I have been having a back-and-forth argument with colleagues about the value of AI in MS research. I therefore asked Google Gemini Pro 2.5 Deep Research the following question:
Thanks so much for this very important and relevant post. I have been a believer in the "inside-out" model of MS for a long time, but there do not seem to be many in the MS community who believe this model. I personally believe that what I refer to as the "conventional MS model" (the "outside-in"/MS as a primary autoimmune disease) was created and exists to benefit the "medical-pharmaceutical-industrial" complex in that it allows and supports the pushing of the immunosuppressive and/or immunomodulatory MS "DMTs" and places the most emphasis on relapses and new conventional MRI activity, such as Gd+-enhancing lesions. I have always disagreed with the USA FDA's labeling indications for the DMTs that state "for the treatment of relapsing forms of MS" (*or in the case of Ocrevus, PPMS), because as I see it, the "DMTs" do NOT "treat MS"-rather, they reduce some manifestations of active inflammation: relapses and new conventional MRI activity. If MS were to be seen, classified, and treated as a primary neurodegenerative disease, at this time, I do not believe there would be any "treatments"-it would be a similar situation to ALS, frontotemporal dementia, Alzheimer's, etc, where the focus would be on any DMTs that might be able to "slow progression," which has not been the primary endpoint of MS DMT trials, except for some trials specifically for progressive forms of MS. It would be MUCH harder and more expensive and time-consuming for the pharma companies to try to come up with "treatments" for MS as a primary neurodegenerative disease vs a primary autoimmune disease.
I would love to hear your thoughts about this and if you feel we will see a shift in the MS community (esp among clinicians, researchers, regulatory bodies, pharma companies, etc) towards the "outside-in" model and away from the "inside-out" model.
I think we will see a shift, but it will be a painful transition. MRI (despite all its virtues), is a half-blind double-edged sword, leading us to neglect the things we cannot see. I fear some newer therapies for smouldering MS may fall on that sword.
I completely agree with your statements here. We know from the research at this point that there is SO MUCH MS disease activity and pathology that cannot be seen on conventional MRI, yet conventional MRI continues to be a major focus and primary endpoint of the MS DMT clinical trials as well as patient monitoring and assessment.
When you stated that "I fear some newer therapies for smouldering MS may fall on that sword," I am wondering if you might be referring to the GEMINI trial of tolebrutinib for RRMS that was deemed a "failure" since tolebrutinib did not outperform Aubagio in terms of ARR, yet we know from research now that relapses are not associated with long-term progression and CDW in MS (the UCSF EPIC study results evidenced this, as one example).
I think one aspect of the transition will also be the role that the revised 2024 McD diagnostic criteria for MS plays during this time in terms of its purported view of the "biological basis" of MS and moving away from the current phenotypes of MS (RIS, CIS, RRMS, SPMS, PPMS).
You might be interested to see this website that Sanofi has created in advance of what seems to be the FDA's expected approval of tolebrutinib here in the USA later this month for non-relapsing SPMS: https://www.rediscoverms.com/en-us/. Sanofi putting this information out is a "double-edged sword" as well, I feel.
I suspect Charcot, c.130 years dead, would be turning in his grave.
Recent thinking has suggested that MS is one disease. Yet this ‘one disease’ has yet to be cracked despite unbelievable advances in imaging, computer power…. What have 40 ECTRIMS annual meetings really achieved? Did the recipients of the Barancik and John Dystel prizes really deserve them given the most basic questions about this one disease have yet to be answered?
MSers wanted one thing from the 4-5 decades of well funded research - to stop the underlying neuro-degeneration which leads to unrelenting disability progression. Some even hoped for a degree of neuro-restoration. Yet the drugs (the MS drugs market is projected to grow to c.$41 billion by 2032)to date have only tackled relapses not the neurodegeneration. Relapses were the low hanging fruit.
The MS Society websites claim that ‘there’s never been a better time to get MS’ and ‘MS research has never been so exciting’. Yet, in 2025, MSers are still moving from crutches to manual wheelchairs to electric wheelchairs. My GP visits two women with MS in their mid-30s who are in care homes. The use of AI in MS research will make little difference to them.
