Avatar

¢vl† ov Nvll

@cvltovnvll

Be it a hypersigil
Posts
cyborg-alchemist
Avatar
river-taxbird

There is no such thing as AI.

How to help the non technical and less online people in your life navigate the latest techbro grift.

I've seen other people say stuff to this effect but it's worth reiterating. Today in class, my professor was talking about a news article where a celebrity's likeness was used in an ai image without their permission. Then she mentioned a guest lecture about how AI is going to help finance professionals. Then I pointed out, those two things aren't really related.

The term AI is being used to obfuscate details about multiple semi-related technologies.

Traditionally in sci-fi, AI means artificial general intelligence like Data from star trek, or the terminator. This, I shouldn't need to say, doesn't exist. Techbros use the term AI to trick investors into funding their projects. It's largely a grift.

What is the term AI being used to obfuscate?

If you want to help the less online and less tech literate people in your life navigate the hype around AI, the best way to do it is to encourage them to change their language around AI topics.

By calling these technologies what they really are, and encouraging the people around us to know the real names, we can help lift the veil, kill the hype, and keep people safe from scams. Here are some starting points, which I am just pulling from Wikipedia. I'd highly encourage you to do your own research.

Machine learning (ML): is an umbrella term for solving problems for which development of algorithms by human programmers would be cost-prohibitive, and instead the problems are solved by helping machines "discover" their "own" algorithms, without needing to be explicitly told what to do by any human-developed algorithms. (This is the basis of most technologically people call AI)

Language model: (LM or LLM) is a probabilistic model of a natural language that can generate probabilities of a series of words, based on text corpora in one or multiple languages it was trained on. (This would be your ChatGPT.)

Generative adversarial network (GAN): is a class of machine learning framework and a prominent framework for approaching generative AI. In a GAN, two neural networks contest with each other in the form of a zero-sum game, where one agent's gain is another agent's loss. (This is the source of AI images and deepfakes.

I know these terms are more technical, but they are also more accurate, and they can easily be explained in a way non-technical people can understand. The grifters are using language to give this technology its power, so we can use language to take it's power away and let people see it for what it really is.

Avatar
slmdrs

If you want to get under tech bros skin, just call it applied statistics.

Machine learning - and by extension language models and GANs - is applied statistics. Most people who build machine learning models know more about statistics than they do about the computers their models run on. So much so that the companies developing these models have to hire programmers to connect the model to an interface ordinary people can use because the people building the models can't do it themselves.

Additionally, knowing that machine learning is just applied statistics, you can also extrapolate everything you know about statistics to machine learning. For instance, in statistics, you have to carefully pick your sample to reduce the likelihood of a confounding variable throwing your results askew. In machine learning, your sample is your dataset, and frequently datasets are either a) picked lazily, b) too small, or c) outright stolen.

Lazily picked datasets can be polluted with data points completely irrelevant to the desired goal. An example is a model being trained to pick the best candidate for a job, but the dataset includes unrelated information such as an applicant's name, gender, or race. These are confounding variables that can make a model perpetuate harmful patterns such as racism and sexism.

Small datasets, even if they are free of confounding variables, are also a hazard since a small sample is unlikely to be representative. Take every attempt companies like Google or IBM have made to make a model that can diagnose specific diseases. Each of these models, when put in the field, frequently failed to diagnose the disease reliably. Even the object-detection models in self-driving cars have this problem, with some (Tesla) not being able to tell the difference between a person or a road sign.

Stolen datasets (or datasets containing stolen data) are an attempt to overcome the small dataset problem by building massive datasets containing every possible, valid, input. And as we've seen with GANs and LLMs such as Midjourney or ChatGPT, a lot of that material is simply yanked from the Internet regardless of consent or copyright. And that's not to speak about the fact that these datasets are often so huge that no one has the time to check for and remove confounding variables, so nascent patterns in the data can manifest in the model's output. (This is how people figured out GANs were trained on stolen art since the GANs can't help but replicate the images they were given.)

Avatar
grison-in-space

I also gotta say, those giant stolen datasets are so vulnerable to data poisoning (as the GLAZE program I've seen waved around does) or even simply incorporating unexpectedly undesirable input that shapes the future outputs. The general hope of the applied statisticians trying to build algorithms and models from these giant, stolen training sets is that there's some kind of underlying throughline pattern that can be divorced from the noise and then re-noised to create new outputs, if only you submit enough samples. But what they consistently keep finding is that if you don't pay attention to identifying and understanding real effects or patterns in the data, they appear in your outputs, too... which is why the ChatGPT "story-builders" have such a predilection for omegaverse.

In that way, it's very much like research. There are certainly scientists in Big Data-driven disciplines who think that feeding more and more and more raw samples into their models will teach us something useful about the infinitely complex natural world in which we dwell. They have been desperately trying to discover something new for like the last ten years, and overall their labors are yielding surprisingly little in the way of useful information. Start small and pay attention to the little pieces. You don't get understanding from unsupervised machine learning, you get it from understanding one piece of a giant model of the infinite complexity at a time.