My Latest AI Hack

Sorry for the lack of posts lately. I haven’t been particularly inspired recently as I’ve been very focused on thesis and whatnot. This one will be short, but I hope someone finds it useful. Enjoy!

I am always trying to find new, better ways to use AI. Like I’ve said previously, it is imperative to use AI make our use of effort more efficient, not to entirely replace our effort.

One of the greatest uses of AI is as a research assistant. I have, many times, found myself asking ChatGPT “find me sources on X.” It sometimes finds me something new, but usually what it turns out is not empirical or trustworthy. Oftentimes it will be a news source, some random no author report, or something unrelated to X entirely.

Using an online library search through your university or Google Scholar will 9 times out of 10 give you better material. My problem is that I have usually hit a point where I, seemingly, have found all there is. With every new query, I just get the same results over and over. Things I have already seen. I rarely find something new no matter how many “advanced searched” I do, adding on new categories. It never seemed to be enough. It is at this point I may turn over to AI to try and find a new source.

I always felt a little dirty or lazy doing this and knew to myself that what I really need is still out there on the library search. It occurred to me, instead, that I should try asking ChatGPT to create a library query. Below is my conversation with the AI. Parts of each ChatGPT response were truncated for clarity sakes.

Me: Create a comprehensive Boolean search query for an online library. The search is about disparate outcomes in police traffic stops.

ChatGPT: Here’s a comprehensive Boolean search query you can paste into most academic databases (EBSCOhost, ProQuest, JSTOR, etc.) to find studies on disparate outcomes in police traffic stops:

(”traffic stop*” OR “vehicle stop*” OR “driver stop*” OR “police stop*” OR “traffic enforcement” OR “roadside stop*”)
AND
(”disparit*” OR “inequalit*” OR “bias” OR “discriminat*” OR “profiling” OR “disproportionate” OR “differential treatment” OR “disparate outcome*” OR “racial outcome*” OR “inequitable outcome*”)
AND
(”law enforcement” OR “police” OR “policing” OR “officer*” OR “patrol” OR “traffic unit” OR “highway patrol”)
AND
(”outcome*” OR “decision*” OR “discretion” OR “enforcement” OR “sanction*” OR “citation*” OR “arrest*” OR “warning*” OR “stop result*”)
NOT
(”fiction” OR “film” OR “movie” OR “literature” OR “television”)

Me: Can you now create a new query for disparities in vehicle searches?

ChatGPT: Absolutely — here’s a comprehensive Boolean search query tailored to find research on disparities in vehicle searches by police across academic databases (EBSCOhost, ProQuest, JSTOR, etc.):

(”vehicle search*” OR “car search*” OR “automobile search*” OR “traffic search*” OR “roadside search*” OR “consent search*” OR “probable cause search*” OR “search and seizure” OR “search incident to arrest”)
AND
(”disparit*” OR “inequalit*” OR “bias” OR “discriminat*” OR “profiling” OR “disproportionate” OR “differential treatment” OR “disparate outcome*” OR “racial outcome*” OR “inequitable outcome*”)
AND
(”law enforcement” OR “police” OR “policing” OR “officer*” OR “patrol” OR “traffic unit” OR “highway patrol”)
AND
(”outcome*” OR “decision*” OR “discretion” OR “search decision*” OR “search result*” OR “arrest*” OR “citation*” OR “contraband” OR “hit rate*” OR “search yield”)
NOT
(”fiction” OR “film” OR “movie” OR “literature” OR “television”)

As you can see, ChatGPT would create very extensive, iterative responses with mutually inclusive and exclusive search terms. For instance, the use of the asterisks in “disparit*” or “inequalit*” that allow for different suffixes to be applied. In the case of the former, it may reiterate as disparity or disparities, for example. It also, as you’ll notice, includes exclusionary criteria at the bottom so as not to pull from illegitimate sources, even though I did not indicate this was intended for empirical research.

I found this to be extremely helpful in finding new sources. It also eliminates the AI bias whereby it won’t be able to give you sources on it’s own if they are behind a paywall or the like. In this case, as well, it still leaves the actual search for materials up to you. Although the AI is defining the logic for the Boolean terms, you can independently edit and enter it into any search engine.

Let me know what you think and if you give it a go. I’m sure I’m not the first person to do this, but I had not seen this anywhere else. There are some job candidate search engines that function like this from what I understand, but I think it has great potential for use in research and learning. Thanks!