Monday, June 24, 2024

UDLCO: AI, AGI, Spearman's g and the g physician

UDLCO summary: A chance discussion on General physical examination roles over to Spearman's g and AGI followed by terminology and their ontology management 




UDLCO transcripts:





[23/06, 05:27] ATP: what is EOM full?

[23/06, 05:36] SgI: Eye Opening?

[23/06, 05:38] AgI: Extraocular Movement

Caveat: Most like interpretation given the context

[23/06, 05:59] SgI: Matches somewhat with Eye opening -relevant for Neuro cases 
mention of Nystagmus separately raises an issue to this expanded terminology

[23/06, 06:03] AgI: My answer was GPT based Sir, hence the caveat was there! 

Shouldn't there be studies on analytics leading to reduced reliability? The contra indications of AI, and other emerging tech

Big tech will not generally be willing to sponsor such research

[23/06, 06:09] SgI: I still worry 
All this talk about LLMs and inclusion of AI in day to day apps falls flat when every EHR type converts to HER while my _teh_ never corrects to _the_

[23/06, 06:17]AgI: It's the classic elephant and the blind men... 

Folks are fitting the elephants legs where pillars should be, replacing ropes with its tail, and in medicine I would say it's trunk is being mistaken for the snake on the Rod of Asclepius. 

And, what is more, to build a larger circus tent, they are fattening up this mammoth. LLM larger and larger mammoth! 

It will be quite a show when this mammoth runs with all the tent attached on top!

[23/06, 06:19] ATP: oops...I was thinking ROM - range of movement with spelling mistake

[23/06, 06:20] ATP: only ocular movement is possible. how can extraocular move

[23/06, 06:21] Abhishek Aye AI: Movement of the extraocular muscles

EOM stands for both
Extraocular muscles
Extraocular (muscle) movement

[23/06, 06:22] AgI: Essentially ROM of the eye

[23/06, 06:34] ATP: btw if a medstudent presents a case like this to me: it is a F. to repeat semester and reappear in 6 months. See the quality of HPE and data capture, leaves so many key things out. OHCM and WashU have such a nice template for HPE. Even 5% of that is not visible here.

[23/06, 06:36] ATP: OHCM = Oxford handbook of Clinical medicine
WashU = Washington University manual

[23/06, 08:00] ALLM: Agree with GPT.

It is extraocular movement and the template is a focused one by someone who possibly limits his her practice to vestibular disorders, a subspecialty of otorhinolaryngology @⁨ATP⁩ @⁨SgI

[23/06, 08:56] ATP: HPE should be common. how can it be so fragmented and disjointed?

[23/06, 08:57] ATP: EOM full. does it make any sense?

[23/06, 09:06] ALLM: Full range of extraocular muscle movement


[23/06, 09:29] AgI: Per SCT

EOM is for the muscles group

They use Extraocular *Eye* Movement, however no mnemonic EEM is specified

Ref screenshots below

[23/06, 09:31] ALLM: πŸ‘†This shows how the data presented by the vestibular specialist can be snomed coded πŸ‘

[23/06, 09:32] ATP: I just need a simple HPE

[23/06, 09:44] ALLM: Yes

However it's likely to be far from simple

Specialization can reduce human complexity to make things focused and simpler

However those who choose to tackle the entire human in all it's complexity require a lot of Spearman's g loading that makes them the ideal g p

Looking forward to @⁨AgI⁩ 's sharing more about Spearman's g loading which I chanced upon recently after following his lead on RPM (aka Ravens progressive matrices). I realised my workflow in medical cognition (aka integrating medical education and practice) till now had been incomplete without it.

[23/06, 09:55] AgI: I wouldn't dare that in_doctri_nation πŸ™πŸ™‚


Sorry about the Wikipedia link. I would need to go back to my clin psy notes, where I last met Spearman. 

A more pertinent "clinical" metric is the *lie score*. Traditionally associated with probability of integrity of the subject / patient. 

Modern clinical interpretation leans towards the possible target metric of lie score being "law abiding". Discipline internalization leads to higher lie scores. 


Again sorry about the wikipedia link. The page reads ok... As of this message

[23/06, 10:16] ALLM: Yes and quoting from your wiki link πŸ‘‡

"The g factor (also known as general intelligence, general mental ability or general intelligence factor) is a construct developed in psychometric investigations of cognitive abilities and human intelligence"

Do you think AGI would have already borrowed and published some insights from Spearman's?


