Hundreds or thousands of associated positions may point at the same general region, and each one on its own is weak. The useful move is to combine them. AI takes the full set of associations, groups the positions that fall near the same gene, and adds other kinds of evidence — whether the gene is active in the tissue affected by the disease, whether the protein it makes is the kind a drug can act on, whether related genes are already known to matter. Each gene or protein ends up with a combined score.
Ordering by that score produces a prioritized list. The top entries are the places most worth investigating first. The confidence bars next to each entry are a compact way of showing how strong the combined evidence is: a long bar means many independent lines of evidence point the same way, a short bar means the case is thin.
Two things about this list deserve emphasis. First, the ranking is relative — it says which candidates look better than others, not that the top one is correct. Second, the score reflects the evidence available, and evidence can be uneven: a gene that has been studied a lot will usually have more supporting data than one that has barely been looked at, so a low rank can mean "little is known" rather than "unimportant."