Even getting to that level would still be useful, but not without disadvantage. Archive searches are awful currently, and any level of improvement would be welcome. Though I'm inherently wary of something that is interpreting something for me.

So, if you asked "show me all sources that disagree with X claim" it will have issues because if Y source disagrees with that claim may be in contention, and it may not present it. That would still be useful, but because of that potential contention you would still need to check each source.

Such as helping read garbled texts at a rate that humans could never achieve, but I doubt the usefulness in the actual analysis of that information. It's a field with large gaps, and confident, but not certain conclusions. Dealing with that level of nuance and inherent gaps is where I see issues.

I'll admit I know next to nothing about genetics, but I'd assume from my understanding is that as a science it can reach definitively correct answers. History can achieve a high degree of confidence, but not certainty, and as we go back that level degrades. AI is already used in the field.

Again it's a research first field. If you're going into something seeking supports and contradictions than you're already off to a bad start as you are going to miss information that might be important, but isn't going to be picked up by you or an LLM

Yes it would be extremely valuable. The issue is that what constitutes as contradiction or support in the field is a matter of significant interpretation, a level of nuance that I do not believe LLMs could reach. It also has the issue of items outside of your potential query being missed.

The huge corpus of information is the primary spot of research. Going through it and making new arguments based on what you found, not what you were looking for is the point of the field. You can't rely on something using existing information to find new information.

AI certainly could summarize information as you've presented it, but it would be of little practical usage as you would have to go and check each individual element before using it. It also discounts that when searching for specific elements it can miss things that aren't directly specified.

Well, generally in the field you would make your thesis after analyzing the sources relevant to your general topic. But, again its a field where the same source might be used to support or criticize an argument depending on the interpretation of the author. Not to mention sources that contradict.

So, what exactly is being augmented? Having the conclusions based of primary sources does not give access to the external information that those authors used. You probably could get it to be oriented in a general direction, certainly that would be acceptable for non-academic uses. 1/x

We simply do not have access to every primary source, and a lot of higher level research involves finding material that is not readily available. So, even for textual sources you could never have a complete database. There is also the issue that physical sources like artefacts can't be used at all.

*counted as a casualty
Secondary sources could help, but not solve the issue, as they may be wrong, outdated, or not actually apply to the case. There would also be the issue of anything based on this would inherently not be new information. Additionally there are issues types and # of sources.

So, if you're just putting them all in a database to pull from you're not going to get useful information. As sources they have to be judged individually, and require information external to them. That information might be as simple as how something as a what counted as a to cultural norms.

Hi, one of those historian friends here. Primary sources aren't used by combining them and synthesizing some sort of "correct" narrative from sorting through their contradictions. They're often contradictory, inaccurate, are inherently biased towards the author, and many times non-textual.