Shinycroak

I think the way to think about "AI" and literary and artistic works is that "AI" mechanically creates what the law calls "derivative works." * A derivative work is distinguished from a work influenced by previous work by a number of tests; originality is one of them. If, for instance, a large language model is asked to create a work in the style of Shakespeare, it is not creating an original work, even if the words have been rearranged; it is not capable of originality. Arguably—arguably!—works that are assembled from other works using a series of prompts might be considered at least partly original, but a great deal of the content is derived from works that were used to develop the language model. If there must be mandatory licensing of existing works to the owners of "AI" systems, let the law assign fair royalty payments to the original authors or arists. Citation * Strong, William S. “Derivative Works.” I...

If anyone actually wanted to achieve medical privacy in electronic medical communications, they could use encrypted email and Signal; there would be no third parties to disclose the information. I don’t know why so many people and medical offices so willing to trust third party information brokers like Oracle and Epic with their communications and medical records. These companies’ business model is gathering information about patients; there’s a huge conflict of interest in relying on them to transmit private information. Two major medical data firms, Oracle Health , formerly Cerner, and Epic , the provider of MyChart, have both been sued over privacy violations. MyChart requires all future privacy disputes to be handled by private arbitration, making it difficult to gain a reasonable settlement should medical privacy be breached.

One of the comparisons I've seen in defense using LLM technology to generate code, is that we after all use compilers to generate code from high level languages. But compilers are by design predictable and if they are not predictable that is an error; LLMs are stochastic in nature and there is so far no reliable way to tell when they are in error without careful examination of their output. Currently no one knows how to make LLMs predictable. There is a second problem: If your coder never tells you, "You're wrong, that doesn't work" or, "I don't think that's what you want" you will never find errors in specification; that is as true with a human coder as an LLM. There seems no way to bound the errors of LLM-generated code. I'm not even sure how to measure the errors; testing does not, cannot, do this. As Dijkstra famously obseved, "Testing shows the presence, not the absence, of bugs." The problem shifts from "Is your code corr...

Today, in a blog post , Anthropic published a claim that “Tools like Claude Code can automate the exploration and analysis phases that consume most of the effort in COBOL modernization.” They claim that Claude can be used to analyze and document existing accounting software, so that it can be more quickly modernized. Has anyone else had the disorienting experience of reading an “AI” summary and realizing that it is wrong, wrong, wrong, but very neatly worded? Why does anyone think the summaries and documentation Claude generates will be any better? And there will be no way to check them…

We know that "AI" will create superficially valid natural language texts in authoritative persuasive language. They are right a fair amount of the time, which persuades us they are reliable, but often they are wrong and sometimes people are endangered by the errors. What does it do with programming language texts? I would expect "AI" to create programs that appear to be right most of the time and sometimes just fail. A superficial examination of the generated code will not catch the problem any more than a superficial reading of the natural language texts uncovers their errors.  I feel queasy. It is certainly going to happen that an "AI" generated program will have a subtle bug that does a lot of harm, and that even will occur without malicious intent upon the part of the users of the technology, let alone deliberate malice upon the part of the owners. No one should trust a system managed by Sam Altman to produce honest answers, or take safety into account...

[I keep bringing these up on Bluesky, so I think it’s time to gather them up and make a post out of them.] This is a collection of articles on the problems of “AI.” “AI”—really, various sorts of generative machine learning models—including generative large language models (gLLMs) and generative stable diffusion models (gSDMs) so far do not live up to the promises of their marketers. A computer you can talk to is one of the great dreams of computing, and the initial releases of transformer-model based chatbots seemed to live up to this. There have been long-standing qualms about this idea, most notably Dijkstra’s argument that the imprecision of natural language was an impediment to correct thinking about computation and to accurate computing. 1 Unfortunately, so far it appears that Dijkstra was correct; gLLMs and gSDMs are notorious for errors and they are not currently designed to indicate uncertainty to their users so that people confidently rely on their erroneous output. There ar...

