This is not accurate for evolution or AI. For evolution, not all animals evolved long necks because evolution is a random search on a large, constantly changing

This is not accurate for evolution or AI. For evolution, not all animals evolved long necks because evolution is a random search on a large, constantly changing space. Giraffes are also not the only animals with long necks. We have no reason to believe no other animals will evolve long necks in the future either. For AI, stating that we throw out 99.9% of the information is wrong or misguided depending on interpretation. A lot of people are fixated on ML 101 belief that models are black boxes and impossible to unravel. This really isn't true, and significant progress on explainability tooling has been made in the past several years. Most of which (by which I mean none that I'm aware of) require changing the underlying AI model. 2 examples: 1) The well known 20 questions genie, akinator. Not really ML, but works similarly to decision trees. You have a space of information. The underlying tree model learns what the most relevant questions are to divide the space to optimize entropy for information gain. 100% explainable. Surprisingly accurate for many people. https://en.akinator.com/ https://en.akinator.com/ 2) Shap values for AI models are one example of tooling to make ai models more interpretable. They show you which features mattered for a single prediction, and by how much in which direction. If we imagine a decision tree again, we can assume that this bears a strong correlation to the path followed through the tree and the associated information gain at each step. If we imagine a complex image recognition neural net, we can highlight which pixels in an image contributed to the outcome the most. It wouldn't be terribly surprising if shap is the tool being used by google here specifically. Focusing on a few important features rather than considering every single is ideal. If I, for example, give you an extensive fecal analysis, psychological, infared, radiological, psychic medium prediction, and survey results from 5 normal people who got to look at the animal- and I asked you to identify if it was a cow or not... It would be appropriate to just say "survey said cow, so its a cow".