There is also systems biology. I can't assess how good their models are. They can get better.
My point is that biology can move forward with more computational work which is based on computational models which are different than wet lab methods. In silico experiments can and do inform wet lab work by providing promising leads which direct wet lab efforts in the right direction. Now I'm back to my first argument, so it's time for me to stop.
That was the earlier point I was trying to make in the thread (I am getting my PhD in computational/systems biology). The types of computational models you're talking about are based off of very limited types of data that can be generated easily in the wet lab, and they often end up spitting out biologically implausible predictions--usually because the model is just trying to do the best it can with very limited information. To improve beyond this, there would need to be a systematic ability to generate specific real data to correct and refine the model's predictions.
Unlike a math or coding problem where you can generate the prediction and check its validity immediately, a biological model would need an unbelievable amount of
unique data requiring very different reagents and protocols to actually refine it to anything worthwhile. For example, you could get RNA-seq data for 10 million cells exposed to hundreds of different drugs, and that model will not be able to generate relevant predictions for what happens to cells exposed to e.g. a specific virus because that involves new gene programs and interactions therein. So you need to revamp your state of the art automated lab to be able to culture viruses. And you end up with a decent computational model of a specific immortalized cell line exposed to a specific virus. Maybe it even performs decently well for predicting the response of a few other similar cell types to a few other similar viruses. Which, as scientists are well aware, may or may not be useful for predicting actual in-vivo dynamics. So you revamp your state of the art automated lab to be able to breed and take care of mice, innoculate them with that specific virus (if it can infect mice at all), properly dissect a mouse without human direction, etc. etc. etc. And this is all only for RNA-seq data. You'd need to keep adding new technology to your automated lab to get any other data type, which all involves different reagents, protocols, and potential technical problems that will need troubleshooting. With the willing participation of hundreds of institutions, you can even do some limited human experiments as suggested by the AI, which confirms that a lot of the mouse model predictions were pretty far off.
Let's say that this is already the way the world is going, and pretty much all lab work will look like this anyways in a decade or two. Plus, in this scenario, we've reshaped society so that financial costs, supply chains for reagents, etc. are no issue. Where does this leave an AI that is trying to solve problems about the human body? After substantial investment in data generation and model refinement, it is very good at simulating the exact data types under the exact conditions that were in its training and fine-tuning data. Which is great, society can collectively save some money and materials on repetitive experiments. For anything unseen in its data, it can make guesses as to what experiments might be worthwhile carrying out and what we might expect to see--a substantial proportion of which are duds because the information available to make that guess is very incomplete.
So, in terms of usefulness to humanity, roughly on par with a grad student that's been given carte blanche.