(A step away from my usual political posting.) There should be website ratings like: You'll never find what you want using the tools for site provides; try using a search engine instead. They usually can ship you what you want but they'll try to sell you everything in the store on your way to finding it and they treat their people and suppliers like dirt. They've got local stores with decent stock but if you go to the website it'll send so many pop-ups that you'll just wish you went to the store instead, people steal things out of their pickup bins, and if they screw up an order you will not be able to reach their customer service.

My latest smartphone update came with an easy-access chatbot button that was hard to disable; I doubt most users will bother, let alone figure out how. There is, I think, a risk of creating cognitive disabilities by providing too-easy access to chatbots, the way one can develop a physical disability by persistent restrictions of motion. To some extent, all cognitive-enhancing technologies do this; people who write don't develop oral and memory skills; people who use calculators don't learn paper and pencil computational skills, people who grow up with photography don't learn to draw. But unlike writing, calculators, and so on, chatbots are not aids to cognition but replacements for it—and a replacement that is controlled by someone else. If a child writes, there are still writing words and ideas that have at least passed through the child's mind; chatbots entirely bypass this, inserting ideas from an external source; routine chatbot use interferes with thought. There i...

If, say, I broadcast a short story on the radio, I have to license the original work. If I print a book, I have to have a license to do so. If I publish a thinly-veiled rewrite of a book without a license, that is copyright infringement. And so on and on. It ought to be copyright infringement to do that with a large language model (LLM) or, for visual art, a stable diffusion model (SDM.) LLMs and SDMs do not exist, do not operate at all, without a body of work to built the models from. Without that training data those models do not exist at all. Therefore the developer of an LLM, SDM, or any other future generative machine learning technology, ought to be required to license any work used for developing that model.

Sam Altman, who runs OpenAI, is a major investor in a firm called Helion (unlocked Bloomberg article), which claims it will be producing electricity from nuclear fusion by 2028. This is the second version of this article; physicist Stefan Urbat wrote to inform me that after 60 years there has been progress in dealing with second-order instabilities.

Why ChatGPT? Why Stable Diffusion? They are destructive in so many ways, expensive to operate, and are probably ultimately going to fail, polluting the information environment for years. They don’t even make good money. So, why?

“Artificial intelligence” researchers mean by the phrase “general intelligence” two things. One is simply a machine learning system that can, like humans, use its knowledge for many different tasks. This is not something that has currently been built; language models produce language, stable diffusion models produce images, and neither seem to have any concepts of the underlying realities these words and images reflect. The second is the “general intelligence,” g , derived from intelligence tests, which is believed by some to be a unifying reality behind intelligence test scores. In a paper on GPT-4 from Microsoft a Wall Street Journal Editorial by Linda Gottfredson, Mainstream Science on Intelligence is cited as providing a definition of general intelligence. It’s an appalling piece, repeating debunked claims about racial differences in intelligence and claiming as scientific consensus hypotheses that are at best debatable and at worst outright false.

Unfortunately, I do not know the details of the law that bears on this subject. Instead, I offer some general remarks on the technology which I hope will be useful contributions. When a painter creates a work, the brush does not hold the copyright, the painter holds the copyright. It is not different with "artificial intelligence." If a painter copies someone else's style, that's a copyright violation. It doesn't matter what tool the painter uses. If someone uses an AI model to copy an artist's writing or drawing style and publishes the result, if this does not fall under fair use, that's a copyright violation - it doesn't matter what tool is used. Because "artificial intelligence" is such an efficient violator of copyright, because there is no way to identify an AI model's sources, and because it is impossible to delete a work from an AI model it is both appropriate to insist on an opt-in model for works used to train an AI mo...

(I wrote, and then discarded, a reply to Claire Berlinski's articles on AI; she entirely believes in the TESCREAL arguments. On the way, I gathered a few links and I figured I'd record them here.)   AI researcher and science fiction author Vernor Vinge's 1993 essay, Technological Singularity , where the term was first used. For one of his fictional treatments of the subject, see his novel A Deepness in the Sky . “The Singularity: a Panel with Science Fiction Writers Vernor Vinge, Charlie Stross, Alastair Reynolds, and Karl Schroeder,” 2013. Link (video.)  “I believe that the creation of greater-than-human intelligence will occur during the next thirty years. I'll be surprised if this event occurs before 2005 or after 2030.” – Vernor Vinge  Seven years to go.

