technology

Artificial intelligence panic

This entry was posted in Commentary and tagged AI computers technology on by

The past year or two have seen the rise of AI software that tries to mimic humans. There are AI chatbots, writers, researchers, and even songwriters. Often they seem more like artificial stupidity. One bot, Grok, has even gotten religion. People are panicking, saying artificial intelligence will take away all our jobs, make it impossible to tell reality from deepfakes, or even destroy the world. It’s far from the first tech panic in history, and unless AI really destroys the world, it won’t be the last.

AI has been around for a long time. Alan Turing wrote a paper in 1950 suggesting that machines can think. The term “artificial intelligence” was coined in 1956. Joseph Weizenbaum’s “Eliza” program, released in 1966, played at psychiatric advice. Some people even found it helpful. In the same year, Robert Heinlein’s The Moon Is a Harsh Mistress presents an intelligent computer as the master strategist in the fight for lunar independence.

Internet users use AI many times a day, often without knowing it. Search engines are driven by AI. Writers use grammar checkers that analyze their writing with AI. The technology decides which posts and ads you’ll see on Facebook. Some of these uses have been around for years.

The difference is that traditional AI works in a limited domain. Software can find pages that match queries, analyze grammar, or play chess. No one program tried to do all these things. Chatbots like ChatGPT pretend to be experts on everything. It works about as well as when people do that. They use a Large Language Model (LLM), which is a crude form of AI that ingests huge amounts of information and spits it back out as answers. Which is basically what Eliza did, but with a much larger information base. The article I linked to above called the process “an extraordinarily sophisticated game of autocomplete.”

AI data centers require huge amounts of power. That’s a clue to how they work: brute-force processing, not anything that resembles human thought.

The old rule “garbage in, garbage out” still applies. These systems are prone to “hallucinations,” i.e., invented information. When people use LLM output as trusted information, they sometimes pass along invented quotations and references to non-existent sources. Legal papers have especially had problems with made-up citations. The problem isn’t so much with using AI as with treating it as the last word. Any computer output needs to be checked for accuracy, but LLM output can feel authoritative.

AI systems have made it easier to trick people and get around security systems. A few comments submitted to this blog have looked, at first glance, like products of human thinking. Not necessarily very good thinking, but human. On a closer look, I could see that they strung words together in ways that even a stupid person wouldn’t. I don’t think I’ve unknowingly approved any AI comments, but I could be tricked. That’s relatively harmless. But if an AI impersonated someone I trusted and got me to give it confidential information, it could be bad.

Some people talk about apocalyptic scenarios. Could an AI impersonate Donald Trump and start a nuclear war? I’m less worried about that than I am about Donald Trump starting a nuclear war. Security systems need to keep up with technology, but that’s not new.

The most annoying thing about LLMs, and a big source of the hostility to AI, is using it to avoid creative thinking. Magazines are swamped with AI-written stories. AI-generated “art” is all over the Web. MASSFILC was sent an offer of AI-written filk songs, and the spammer even wanted money for them. Search for “AI funeral eulogy” on your favorite search engine, and you’ll find a bunch of offers so you can let the world know how bad a computer feels about Grandma’s death.

Another issue comes from a strange corner of the Internet, arguing that AI systems are conscious, running them is slavery, and shutting them down is murder. That’s so full of absurdities it’s hard to know where to start. What is conscious? The hardware that’s running the code? The software itself, like a spirit inhabiting the computer? We don’t know exactly how consciousness arises, but it’s an aspect of a body that experiences and interacts with the world. It isn’t a spirit that exists independently of our bodies or a demon that possesses them. Everything an AI system does is determined by code that runs on it, and the hypothesis of awareness isn’t necessary to explain anything. Computers that “wake up” are just a science fiction device. If some future system uses a “positronic brain” or whatever might enable awareness and purpose, that’s a different story.

Advancing technology presents new challenges. It always has. But the sweeping hostility to AI we’ve seen lately is just the latest tech panic.

Standard Reader and the Decentralized Web

This entry was posted in Commentary and tagged Internet standard.site technology on by

Thanks to Debbie Ohi, I’ve found a reader for sites that use Standard Site lexicons. Appropriately enough, it’s called Standard Reader. It’s still rough, but I like what it’s aiming for. You can search for publications from various sites and subscribe to them. Once you have some subscriptions, your recommendations are based on them and nothing else.

Supposedly you can log in with an existing Bluesky account. I couldn’t figure out how to do that. The first time I tried to register, I got a Captcha with incomprehensible instructions. The second time I got one that made sense. Here’s what I’ve subscribed to so far.

  • Debbie’s Blatherings – by Debbie Ridpath Ohi
  • Aaron Ross Powell
  • Standard.site
  • Atmosphere Community
  • Connected Places

Supposedly I can change my handle to my own domain, and there ought to be a way to add my site to the index; I haven’t figured that out either. Some of the tools I’ve seen are tied to Bluesky, which they shouldn’t have to be.

I’m still working on publishing this blog to the ATmosphere, as the network of AT protocol services is called. A lot of this stuff is still at the experimental level. Some of the approaches being taken won’t scale up; you can have a list of a thousand sites, but not a million. Better discovery tools will be necessary, and their maintainers will have to defend them against spammers.

Returning to the decentralized Web

This entry was posted in Commentary and tagged Internet standard.site technology on by

The original idea of the World Wide Web was decentralization of information. Anyone with an Internet connection can set up a website. Others can link to it so that people will discover it. Somehow, though, it’s lost its way. Big “social media” sites dominate it. They let you find a lot of people you know or are interested in in one place. But this advantage comes at a price. On most of these sites, the software controls which posts you see and who is recommended to you. They’re more vulnerable to government censorship than smaller, more agile sites. People complain about them but keep using them.

It’s getting worse. Some countries are banning social media access by young people. There’s talk of legally requiring people who post to these sites to disclose their identity. This will discourage the expression of dissent and intimidate people who want to discuss problems they aren’t even comfortable telling their own families about.

The solution has always been there. It’s decentralization. Blogs and forums hosted on the creators’ own machines or for a monthly fee on a server. The RSS and Atom protocols let you subscribe to most of them without handing them any information. The main problem is finding the ones you’re interested in. Some work is now being done on a set of lexicons built on the AT protocol called standard.site, which will help with discovery and identity.

A lexicon is a schema for metadata; standard.site lets you publish information about your website in a way that lets it be aggregated. If you create a record for your website, it includes at least the name and URL, and it can include things like an icon, a description, and labels. What you do with a standard.site record once you’ve made it isn’t clear to me; I assume you need to submit it to an aggregator that lets people search for sites that interest them. There could be lots of aggregators, catering to different interests. If the idea catches on, search engines might use them to improve the relevance of their results.

There are some client applications for standard.site, but I’m still looking for one that I’m comfortable with. Ironically for a discovery aid, the name makes finding information difficult; a search on “standard.site” turns up mostly irrelevant results.

If you know more or have other ideas for decentralization, let me know.

New Hampshire’s questionable alert system

This entry was posted in Commentary and tagged New Hampshire privacy technology on by

The state of New Hampshire has a system where you can sign up to get emergency alerts. In itself, that’s a good thing, but the way it’s run raises red flags. The registration site is on genasys.com, and registering requires providing more personal information than necessary. The probability is close to 1 that Genasys sells this information. It’s also easy to exploit the system to annoy people.
(more…)