Heuristic Search Episodes

When exploring the concepts of clinical decision making, we recently coined a term that we think will be useful: Heuristic Search Episodes (HSE).

What is a Heuristic Search Episode?

These are typically short and sweet searches that will satisfice the problem at hand.

They are quick and dirty. But a bit more in-depth than a simple, single Google search.

They are nothing like a Systematic Review, not even as full as a Scoping Review.

The vast majority of searches in our lives do not involve note-taking: as soon as we have a good enough answer, we stop. If we don’t remember it in future, we will probably simply repeat the search.

HSEs are more likely to be used for that middle ground of problem: sufficiently complex answers that it is worth making a note, and also likely enough to be repeated that there is value in such note-taking. Indeed, with many repeats, we are then more likely to branch into note-making. [[Note-Taking vs. Note-Making]]

Clinical HSEs

Let’s explore a recent interesting clinical example with POCUS: point-of-care ultrasound — notes were made in Zotero and PBrain. Using Elicit.org and Google Scholar did quite well but some other AI-based tools (scite.ai and Semantic Scholar) did not.

This allowed me to collate and condense some information which refuted my previous stance that POCUS was not really evidence-based and had little to support its integration into regular healthcare provision. This was useful and timely for an upcoming panel discussion on the topic.

Ad hoc episodes

Another HSE type is when you come across something serendipitously (See [[Serendipitous Browsing]]) and want to make a note about it somewhere because “it might be useful later”. This is part of my Personal Knowledge Management (PKM) process or toolset or approach.

I use TheBrain (formerly known as PersonalBrain) a lot for that – so much so that I have created a specific workflow to facilitate it. I send the link to an email alias called PBrain (which comes up easily because “PB” are not commonly associated letters). The actual address is d.avid.topps AT gmail DOT com which takes advantage of the trick in GMail that you can embed extra periods without the destination being affected. There is then a Filter in my GMail that places these messages into their own folder. Then when I have time, I can browse this folder and place those messages into the PBrain app.

The advantages of using the PBrain app for this are:
1. Easy capture
2. Linking to related Thoughts (which often comes in useful later)
3. Excellent Search
4. Annotating with comments on why this might be interesting.

In the [[Serendipitous Browsing]] note, I comment about capturing direct to Notion. But in my current workflow, it is better to keep using the PBrain app, partly for the above reasons, but I can now also easily cross-link to the rest of my PKM suite. I can also place the note into an extended Context of related thoughts — this makes later discovery more interesting when I come back to Thoughts I had long forgotten about.

This latter serendipitous aspect happens often enough that I sometimes simply go wandering in PBrain, just for fun.

More GPT-based search

On 19mar2023, I came across Aaron Tay’s blog post where he is now quite keen on Perplexity and BingGPT — more so than Elicit.org, which I found surprising.


Looking at my searches in Zotero, I ran the same search about POCUS at [Perplexity](https://www.perplexity.ai/) and was similarly impressed. See https://www.perplexity.ai/?s=u&uuid=4ac47d1c-216a-425a-a4de-cb526a025d1b — this URL does make the search reloadable, which is convenient. But this may not be publicly accessible.

This was quicker and easier than doing the 3 rounds of Elicit search, plus using the Starred selector method. I think that I might have been more skeptical if I had not tried Elicit first.

In Aaron’s post, he tells you how to make the search more restricted than just general internet but he seems quite impressed as to how much better GPT is when combined with a search engine, rather than just using it by itself. This is now termed Retriever Augmented Generative (RAG) search.

I also get the impression that this whole area is going to change quite fast. Worth keeping an eye on.

Annoyingly, Perplexity uses Twitter for sharing. And Discord for the forum – sigh. I find that Discord chatter is quite general, not academic.

I don’t see a way to load the references into a reference manager as a batch. The research grid that is generated by Elicit.org is better in this respect.

On 14dec2023, when I look another look, there is now a library and Collections where your previous searches are stored: https://www.perplexity.ai/library — but the library does not have all my recent searches – I may have used a different login for those.

There is also now a Pro version – not cheap – and not essential for the kind of searches that I am currently doing but might become useful. e.g. https://blog.perplexity.ai/getting-started#copilot

Bing GPT

Is now enabled in the right sidebar when you use Edge as your browser.

Gave a similar answer to Perplexity but with simpler interface.

Also makes it harder to retrieve the source materials. You have to trust it more.

Another clinical HSE on antidepressants and long QT – nicely summarized and reasonably accurate – but I can only say that because I already did the search. In itself, there is not enough information to provide confidence that the answer is correct. Similar to using ChatGPT – see [zotero://select/library/items/MHXBL2CK]

More insulated results

After using both these tools, I found I was somewhat dissatisfied with BingGPT. Not because the answers were poor or wrong. I just couldn’t really tell. When using Elicit.org, with the 3-layer process, I felt somehow more connected to the process and sources, which gave me more trust as a knowledge worker. Ongoing testing with Perplexity has been more encouraging — it seems to be improving with time and continuing upgrades.

For Joe Public, he will not want the extra hassles, and likely may not be able to tell the difference anyway. I might find myself using Perplexity more where I want the quickest/dirtiest answer