During a Mastodon conversation between Signal Foundation President Meredith Whittaker and data scientist and activist Emily Gorcenski it was pointed out that most of the techniques that are called “AI” date to the 1970s and 1980s. Some predate modern computing. Ms. Whittaker: Deep learning techniques date from the 1980s, & “AI” had been hot/cold for decades, not slow until 2012. There was no new “single idea” in 2012. What WAS new, & propelled the AI boom, was concentrated resources (data/compute) controlled by tech cos. – link Ms. Gorcenski: Backpropagation dates back to the 60s. Deep learning neural nets used to be called “group methods of data handling” and date back to 1978 or so, when the first 8-layer (polynomial) neural network was developed. Fuzzy approximation networks and radial basis function networks hail from a similar era. Weiner explored the polynomial chaos in the 40s, the Karhunen-Loeve transform predates that iirc. - link Everything that could be do...

Because of my limited knowledge, I have chosen to present these as disconnected notes rather than a more organized essay. However, I have not seen many of these thoughts before, so I hope this short note adds something new to the ongoing discourse. It’s not, really. Not intelligence, anyway. It doesn’t know truth from falsehood, or right from wrong. The technologies that are called artificial intelligence, ( diffusion models and large language models) , are basically very large grammars. They seem to replicate part of brain visual and speech centers, but no other neurological functions. In brain damaged people, there is a thing called confabulation . Confabulation is what happens when a damaged brain reaches for a memory and finds it’s not there. It just fills in the gaps. And I think that’s pretty much what an LLM does; it seems to be a replication of part of a brain but it has no real memory or logical capacity or ethics. A human author knows to fact check and not to plagiarize...

There is a tension between microblogging as social activity and reading a microblog site for news. One wants to know the official announcements of, say, the state of Texas or the Tory party. Yet if these things came from your friends or even casual contacts you would ignore or block them. This is the balance one negotiates on Twitter. To effectively replace Twitter a site or a service has to support this balance.

When FDR became president, one of his administration’s first acts was to stop a shipload of gold from leaving the USA. It was, literally, a denial of service attack on an already deflationary economy. At that time, the US economy was in such a poor state that the very wealthy thought the best thing to do was to protect their wealth by converting their banknotes to gold and removing the gold from the USA, thereby making any economic improvement impossible – a denial of service attack on the US economy. Metallic currencies have the failing of being inflexible in supply. In a financial system based on gold or silver – “specie” – the wealthy can simply cash in their chips and take them home, and the economy freezes because, literally, it is out of money. In that situation it is not that there is less labor, fewer farms, stores, or factories, or less demand – there is a shortage of money and barter becomes the only way to do business. Cryptocoins combine the problems of metallic and fiat ...

This is all very discouraging, but let me offer a few rays of hope. The objections I have raised are answerable. A centralized sign-up process, user-oriented focusing tools, social ownership, and an ethical governing organization could provide a better version of microblogging. We might even be able to preserve some of the best things about federated models. If one were to design focusing and filtering tools that worked for the users rather than the network owner, they would have both automatic and manual elements, and be subject to direct control by the user as well, which is something the commercial sites cannot allow, since users would block most advertising. But if this is going to come to pass, someone is going to have to take the lead and a great deal of resources would have to be put into this proposed system. It surprises me how short this section, and the section on ideological problems, are. Partly this is because they are a short distillation of years of experience but, a...

Your first problem, if you decide to get a Fediverse account, is which Mastodon server to connect to. I believed the Mastodon propaganda, and picked out a small site from the list at joinmastodon.org . That small site turned out to be run by fascists and does not allow one to cancel one’s account. I left and moved on to a small political site…which kicked my moderate liberal ass out for being too radical. I then decided that being one bird in a large flock was a good idea and signed up for an account at mastodon.social , the Mastodon mother site. Many people won’t get past this first barrier. If you want to try, as a general rule, avoid small Mastodon (and other Fediverse) sites unless you personally know and trust the management; some of them are run by abusers, and you sometimes can’t find this out except by joining and sustaining abuse. Others are run by various sorts of snobs. (By the way, why the devil are Mastodon messages called “toots,” a US ...