[23/06, 10:21] AgI: I assume you refer to AGI oriented research literature. 

Biomimetic methods do build upon psychological theories. Such research is more likely in systems psychology and systems neuroscience. 

Most of us into hands on synthetic intelligence research may consult psychology for insights, but neuronal circuits guide more of our biomimetic approaches. 

Synthesis fails easily, analysis never fails. It may only be proven wrong on a later date.

[23/06, 10:22] ALLM: Wow! On the second last line! Looks like a potential slogan


[23/06, 10:23] ALLM: It's like Brahma fails, Shiva doesn't!

[23/06, 10:32] Snollm: Unless information is pooled and analyzed, knowledge is lost. We might just as well work only with paper. Write it, forget it, trash it. VoilΓ . Easy-peasy. And when others make vital knowledge discoveries, pay through the nose to get to even know about it. Penny wise, pound foolish. Egad!

[23/06, 10:43] +91: First get information...

[23/06, 10:45] ALLM: Data capture systems is the first step

Unfortunately mired with Orwellian fears


[23/06, 10:47] +91 : So solve that first than lobbying govt to buy SNOMED 15 years back...

[23/06, 10:58] ALLM: Did some searching myself!

As suspected someone has been there done that in 2018!

Their article below is just a feeler though and I find even the Wiki page on "g factor" has quoted generously from it's intro while I quote below from it's conclusion πŸ‘‡

"Can the g factor play a role in AGI research, 
then? 

In light of our discussion, the answer can be either positive or negative depending on the very aim of AGI. 

A weak or instrumental notion of g, like the psychometric g, can play a role in 
AGI projects characterised by an emulative approach, where the goal is reproducing a human level intelligence regardless of details about its 
neurocognitive or biological architecture.

 Here, the psychometric g, as assessed by IQ tests, might help to evaluate the intelligent behaviour of artificial systems besides other behavioural 
tests—e.g., Turing and Nilsson’s tests. 

By contrast, a strong, neurocognitive notion of g is involved in the discussion about the composition of human intelligent systems, the causal 
interactions among parts, and how to artificially reproduce these aspects. In this respect, the possible role of the g factor in AGI research depends on empirical data in neuroscience. As we 
have argued in this paper, this role of g in AI is dubious."


[23/06, 11:01] AgI: Conclusions are as above! 

Rather than psy theories, I personally find more tangible material for synthesis in the nearby bhoot bangla!

[23/06, 11:03] Snollm: Eh? What? It’s getting recorded and then callously discarded. Think of all the files and prescriptions and discharge summaries and investigation results that get generated in one day in all of India. And then compute the colossal waste, knowledge-wise.

[23/06, 11:04] Snollm: Let’s be honest and admit that even a simple and decent system isn’t available for the doctors to work with.

[23/06, 11:07] ALLM: Yes the stumbling block is a collective fear of an Orwellian future exploitation although what we may be actually witnessing is more and more individual anguish on why no one's evaluating what has been already captured!

"Orwell never realized we would be buying the cameras ourselves and our greatest fear would be that no one was watching." RT

More here:

[23/06, 11:07] Snollm: Classic chicken and egg. Without semantic interoperability there’s no usable information (not just data, mind) capture. Without controlled medical vocabularies (read SNOMED), there’s no semantic interoperability.

[23/06, 11:11] SS: Since SNOMED CT (unlike ICD) cannot work without an EHR system, becoming a country member of SCT before ensuring nationwide EHR/EMR use may not have been a judicious move by any of us.

[23/06, 11:13] SS: India is a country member of SCT since March 2014 but as of June 2024 we hardly use any EHR/EMR.

23/06, 11:16] AgI: Typically Pill AI refers to patient compliance solutions


Caveat: The term may inadvertently be misconstrued in our social context.

[23/06, 11:18] Snollm: Actually, this is not true. Lots of organizations, both public and private do use it, although probably not harnessing its full power. NRCeS has a rather long list of such organizations.

[23/06, 11:19] SS: What is the % of use in a country of almost 1.5 Bn population?

[23/06, 11:21] SS: Most importantly, how many medical colleges in India (out of 706) do actually use any EHR system for any meaningful purpose?

