Computers should be tools for the people but — hidden in a malaise of fatalism — the AI bubble has made computers tools to push people around. Soon, or in some time, this will change. We have some thoughts to share with you in the meantime.

Morro Bay, California

Preamble

The purpose of computers is human freedom. — Ted Nelson
  1. We are not “anti-AI.” We are against abuse, destruction, and exploitation: of people and our environment, as well as lies and manipulation that provide cover for the above.
  2. We are unsatisfied with the confusing, inaccurate term “artificial intelligence” (AI), but grimly accept it because it is widely-used.
  3. Our demands relate primarily to economic relationships, discourse and information software; we don’t see concerns we omit — such as regarding the environment and material created by AI (“slop”) — as less important, only out of focus; if you are prepared to complete the picture, contact us.
  4. We are enthusiastic about the real potential of AI and large language models (LLMs), but think that their capabilities have been overstated.
  5. We all have a horse in this race: from alternative technologies seeking customers to alternative worldviews seeking adherents.
  6. We think that the overriding character of AI today is that of a speculative bubble: at a scale for which any possible outcome will bring significant hardship — we must avoid a disaster like that which followed similar periods, particularly the 1920s and 30s.
  7. We think we can build a future that protects individual liberty and the general welfare by — among other things — using computers more humanely.
  8. Our manifesto will be stronger with your feedback and comments: please provide them.
  9. And, if you agree, please comment to sign your name.

Argument

1. Economics: an AI bubble

Circular financing among AI companies
We are in a bubble. Generative AI does not do the things that it’s being sold as doing, and the things it can actually do aren’t the kind of things that create business returns, automate labor, or really do much more than one extension of a cloud software platform. The money isn’t there, the users aren’t there, every company seems to lose money and some companies lose so much money that it’s impossible to tell how they’ll survive. — Ed Zitron
We have a history of net losses and may not generate sufficient revenue to achieve or sustain profitability in the future. — Anthropic (October 2026, draft S-1 filing)

The current AI bubble — perhaps best charted by Ed Zitron — is the largest in a series of tech bubbles: dot-com, crypto, with overlapping sub-bubbles like “web3” and “metaverse.” Though Open AI CEO Sam Altman has called the economic conditions around AI a bubble, many disagree, or agree that we are in a bubble but claim that this is not a problem.

Bubbles happen when excitement about a speculative investment (which can be in anything, from shares in public companies to flowers) takes on a life of its own, and investments pile in more due to momentum than the desire to build something useful.

Bubbles are neatly regressive: putting money in the pockets of those who got in early and got out early, fleecing those whose who came late to the party, with onlookers shouldering the economic fallout. Bubbles precede our most cataclysmic economic moments and the succeeding social conflicts, most obviously the roaring 20s — great depression—authoritarianism/WWII cycle.

However, many have argued that the AI bubble is a good thing, because we might get something useful out of it, such as the “dark fiber” laid down during the dot-com bubble.

There is, of course, nothing close to a guarantee that anything of the sort will happen. Datacenters, much in demand today and often cited as a useful asset that could come out of this bubble, are quite unlike fiber (which is essentially incorruptible glass buried beneath the street):

  1. Datacenters are highly specific to AI loads which may go away
  2. And wear out

Mistakes that turn out to be profitable, like dark fiber, aren’t arguments for bubbles; something either has a strong argument for it in the moment or was unjustifiable in the moment, and it paying off was merely good fortune.

Bubbles and crashes are failures of planning and capital allocation. If bubbles were actually desirable, it would be possible to find a bubble proponent who would tolerate one in an economy they control, e.g. their household or within their own company. Rather, bubbles are good only when someone else picks up the tab.

For those who think all this investment is quite sensible, i.e. that we are not in a bubble, one must ask: given that the U.S. economy would likely be in recession without AI, what are we doing this for?

Isn’t the rest of the economy — farming, transportation, healthcare, arts, etc. — the place where the actual purported benefits of AI will reach us? The gains in productivity and innovation promised by AI-proponents must be delivered to us through things. How long, therefore, should we stifle the rest of the economy while we wait to AI to pay off? And what will be left?

Bubbles hurt when they burst, but usually the sooner the burst comes the better, as bigger bubbles hurt more. Finding a way to avoid popping the bubble — such as through state intervention — stands protect those invested, while keeping the capital allocated to the wrong places.

We need to leave behind an economic regime buoyed by a series of fast-growing and capital-intensive fads, and transition to a stabler economy with broader investments and benefits.

2. Messaging: fatalism, hype, and “CEO said”

We too often let people get away with telling us what the future is. During the NFT era, many said and a few believed that NFTs would change our entire way of life.
The greatest myth of technology is the myth of determinism: the belief that technology is determinate, objective, inevitable, that the nature of computers is given. […] The computer world deals with imaginary, made-up stuff that was all made up by somebody. Everything you see was designed and put there by someone. — Ted Nelson
The most important thing about a gadget isn’t what it does; it’s who it does it for and who it does it to. — Cory Doctorow

We have internalized the narrative that the way computers work is preordained, and that it is necessary that computers work the way they do currently. We can thank this narrative for the idea that AI is the next step in computing.

