That should always depend on the project at hand: if the project is primarily in a specific domain of AI research, then you need reviewers working precisely in that particular domain of AI; if it’s in ethics, then you need experts working in ethics; if it’s interdisciplinary, then you try to get reviewers from the respective fields. This also shows that it will be rather difficult (if not impossible) to have an expert team competent to evaluate each candidate project. Instead, the team should be competent in selecting the adequate expert reviewers (similarly to journal editors who invite expert reviewers for individual papers submitted to the journal). Of course, the team can do the pre-selection of projects, determining which are worthy of sending for expert review, but for that, it’s usually useful to have at least some experience with research in one of the relevant domains, as well as with research proposals.
That should always depend on the project at hand: if the project is primarily in a specific domain of AI research, then you need reviewers working precisely in that particular domain of AI; if it’s in ethics, then you need experts working in ethics; if it’s interdisciplinary, then you try to get reviewers from the respective fields. This also shows that it will be rather difficult (if not impossible) to have an expert team competent to evaluate each candidate project. Instead, the team should be competent in selecting the adequate expert reviewers (similarly to journal editors who invite expert reviewers for individual papers submitted to the journal). Of course, the team can do the pre-selection of projects, determining which are worthy of sending for expert review, but for that, it’s usually useful to have at least some experience with research in one of the relevant domains, as well as with research proposals.