[23/06, 11:23] BEMR: Prof, also how many colleges, grad students have any intro to Health Informatics? Without a  culture of data usage, how do we even think that EHRs would be of use to Drs.

[23/06, 11:25] Snollm: Working SNOMED is challenging. One has to understand it well enough to be able to design systems that can utilize it effectively. The current levels of understanding is still way low amongst most. Probably the SNOMED-BAHMNI collaboration will have some significant impact since all that the implementers will then need to do is to use the solution out-of-box. The BAHMNI solution is not the be-all-and-end-all one, but definitely a good starting point.

[23/06, 11:28] Snollm: To use an EMR one doesn’t need any informatics knowledge. Just being able to use it is enough.

[23/06, 11:31] BEMR: Thats why I said "intro" ! and you would be surprised .. btw, we setup an instance for MUHS .. and are getting some positive feedbacks

[23/06, 11:40] Snollm: Even any introduction isn’t necessary. Most of the current generation of medical students are pretty computer savvy. Sadly, almost all digital health information solutions are appallingly designed. Not intuitive. Terrible UI/Ux.


[23/06, 11:41] ALLM: People moved away from compliance to adherence and concordance decades ago!



[23/06, 11:42] BEMR: I am wondering if in that case courses for Health informatics is useless if being computer savvy would cut it.

[23/06, 11:42] BEMR: anyway, I am not from medical background - so what would I know! I am only reflecting on what I been hearing for so long ..

[23/06, 11:43] ALLM: And spend most of their time memorising MCQs from their mobile computer screens

[23/06, 11:43] AgI: Definitely new terms for me. How prevalent in quality of care reviews?

[23/06, 11:47] ALLM: I must admit i too was amazed by these terms while editing this piece in 2011πŸ‘‡


[23/06, 11:47] Snollm: Health informatics is required to make use of the information captured and available for reuse and analysis. To use a system one needs to know how to use it. No more, no less.

[23/06, 11:48] AgI: Concordance, then, is most affected by ease of information access (for patients)?

[23/06, 11:48] BEMR: Sir, with due respect, Health Informatics is not limited to data capture, system of record, or EHRs.

[23/06, 11:50] kk : Ji I think when I know how the captured data is used meaningfully then I may capture complete data by using the tool

[23/06, 11:52] ALLM: You got it

[23/06, 11:55] Snollm: Knowledge of health informatics isn’t a basic requirement for using a system. A well-designed system that uses SNOMED (let’s say for argument purposes) will make data capture, query-retrieval and even analysis, workflow automation, decision support all happening in the background. Users of properly designed systems with “SNOMED inside” wouldn’t even be aware that any code system is being used at all.

[23/06, 11:56] ALLM: Here's more about how unlimited it can be (shared previously here ad nauseum)πŸ‘‡


[23/06, 11:58] ALLM: Yes I think she meant knowledge of health not health informatics is important to make meaningful use of captured data

[23/06, 12:05] Snoll: I’m assuming that the person who’s entering the information is having sufficient knowledge and understanding of the importance of the observations and has the ability to make the necessary inferences. That’s axiomatic.

[23/06, 13:51] Snollm: I don’t know how to explain it any simpler. Sincerest apologies

[23/06, 14:00] Snollm: Thing —> Transport (—> Cart —> Handcart, Bullock cart, Horse-driven), (Wheelers —> Bicycle, Tricycle, Unicycle, Scooter, Motorcycle, Four Wheelers), (Motorized —> Car, Aircraft, Helicopter), (Can fly —> Aircraft, Helicopter)

[23/06, 14:00] Snollm: Now you can see how SNOMED is arranged in a poly hierarchy. πŸ™‚

[23/06, 14:16] AgI: Modern medicine and AI appear to be in a self fulfilling prophecy loop! 

So what is nature vs synthetic? 

Modern medicine vs traditional medicine
Or
AI based practice vs "traditional" modern medicine practice? 

Maybe, Nature has an anecdote in how a migratory song bird matches or betters predictions for the hurricane season! 