MS is often portrayed as the poster child of neurological diseases. Not difficult given the dire progress relating to MND, Parkinson’s and Huntington’s. But with so many basic questions yet to be answered, the MS research field needs a shake up and a kick up the backside. Too many broken promises (remember NMSS Promise 2010?), too many progressive MS trial failures (MS Smart, MS Stat)…… I hope that AI doesn’t just generate another thousand+ rabbit holes to keep Professors of neuro-immunology, ever expanding research teams, and an army of PhD students busy for the next 25-30 years.
Not sure if this is correct. Once we learned how to do progressive trials, we have had some success, i.e. ocrelizumab, siponimod, tolebrutinib, ...
I will be presenting the O'Hand study at this year's ECTRIMS, which is based on insights from our PROMISE 2010 work. It may not all be bad news or no news at all. ;-)
I strongly favour the inside-out view. Otherwise it is difficult to explain evident abnormalities brewing in pre-lesional “normal appearing” white matter (NAWM), that are measurable up-to 2 years before a Gad enhancing (or new T2) lesion appears.
I worry there is a false dichotomy here, between inflammation vs. neurodegeneration. Plenty of what is coming from inside-out (and secondarily propagating Gad lesions) could still be inflammatory. Smouldering inflammation (and not just microglia) compartmentalised within the CNS could rationally be a major driving factor for the mitochondrial dysfunction and eventual axonal degeneration seen in MS. Indeed, cortical lesions (that often co-localise with meningeal B-cell follicles) independently correlate with metrics of axon loss in the NAWM.
Neurodegeneration risks becoming an umbrella term, that incorrectly swallows up all the smouldering inflammation which: 1) we cannot see on conventional MRI and 2) fails to respond adequately to the current crop of anti-inflammatory therapies.
The risk of inaccurate terminology is that we may prematurely divert attention to neuroprotection (which I agree is important), before fully addressing the unmet need of smouldering inflammation (compartmentalised within the blood brain barrier).
If your house was on fire, you’d probably wait for the fire brigade to fully put all the flames out, before you start to redecorate and install new furniture.
You definitely still have a role in the brave new world of AI! But not all of your peers will. The difference is how you centre patients. Whether NHS, academic or AI generated information and data, you’re making knowledge accessible to support informed choice, respectfulness and dignity - and through that helping patients feel less isolated and hopeless. That’s irreplaceable. Those who provide a poor service are on borrowed time. AI will change the horizon on patient experience and options. People will always prefer people - but where human medics let them down, AI-assisted healthcare will increasingly offer an alternative. It won’t offer the quality that you do but how many of your peers can either.
Thank you for the opportunity to comment on the back and forth we have been having with others about the potential role of AI, and this example. My, and others', comments in the to- and fro- were that the level of thinking and interpretation required were mainly the domain of the human brain, and that i was unclear if AI could provide this; although AI could facilitate this. You commented that you expected AI to achieve the levels of thinking and interpretation.
My thoughts are:
This is very nice summary to read. Is it accurate? To be absolutely honest, that's hard for me to evaluate as some of this basic science is out-with of my expertise and that would be the role of those people with the expertise to debate. but let's assume it is.
Additional important issues for me are:
This, i presume, is a synthesis of available literature. I see no QC. To what extent are the various components substantiated and accurately interpreted, rather than the interpretations of others?
Whilst the discrepancies for the inside-out and outside-in hypotheses are mentioned, what alternative hypotheses are generated from them by AI, or support for that, rather than to say that, at the end day, MS may be a combination of the inside-out and outside-in hypothesis. Feels like a bit of a cop out.
There is mention that MS doesn't meet criteria as a canonical autoimmune disease. Given concerns raised over the years about the suitability of the MS diagnostic criteria, and their developments, how suitable are these canonical autoimmune disease criteria. Again, above my level of expertise but others should comment, critique and challenge.
AI can highlight discrepancies, but to what extent can AI help us understand the nuances meaning of discrepancies from theory and expectation.
Can AI help us determine exactly what work is needed to clarify the specific work required to progress out understanding. For example, in the section titled "Implications for research and therapeutics" my understanding is that AI has trotted out the list of suggestions of others. Nothing new. Perhpas my expectations are unreal
So, as of yet, whilst i can see a very clear role of AI for summarizing and synthesizing available information, whether it be written or numeric, it is the critique and interpretation of that information, and the thinking about markedly alterative ideas, that is required for scientific development. Here the human brains of unique individuals become irreplaceable.
