News is that Regime put out a new law, bypassing Congress as usual, in the form of an “executive order”, about “AI”. The new law’s title is “Safe, Secure, and Trustworthy Artificial Intelligence“.
The new law has the Cult of Safety First!’s favorite word in it: safety. It’s got the woke favorite: equity. It’s got liberal’s favorite: civil rights.
Does it have any of our side’s beloved words? To an extent, which we’ll come to below. But this is only important if we are to trust the Regime, which we are not.
On the whole, this is yet another encroachment of Experts into an area that they have not yet insinuated their hooked tentacles. Experts in an Expertocracy never have met, and never will meet, a thing that does not need their watchful eye and super-brained guidance.
Anyway, let’s look at this new law.
First, terms. There is no such thing as “artificial intelligence.” Although if there were, Kamala Harris, who is the supposed entity behind this new law, would be the best example. There are instead two forms or kinds of “AI”: (1) surveillance and data collection, (2) statistical models. And all models only say what they are told to say. That’s it. Both are used at times in tandem, of course, such as in facial recognition predictions.
I emphasize predictions because that it exactly what these “machine learning” models are doing. Call “AI” whatever fancy name you want to give these statistical models, but they are only making guesses. Usually good ones. Sometimes excellent ones. But not always. They are imperfect as all models are imperfect.
Models (hard-coded in machines) which scan UPC codes are better than guessing which face belongs to which person, the latter obviously more complicated than the former. (Fake images and the like are also predictions in this sense.)
The advantage “AI” models have over, say, the kind of statistical models used by sociologists and the like is that “AI” models are put in predictive terms, which is why they are so good, whereas old-school stats models use an outmoded over-certainty-generating method which worships model parameters (the “guts” of models), which are of no use in any real-world application.
As an aside, we (a very small handful of us) have been trying to get old-school stats to shift to observables instead of parameters. In vain. But never mind that here.
Let’s take both kinds of “AI” as used in the new law in turn, saving the newsworthy part for last.
Surveillance
The Regime, if you can believe it, makes the right sounds here. They say things like this:
Protect Americans’ privacy by prioritizing federal support for accelerating the development and use of privacy-preserving techniques—including ones that use cutting-edge AI and that let AI systems be trained while preserving the privacy of the training data.
Strengthen privacy-preserving research and technologies, such as cryptographic tools that preserve individuals’ privacy…to promote the adoption of leading-edge privacy-preserving technologies by federal agencies.
Evaluate how agencies collect and use commercially available information…
None of this should be taken to mean the Regime itself won’t collect, and use as they like, every scrap of information on you, as the NSA does now. They mention crypto, which also sounds good, but do not forget that the Regime always demands “back doors” into it, so that they can do the obvious.
This becomes mostly air when you realize most people don’t care. After all, they carry, and even pay to carry, tracking devices wherever they go. “Credit agencies” already have most or all of your transaction data, so that they can sell “credit scores”—which themselves are predictive models. “AI”, too.
And all of this will be made a whisper of an echo of an aside when the Regime institutes fully electronic money.
Pattern Recognition
The big news was this. Pattern recognition, as everybody on our side already knows, is verboten, and cancelable, when it is engaged in for any purpose other than to DIE. It is illegal already in certain sectors, like banking. In a sense, the same algorithms forbidden there are used with wild abandon by universities and companies, but in reverse so that they can achieve full DIE.
Meaning there is, already, no common sense or consistency to “AI”. We see the same two-mindedness in the new law.
Provide clear guidance to landlords, Federal benefits programs, and federal contractors to keep AI algorithms from being used to exacerbate discrimination.
Address algorithmic discrimination through training, technical assistance, and coordination between the Department of Justice and Federal civil rights offices on best practices for investigating and prosecuting civil rights violations related to AI.
Ensure fairness throughout the criminal justice system by developing best practices on the use of AI in sentencing, parole and probation, pretrial release and detention, risk assessments, surveillance, crime forecasting and predictive policing, and forensic analysis.
The Regime does all it can to pretend not to see crime statistics, at least for the old-fashioned reason that if it can’t see what is happening, then it can pretend what is happening isn’t.
As we know, “discrimination” only works one way: Victims can’t be “discriminated” against, Oppressors can, and must be. So any algorithm that is formal—encoded in any way, or that can be discovered by statistical means by external agencies on the hunt for “discrimination”—will be ruthlessly prosecuted. The Regime is determined to DIE, and DIE it will.
Models that make excellent but unwanted predictions about Victims will be purged in just the same way as bearers of bad news used to be flayed alive.
Look for an agency, or agencies, to be staffed policing algorithms that make too-accurate Victim predictions. To avoid dings, algorithm writers will have to hard-code DIE into them. That’s the only way to get them to work.
In other words, if the algorithm is judged by its predictions, in the sense that people successfully use these predictions to make loans, approve applications, and that sort of thing, then where Victims have worse outcomes, the algorithms will be blamed. Even if writers have done their utmost to strip all obvious, and even all not-so-obvious, Victim characterizations from the algorithms.
Any characteristic that is associated with worse outcomes that Victims have greater proportions of than Oppressors must be excised with extreme prejudice. But that means everything, in the end. Which means everything has to come out, which makes the algorithms useless.
Thus the only way to beat this new law will be to do as colleges do, and select for Victims. The DIE will have be to be hard coded. The models can still do a good job on Oppressors, but Victims must be given a pass. This makes “AI” a technological kind of anarcho-tyranny.
Which is a reminder that Diversity is our weakness.
Oh, there are also words in the new law like this: “Develop principles and best practices to mitigate the harms and maximize the benefits of AI for workers”. This will be ignored. Because replacing bad workers by machines is the only hope to avoid DIEing completely.
Subscribe or donate to support this site and its wholly independent host using credit card click here. Or use the paid subscription at Substack. Cash App: $WilliamMBriggs. For Zelle, use my email: matt@wmbriggs.com, and please include yours so I know who to thank.
The real experts in AI are working to develop software that does useful things. This is yet again a case of adverse selection (the people who can't do, regulate; the people who can do have too much to do to be able to interfere with those who regulate).
If I may quote a fellow Portuguese AI researcher, Pedro Domingues, whose twitter feed is full of these gems about AI:
'The problem with Geoff Hinton and Yoshua Bengio [two of the very few people in the "regulate AI" that actually have real expertise-JCS] is that they used to feel like they were driving the deep learning train, but now it’s left them behind. From there to worrying about out-of-control AI is only a small step.'
The rest of the "Experts" are the usual public intellectuals that read about the topic of their putative expertise in a blog post on Monday for the first time and are collecting speaking fees for talking about it on Wednesday, because shamelessness and the gift of blarney, associated with a friendly media that will silence critics and disappear their previous nonsense once it becomes clear it was nonsense, are their superpower.
Again, from the Hitchhikers guide: '...The Ravenous Bugblatter Beast is so mind-bogglingly stupid that it thinks that if you can't see it, it can't see you. Therefore, the best defense against a Bugblatter Beast is to wrap a towel around your head...'