From the paper, 
β„šπ•¦π• π•₯𝕖
πšƒπš‘πš’πšœ πšπš’πšœπšŒπš˜πšŸπšŽπš›πš’ πš™πš˜πšπšŽπš—πšπš’πšŠπš•πš•πš’ πš‘πšŠπšœ πšπšŠπš›-πš›πšŽπšŠπšŒπš‘πš’πš—πš πš˜πš›πš—πš’πšπš‘πš˜πš•πš˜πšπš’πšŒπšŠπš•, πš–πšŽπšπšŽπš˜πš›πš˜πš•πš˜πšπš’πšŒπšŠπš•, πšŠπš—πš πšœπš˜πšŒπš’πšŠπš• πš’πš–πš™πš•πš’πšŒπšŠπšπš’πš˜πš—πšœ πšŠπš—πš πšœπš‘πš˜πš πšœ πšπš‘πšŠπš πšπš›πš˜πš™πš’πšŒπšŠπš• πšœπšπš˜πš›πš–πšœ πšœπš’πšπš—πš’πšπš’πšŒπšŠπš—πšπš•πš’ πšŒπš˜πš—πšœπšπš›πšŠπš’πš— πš‹πš›πšŽπšŽπšπš’πš—πš πšœπšŽπšŠπšœπš˜πš— πš‹πšŽπš‘πšŠπšŸπš’πš˜πš› πš™πš›πš˜πšŸπš’πšπš’πš—πš πš›πšŽπš—πšŽπš πšŽπš πšŽπšŸπš’πšπšŽπš—πšŒπšŽ πšπš‘πšŠπš πš‘πšžπš›πš›πš’πšŒπšŠπš—πšŽ πšŠπšŒπšπš’πšŸπš’πšπš’ πš’πšœ 𝚊 πš™πš›πš’πš–πšŠπš›πš’ πšπšŠπšŒπšπš˜πš› πš›πšŽπšπšžπš•πšŠπšπš’πš—πš π™½πšŽπšŠπš›πšŒπšπš’πšŒ-π™½πšŽπš˜πšπš›πš˜πš™πš’πšŒπšŠπš• πš–πš’πšπš›πšŠπšπš˜πš›πš’ πšœπš˜πš—πšπš‹πš’πš›πš πš™πš˜πš™πšžπš•πšŠπšπš’πš˜πš—πšœ.
π•Œπ•Ÿπ•’π•¦π• π•₯𝕖


[23/06, 14:51] sno 2: This is a typical ontology to represent entities and relationships. Adding properties/artributes to the relationship ( the arrows/links) makes it even more interesting to derive knowledge thru graph algorithms . This allows one to explore these graphs - typically used for clinical research, networks etc . These are techniques to build knowledge systems- typically solved using computer Languages such Prolog and Lisp. Today these concepts are used in graph Databases and Knowledge graphs
.
[23/06, 15:29] sno: Knowledge graph - example Chinese medicine representation.

[23/06, 16:31] gcdac: SCT is a poly-hierarchical ontology of unique coded clinical terms. 

Drug-Drug interaction is dynamic chemistry knowledge having clinical significance. It is also an evolving domain of clinical research and safety.

When we designed the Drug Model in SCT, and studied such models in other countries, having DDI information or relationships in it appeared as overloading the purpose of the clinical ontology. 

Some even wanted to include regulatory information (banned, schedules, etc). Some wanted packaging and price to be included in brand concept. There are arguments for & against these. We chose to keep SCT Drug Model simple and basic so that such systems can be built using it.

One can make a DDI database with SCT Drug Extension, just as one can make an inventory and billing system using it. One can use SCT DE with DDI to make ePreacription with alert/warning.

btw, NRCeS provides a Drug Information Service Bundle (DISB) as a separate package where some additional information are provided. 

NHA aims to use both of them and some more info to build the ABDM Drug Registry.

It is a wip.

[23/06, 16:34] gcdac: So, if you'd like to build an authoritative DDI database for India, you'll most probably find SCT with CDCI best candidate to start your work with. πŸ˜€

[23/06, 16:37] AgI: Or, take an existing DDI db and re-encode in SCT?

[23/06, 16:43] AgI: Also, what are the license terms for this work? 

If free for India, as in other NRCES deliverables, will it be feasible to host it open access and contributions? Contributors agreement can enforce any necessary IPR clauses. 

Some quick turn-arounds I envision are 
1. translating to other Indian languages not already supported (and no one is translating drug names, maybe only translitetation). 
2. Open community review, as well as early review

The standard version will always be the ones released by NRCES

[23/06, 16:58] Snollm: Now you’re beginning to get SNOMED 😁

No comments: