still edge cases even then, isn't Alphafold generative?
Ok so full disclosure I am 100% not an AI or techy person and am coming from this from more of an biology/biotech side of things -which is definitely more my wheelhouse. Also you sent this as an ask so I will be yapping to my heart's content
From what I understand, even in more technical circles and even avoiding techbro classifications like AGI and ASI (which get into their own culty questions) the taxonomy about different types of AI can get really muddy.
Things like alphafold are in that grey area - mainly because the model has changed! Alphafold (2018) used deep machine learning to predict distances between amino acids. You will find some people arguing ML isn't really AI! Alphafold 2 (2021) wasn't gen AI either, its classed as discriminative AI - it isn't just making shit up, it's specifically taking an amino acid input, searching available databases and predicting likely protein structures. It's more of a puzzle solver than a creator - you still have to give it pieces. Alphafold 3 (2024) is technically gen AI - it's using a diffusion module as well to broaden the scope and speed of the molecules they can generate. It's the most grey area one and you will get a lot of people with very strong opinions about if it should be classified the same as other diffusion models like image generators and chatgpt based on arguments that are largely too technical for me to understand lmao. I will say that Alphafold is a technology that does have the potential to be incredibly beneficial in the medical field, and that several pitfalls of other genAI are avoided by not stealing people's IP, having infinitely more guardrails, and being used by people who understand it's limitations and don't treat it like an all knowing God. There are absolutely concerns and limitations with it too though!
When I was talking about non genAI I was think more about all the smaller (but often very useful) kinds of AI that get used. They're not uncommon at all, but I would say are edge cases in terms of public awareness? They often fall into the category of things people don't think of as AI, because the term is so nebulous and mysticisied.
Traffic analytics, predicting when and where demand for stock will change, flight refuelling, predictive text on your phone, the computer players in chess - are all different forms of AI They aren't generating anything, use different methods based a lot more on statistics and mathematical models, and like alphafold require humans to keep a close eye on them and make the actual judgement calls.
You get a lot of analytical and predictive AI models in academia as well! Statistical analysis and fields like biotech and genetics it's almost impossible to avoid - because you are doing such high levels of computation that it wouldn't be possible to conduct that scale of analysis without a form of ML. But it is all incredibly specific tools that are not generating anything new - k-means clustering algorithms, tools for analysing phylogenetic relationships, monitoring bodies of water to predict E.Coli levels based on turbidity/pH/temperature etc!
I was speaking to an ecologist a few months ago who was training a model to recognise and identify specific species of animals picked up in camera traps for a specific region. Often for the length of the research period you end up with 100s of GB if not TB of raw data to sift through - and being able to use model like that means you cut down the time and funding required which can be crucial of you are trying to convince governments to give you the time and money to help save endangered species
They all work differently and have vastly different impacts on the world (some are helpful, some dangerous) and need individual conversations instead of this whirlwind we're stuck on at the moment
