What is AI not?

AI is not magic¶

Biology is much more magical than anything in a computer.

There is still a lot about biology that we do not know and biological systems are a part of the natural world, that we study through science formally and that we learn informally about through senses during our lived experiences. AI is a part of a computer, it is designed so, at some level every bit of it is known, if not all well understood at all possible levels of abstraction.

Generative AI is not automation¶

Machines can automate things. In general, people expect automated processes to be highly reliable. Generative AI is designed to generate text based on a pattern that starts with a prompt.

Automation can save a lot of time, but inserting a generative model in the middle may insert randomness into processes that do not need any. There are a lot of problems that, with a little bit of care, could be reliably automated but a Generative AI tool is probably not the best way.

Generative AI is not an authoritative source for factual knowledge¶

It will not give the same response to a given prompt every time.

LLMs will make mistakes.

They meet the philosophical definion of Bullshit Hicks et al. (2024)

Generative AI is not good at math¶

Fundamentally, an LLM cannot do any actual calculations, unless it is an agent than can operate a tool (like a calculator). They sometimes provide and can be strategically prompted to give the right answer and special training procedures have shown progress in this.

Generative AI is not search¶

Traditional, strict search algorithms, a type of algorithm for finding an item in a set. Modern, large scale search engines, like Google, have been machine learning- based since their inception. They were learning what page the person was most likely to choose to read[^click] given the query.

Generative AI, at base is more like a compressed copy of the internet that is then expanded to produce an answer, randomly because it is lossy compression^[1].