This puts us in a malaise of fatalism, making it easy to believe that we must accept what we are given — including LLMs and AI — or risk being left behind. We — the people — create technology and computers; we could drive it in a different direction if we wish or abolish it altogether. If people are being left behind, other people are leaving them behind, not technology.

It’s easy to see technology as something like the laws of nature — offering us gifts like solar power or antibiotics — but it also shares a great deal with fashion: a succession of trends, driven not just by utility, but significantly by imitation and novelty. Packaged technology — like computers, operating systems, applications — has much much more to do with peers and predecessors in their category and with cultural norms regarding how we work with information than it have to do with logic and physics.

Ignore — as Cory Doctrow advises in The Reverse Centaur’s Guide to Life After AI — those who say “there is no alternative” (TINA). IBM said there was no alternative to the mainframe, and DEC proved overwise with the minicomputer and later Apple with the microcomputer; Microsoft said there is no alternative to Windows and Linus Torvalds proved otherwise with Linux.

AI fatalism rides along with two other ideas:

  1. AI will save the world (e.g. cure cancer).
  2. AI will destroy the world (e.g. kill us in a “Terminator-style” scenario).

These are really two sides of the same idea: that AI will, sometime in the future, attain awe-inspiring abilities and change the fate or our species. These claims are usually broached in terms familiar to most people but without practical specifics, or are otherwise nonsensical.

This serves to distract people from real questions relevant to AI today — whether and how to implement AI, how to handle datacenters, whom to complain to that you have to talk to a machine to get customer service, how to protect your data, how to respond to slop — and have them focus instead on fantasies.

This distraction gains much ground thanks to “CEO said” journalism, first christened by Chris Kottke, wherein journals will print nonsense if a CEO said it. Journalists have a moral obligation to challenge the facts and perspectives of their subjects, and especially the executives of influential and well-capitalized companies.

We should evaluate AI and any other technology on its merits, without distraction from fatalism, hype and nonsense.

3. Humanity: pretend AI personalities

Clippy
You’ll have these little “Max Headroom” characters with bow ties, who you have to talk to. […] Having to deal socially with inane objects that pretend to understand you, but don’t, is much worse than a cocktail party. It’s absolutely unnecessary; it’s a council of despair. — Ted Nelson

We anthropomorphize everything, and technology companies are quick to encourage this: from the original Macintosh’s jaunty mouth, to Microsoft’s Clippy.

Large language models are so good at talking that many truly believe them to be conscious or to have personalities. LLMs have neither personhood nor personality: monitoring them to use and respond to first person pronouns is deceptive — more so, say, than Clippy because LLMs are so good at talking.

Names and personal pronouns are for intelligent beings with personalities: having people use them for LLMs puts the user in a posture of equality with the machine, which, given the growing power-imbalance between the supercorporations fielding LLMs and people, puts the machine above the human in the power relationship.

Conversely, we and our corporations and governments find it just as easy to dehumanize people: sometimes through lurid oppression, sometimes merely through bureaucratic drudgery. It is not fair to expect someone hanging onto their dignity by a thread to talk to a machine as though it has a personality (more so to require them to do) in order to proceed with whatever work, government or customer service task is required of them.

The prospectus of the LLM includes a promise to help users manage information: particularly to find the information they need, or to contextualize their work. Such solutions, however, rarely actually organize the underlying information.

While people are making some progress, such as with graph databases and markdown files, AI’s insights are typically not stored (nor is the underlying information modified) as durable links/folders/database rows, to shape information for a future date or for other people. While AI can work between silos (e.g. large information repositories like Drive and SharePoint) it creates no workable, standalone connections between them.

LLMs promote personality at the expense of durability and interoperability. AI gives users the impression that something conscious is helping them to get organized, while their (and our society’s) information’s chaos deepens.

Machines are tools: they should respond predictably to unambiguous inputs. Putting a human persona on a machine does nothing but diminish the real human.

4. Information: siren servers and the new enclosure

A memorial to 40 people slain while protesting the enclosure of the common land which supported them. | Image by Richard Croft
An economy where advertisers thrive while journalists and artists struggle, reflects the values of a society more interested in deception and manipulation than in truth and beauty .— Jaron Lanier

In Who Owns the Future?, philosopher Jaron Lanier laid out his theory of the “siren server” — that modern companies have developed a new business model, which generates a competitive advantage by having more data and thus greater insights to leverage.

This is a new form of competitiveness: it is not the same advantage that was afforded, say, to IBM by its huge network of relationships, partnerships and salespeople; or its ability to spend colossal sums on research and development.

