and unimpeded manipulation and influence that can ignore demands for improvement.
> What should be desired, is a return to quality and respecting the user (or at least taking the time to understand them), which unfortunately seems unnecessary now with a global reach and unimpeded manipulation and influence that can ignore demands for improvement.
I know the assumption is always that Big Greedy Co™ is hiring the worst possible employees to save money, but I think it's far more likely that the scale of work required far exceeds the labour required to provide it.
Customer support being a bad experience has always to do with the company policies or a general lack of training.
For company policies, an AI would be rigidly trained and limited to always minimize loses to the company, since you can game the AI if it's too lax by saying the exact prompts or keywords.
For companies that do not even provide basic training except for a FAQ sheet, I do not think replacing the human with an AI is going to improve customer experience, because, a human(IC or manager) might be driven by motivation of compensation or job security, to learn more than what is provided, to do their job well.
What does AI do if it doesn't have a good answer? How do you know it gave you a good answer? There are no qualities like 'that person sounded a bit useless' or 'sounds like they're reading from a script' or any of those other wonderful interactions we have with poor customer service was can judge with sentiment/intuition. You just get 'the AI answer.'
Big greed co will hire the cheapest employees, and will do exactly the minimum amount of customer service they can get away with, so long as their profits are sustained.
I submitted a bug yesterday and got an automated response asking if I'd tried turning it off and on again... now that's not AI but I thought it was funny.
I asked it to forward my email to their devs, human or not ;)
I just switched my banking from the largest national institution in Australia to a local credit union and another bank that has always had an online/phone only model and is good at it. I get the best combination of human customer service and good value accounts and leave behind the abysmal contempt for customer service demonstrated by the major, with all its infuriating new AI mediocrity. I can't be the only one?
Not everyone has the grit to hold up to the customer retention tactics that companies use once they learn you’re leaving. Offering a short term discount or a last-ditch attempt to look into the customer’s otherwise long neglected issue costs much less and retains most customers out there.
You of course already have this for the most part, it's just "premier support" vs "faq that doesn't answer you question" for most people. It's just the "premier support" providers will be even more thinned out than they already were.
I'm not entirely sure about the centralization part. We already have models that can run on consumer hardware and are freely available to use for anyone (e.g. code-llama 34B is actually a viable gpt 3-5 replacement, if not slightly better)
Training these is still out of reach, but fine tuning is getting close (LoRA) and running them is almost easy at this point.
The products we've built so far are power-centralized, but augmentative. Where we go from there is up to people, not the nature of the technology. My hope is decentralized and augmentative, but the worst case scenario is indeed centralized and substitutive.
By centralisation I don't necessarily mean of the models. It might be the training, training your own model requires expensive hardware (even just a high end graphics card is out of reach for most of the world).
But also, a model running on your phone generating AI content is likely to be cheaper and not as good as human curated content in whatever form that is.
I think compute can be decentralised, while the power is still centralised, or at least those at the low end lose out on quality.
There was a time where computers took up the space of whole rooms and had much much MUCH less processing power than the phone I am currently writing this comment on...
oh, believe me, regulations would be lobbied. I.e. to run an AI model as a part of some service, you would need to get certification (a lot of money). Even now you could see a couple of talks on tech conferences from "non-profit" organisations. Also, web services are highly centralised, so there are all the chances to get highly centralised AI services too.
Hatred of phone calls aside, I heard an interesting take from Alex Hormozi (paraphrased): "You're on the phone and you say, are you human? No? Oh thank God, because you know the AI has a thousand times more experience than all the humans combined."
Is that supposed to be a serious opinion about contemporary AI..?
Because that's crazy to me. Humans can be reasoned with. AI can't. My experience with the likes of ChatGPT tells me that if the AI is wrong about something (which it very very often is), there's no point trying to explain to it that it's wrong or how it's wrong, it will say something like "You are right, sorry for the confusion." and follow up with the same or a similar error again.
AI might eventually become an alright first line, but losing the option to speak with a human -- an intelligent entity which can actually be reasoned with -- seems dystopic.
Having worked in customer service before, the only time customers tried to reason with me was when they were wrong. For at least a large portion of customer service calls, the ability to reason is a negative.
Customer calls to say the product is faulty, and with a simple question they admit they didn’t connect the ground wire. They then try to reason with me that they deserve a warranty replacement because their mum just got cancer - fuck reason on cs calls. (Reason might still have a place on complex situations [which yours never is], customer retainment, sales etc)
Even as the top tier tech support for a complex product, 99% of my calls could have been dealt with well by chat gpt in its current form. And my customers would have a more concise outcome, without the variability of my mood, hold times, judgement of their tone etc to effect my advice.
I can’t wait for more ai cs so I can stop talking to a fuckwhit, and talk to an ai that’s almost certainly far more equipped to handle my query.
I'm not talking about a customer arguing that they deserve warranty. I mean if the human CS agent misunderstands the problem, the customer can explain the problem better and try to make the human CS agent understand. If the "AI" CS agent misunderstands the problem, there is no recourse.
> Is that supposed to be a serious opinion about contemporary AI..?
No. We're talking about the future. ChatGPT doesn't have millions of hours of customer service experience. It has zero!
(Well, unless you count the RLHF stuff, but my point is it isn't actually learning from human feedback, although we don't actually know what they're doing with those thumbs up / down buttons...)
> losing the option to speak with a human
Already the case at many companies, sadly. (In those cases, if not elsewhere, it's going to be a significant improvement.)
I think that's a point a lot of people are missing. If AI can already talk about certain topics better than most humans, imagine in the future. There may be a time when talking to humans may become really underwhelming compared to all-knowing AIs.
Why have all-knowing customer-support AIs waste their time and resources talking with underwhelming human customers such as myself? It would be much more efficient for customer support calls to be made by all-knowing personal assistant AIs, which can surely explain the nature of the problem much better than me, as well as being better able to infallibly put in practice whatever the all-knowing customer-support AI suggested.
The AI will have more experience composing bullshit put-offs, but even less access to the systems that control the systems that are exhibiting problems customers currently experience. We'll see the problem of layered customer-service (where the first two layers have no power, other than to maybe turn on some pre-configured bonus/offset/coupon) made worse by further automation.
Maybe in future. Right now it's infuriating and miserable, to the point I will absolutely switch any service provider the moment they subject me to this torture. Old systems with "Press 1 to talk to a person" were bad enough, the AI-based ones are so much worse.
My hope is that a desire for authenticity prevents this from happening – whether that's a strong bias towards human content creators, towards speaking to a human on the phone for customer support (already something companies try to win customers on), or even winning customers on well-paid humans cooking their food for them (something that seems to be increasing).
Unfortunately, I suspect we will get a two-tiered system, where the "middle class" (whether that's disappearing is another question) can afford human content/human support/etc, and the working class are forced to endure poor experiences with AI generated content and so on. This may even get worse over time if, say, AI hits education and provides a worse quality education, but that's probably no different to what we already have with public school funding issues in the US/UK and many other countries.