Okay, stop. Stop. This is called statistics. People have been using it to analyze the human genome for decades, since we first had the sequence. I wrote my thesis on such work a decade ago. This is not "artificial intelligence". Enough already you clods.
While it's true that deep learning reduces to "curve-fitting with stochastic gradient descent," in practice, working with deep neural nets is and feels qualitatively different, even when they are used as universal function approximators within a commonly used Bayesian framework, as is the case here.
Unlike the simpler statistical models of yore that identify easy-to-understand (e.g., monotonic) relationships between variables, deep neural nets can find relationships that in practice cannot be articulated except on their own terms -- that is, they can be articulated only as long sequences of non-linear transformations.
In my view, the word "intelligent" is probably the best we have today for describing models that can learn to recognize patterns in data at such high levels of abstraction, even if this kind of intelligence is very narrow and only of a perceptual nature (i.e., there's no reasoning, planning, creativity, etc.).
Things like Breiman’s ACE (alternating conditional expectation) are also qualitatively different, in many ways more magical, but it is still just statistics.
AI has a hard time telling cows apart from traffic lights under the wrong lighting conditions.
So we have AI looking at a data set, and it notices this anolmalous, shadowy outline of an interesting shape in the data. This AI is the same class of software that can be duped into misclassifying a face for a celebrity with some abstract, bold halloween makeup.
So, do we trust that everything interesting the AI picks up on is assuredly a new finding?
How do we know that the AI hasn’t mistaken a potential pathway for mutant future descendants as ancestral artifacts? Maybe it has noticed something that can’t happen yet, which could be a possible future reproductive vector?
Or, maybe it really is an ancestral artefact, but of something that never happened. Just a weird coicidence of justified end-of-line characters all lining up perfectly, when you drop certain rows.
100% agree. Worse; using deep learning to "find" alleged non-human ancestors is probably a case study in baloney "research" results.
Go look at how Neanderthal admixture was discovered by Patterson's group; they used F-tests. If you can't find it using an F-test, you can't find it using dweeb learning, which is always going to be less data efficient than a simple hypothesis testing technique.
Layman logic dictates that math-related anything lowers clickbait potential.
Unfortunately, the author has a vested interest in getting as many people to _read_ their story as possible - especially so if they're a self-described "freelance writer" like this one.
