Basil Bekdash

Thanks for the systems perspective Stefan. Modelling the organisation(s) is one of those foundational pieces of work that would not only support technical/digital evolution including "AI" deployment but also many other aspects of healthcare operational management and practice. From a clinical perspective I remain highly sceptical of the readiness of the system, professionals and society for generative AI such as LLMs and I am also concerned about the deliberate design of a lot of GenAI that explicitly mimics human outputs without explicit disclosure of the source. I would make a key distinction of the use of LLMs and Gen AI between "hot" and "cold" usage by which I mean use in live workflows (I will stick to clinical as that is my main domain but equally might apply to others) versus in preparation or analysis of material where there is no time pressure of the sort that is almost the norm in clinical practice. The reliance on human checking will fail and as you allude to the existing ways of working are not being adapted (and in practice may not be capable of adapting) to the insertion of LLM based technology as currently design/implemented. It is mainly the failure mode that is the issue even if effectiveness was not in question. Detecting errors and therefore preventing harms is demonstrably harder in a human like output which when produced at greater volume, as LLMs tend to, and under time or intensity constraint is more likely to occur. To go back to your point, there are many foundational pieces of work, architecture, systems change and more that should be in place before introducing a significant disruptor into a complex often chaotic and high risk environment.

Hi Benedict, Would be interested in any thoughts you had on this very adjacent topic.....https://www.pslhub.org/forums/topic/2396-reimagining-health-and-care-data-systems/

Over the past few years I've pondering what is wrong with most of the data systems I interact with. To be honest I've been pondering this for most of my career as I was frequently sent to the dungeon (aka medical records) to review, sometimes also find on the stacks when that was allowed and synthesise it into something. That evolved to doing the same digitally and that is what most data analysts and scientists spend most of their time doing. Granted they generally use more sophisticated techniques then I ever did. That is probably the first concern.....data capabilities have become incredibly good but in big data before it and AI more contemporaneously the problem is as all good data/computing people know garbage in garbage out (for some reason I prefer it to the more British, and therefore correct, rubbish!). Amalgamating data sets, federation (good and bad forms of this....no I don't mean the FDP specifically), SDEs, data linkage etc all are held up with the promise of moving us to the truth. But the real truth is that improving data quality is a wicked problem. So what am I getting at? Well I am interested in how people feel about data management, what they might think about how to shift the entire ecosystem if in fact that is needed to tackle said wicked problem and what the critical non-technical considerations need to be. If there is an appetite I will elaborate and we'll see if we can get some conversation going.

Last week this was published and I am interested in what the broader community thinks of it. I already have a feel what @Clive Flashman thinks but what about the rest of you? https://www.england.nhs.uk/long-read/medical-device-regulation-for-ambient-voice-technology-products/ It certainly felt to me that the initial reaction was a widening of the gap between the concerned and the enthusiastic rather than the fabled and politically much loved "coming together" So.....help or hindrance and what are thoughts with respect to the timing in relation to NHS policy versus the pending outcome of https://www.gov.uk/government/groups/national-commission-into-the-regulation-of-ai-in-healthcare?

@Tauqirashraf Based on scope and longevity....probably the diagnostic service digital platforms. Laboratory and imaging systems (PACS) have given all clinicians (with appropriate rights) access to far more diagnostic information including the primary data, not just reports. They have been around for decades and unlike primary care systems which are even older are a fairly generic and consistent service across settings. All digital systems introduce (often new) risks but hopefully are at worst risk neutral...impossible to truly quantity but there are incontrovertible benefits to digitisation in terms of access, efficiency and secondary uses even if they are far from perfect even for those functions! As or the last 2 questions.....there may well be strategies that answer your question but the fundamental deficit is in the foundations of digital systems which are not truly fit for purposes and require workaround upon workaround to deliver the reasonable functions of today and proposed functions of tomorrow. The future is always the day after and isn't on the agenda if that isn't too cryptic!