Siren servers do not compete by innovating (at least not when they’re mature) — they compete by being big. As a result, siren servers are incentivized to get as big as possible, as quickly as possible, leading to business practices directly misaligned with user needs:

  1. Free/artificially low pricing to enroll users, who are then monetized by degrading the performance of the service and/or by new fees. (Don’t confuse this tactic for the “free trial” approach; the siren server fees are a surprise.)
  2. Barriers to leaving, such as incompatibility (e.g. enterprise information management solutions that cannot talk to each other).
  3. Invasive advertising.

Some siren server systems felt like progress at the time, but with any kind of hindsight, this is not progress. As Doctrow outlines in Enshittification, would anyone claim, for instance, that:

  1. Today’s large, closed social networks would be worse if they allowed some measure of interoperability, e.g. that one could message one’s friends there without an account (as one can with email)?
  2. People like to be followed around by advertisers?
  3. People feel good about or don’t have a bad taste in their mouth from having rushed onto free/cheap platforms in the 2010s, to find those platforms turned into cattle-pens?

It’s arguable that modern LLM companies represent the perfection of the siren server model. They aggregate incomprehensible volumes of data, in order to deliver an informational advantage to their users.

As Lanier has also observed, this results in a peculiar and new type of economic displacement: of workers (e.g. translators, writers, designers, etc.) by tools that simulate them.

Lanier points to the example of translators, who have been obviated significantly by translation AIs that used their translations as raw material. The fact of simulation is critical: AI doesn’t displace people with a new tool — unlike and regardless of the moral nature, say, of replacing horses with cars or canals with trains — today, thousands of people have been and will be displaced by simulations based on their work.

Currently, no legal protections exist, as Doctrow explains in Reverse Centaur. There is much discussion of copyright, but this relates primarily to AI firms’ illegal use of copyright material, i.e. downloading without paying for it. We face a new question: how should we treat the aggregation of people’s material to form systems which then put the same people out of the job?

Throughout human history, we have created new rights to respond to new situations: working time, copyright, patents, zoning, etc. Copyright, for instance, became relevant after the printing press made reproducing texts inexpensive. In some cases, like working hours, many unions and managers have come to acceptable settlements. (Though there is still much to discuss.)

In other cases, such changes gave way to colossal and irrevocable wealth transfers. Perhaps the most relevant example was the enclosures in Britain. For centuries, poor farmers had the right to graze cattle on the commons — land then considered inferior for farming but still useful for the people to support themselves.

As agricultural technology improved, this land became profitable and, over several centuries, landowners enclosed the commons and denied commoners the right to graze. There are still commoners and there is still common land, but this displacement was large enough to cause the exodus from the countryside to the cities which peopled the industrial revolution.

Large organizations are launching an enclosure today: they are aggregating information we thought had no value, and giving it a value. We think that we should form a new settlement (for instance Lanier’s “data as labor”) which provides both for the right of individuals and groups to scrape and study information (including “training” AI) and for the people to make something of their newly valuable data.

Beyond this, we need to wake up to the fact that anything we create can potentially be used as “training data” and therefore must never again simply create (and share) something without first asking who will use it, whether we approve of that usage, and whether therefore it requires protection.

5. AI by force, and as force

An interface is humane if it is responsive to human needs and considerate of human frailties. — Jef Raskin

All tech companies are fond of making us aware of their new features. Companies are fond also of pushing unwanted new features, such as by taking up screen space previously available for other uses, or making a feature that previously stood alone accessible only via a new one. This has reached its zenith with AI.

Appealing to the idea that software should help the user, and should act as a tool, software should not interfere, block, or interpose itself between the user and their work. (This is distinct from “doing precisely what they want, immediately” because software is duty-bound to protect a user from accidentally deleting their writing, for instance.)

Software companies can make people aware of new AI (and other) features without violating the contract set out above. Meanwhile, users should have access to simple, conventional data access features: such as searching, filtering, results lists and tables when companies can make (and have previously made) them available.

Bosses also frequently insist that employees use AI, and some job applications make AI-use a criteria (rarely for the application itself, ironically). Given the harm and uncertainty wrought by AI, bosses should afford conscientious objectors respect and possibly rights, as we do for people with their views on war, social relationships, the environment, etc.

Meanwhile — as part of a broader trend to deny users access to effective customer service (hiding phone numbers/contact information, customer help labyrinths) — companies are now making customer service available only through AI, and are scaling down their human teams. Companies have every right to deploy tools to help (some people prefer AI customer support) but not to remove avenues of communication and accountability, especially in situations where only humans can provide meaningful help.

Software should help its users, companies should help their customers and respect their employees — let us not let AI confuse straightforward questions.

Demands

Therefore, we demand an end to:

  1. The AI bubble.
  2. Fatalism, hype, and to “CEO said” journalism.
  3. Pretend software personalities, possessed by AI or otherwise.
  4. Siren servers — companies that compete only through size and data aggregation.
  5. Companies forcing AI on users, to