Given my last viewpoint, is AI, when set in a Kuhnian frame of reference, a winning combination for speeding up scientific developments?
To understand what i mean by this question, and why i think it is relevant, I need to explain a bit about the work of Thomas Kuhn.
Thomas Kuhn wrote a challenging-to-read book, The Structure of Scientific Revolutions (1962), and subsequently wrote and thought extensively and deeply about the issues contained therein. In his book, Kuhn proposed that major developments in science come from paradigm shifts in thinking; where a paradigm shift represents a very substantial, often radical, change in opinions or theories, rather than just a small improvement in understandings. Kuhn argued that paradigm shifts are a natural part of the cycle of scientific development.
Kuhn's cycle, in simple terms is this: we have an idea about something scientific - an hypothesis - we gain information over time that supports and refutes the hypothesis. Over time, evidence against our original hypothesis mounts and new hypotheses are proposed. Tensions rise and tempers fray in the scientific field between proponents of the new and original hypotheses. The new hypotheses are typically ignored, dismissed and/or ridiculed by developers and supporters of the original hypotheses until finally a new hypothesis is accepted because the evidence for it is overwhelmingly greater than for the original hypothesis. Then, the original hypothesis generators said they knew this all along and it was their idea in the first place. And the cycle continues.
Kuhn's work was very controversial when it landed. Not because of the idea of a cycle of development with paradigm shifts, as he gave multiple exemplars from physics, but because of his ideas of some of the related issues. There were three issues proposed by Kuhn that scientists found particularly jarring:
First, Kuhn said that most scientific research adds nothing to the field in terms of development. This, Kuhn said, was because most research was not aimed at solving the difficult problems in a field. Unsurprisingly, as Kuhn suggested that this comment applied to most research in a field, it was not well received (although many quietly agreed in terms of their own work!).
Second, Kuhn suggested that people responsible for paradigm shifts were from out-with of the mainstream scientists of the field of interest. This was not well received for obvious reasons.
Third, and the most jarring, was that Kuhn argued that a major cause of hindrance to the scientific development of a field was the scientists working within that field. He cited the propagation of old ideas, doing the same thing over and over again, resistance to new ideas and new personnel, lack of inclusivity, and failure to undertake problem-solving coordinated targeted work. In saying this Kuhn had released a very large lead balloon.
Whilst Kuhn's work was controversial there are things the MS field might reflect on, and that AI could help facilitate; hence my question. There has been a gradual build up over time of evidence against the original ideas of the biology of MS (outside-in hypothesis). A new hypothesis, the inside-out, has been proposed and increasing, but incomplete, evidence for it provided. This sounds uncannily like a Kuhnian cycle moving toward a paradigm shift. But is the inside out hypothesis the paradigm shift required or is their another alternative? Can AI help facilitate our speed round the natural Kuhnian cycles we are in? What coordinated, targeted, problem-solving research in MS is required? What work out-with of the field of MS can help us tackle these issues. Can AI help the MS field maximize inclusivity to fast-track our work and ultimately improve the opportunities for people with MS.
Finally, but importantly and very relevant, Kuhn also wrote on the importance of measurement in science; The function of measurement in modern physical science (1961). He highlighted not just the importance of measurement - i.e., its quality not just its use - but that the role was to highlight anomalies (discrepancies from expectations). This aligns with his cycle, where the testing of hypotheses identifies the anomalies that drive reconsideration of our theories, and modification of existing hypotheses and/or the generation of new hypothesis. If we took note of this we would pay much more attention to our measurements in MS research and the anomilies they highlight.
So, in short, AI to help facilitate progress. But the hard yards of thinking and interpretation require the human brain. Therefore, to maximise our ability to solve the main problems in MS we need AI, our clinicians and researchers to think more, think harder, think deeper, i believe, invoke Kuhn's work better our measurements.
As I have responded well by treating my MS as EBV driven for twenty years, and taken an antiviral as a consequence, I'm sure it is an 'inside out' pathology.
What is missing, in majority of articles, is a description of how that infection actually changes things.
My thinking is the damage comes from purinergic receptors being inappropriately opened, allowing an influx of calcium that causes cell death. That destruction is what the immune system reacts to, creating the impression of 'outside in' activity.
Let’s be honest it’s going to affect everyone/everything in the coming years , there’s no turning back so might as well embrace it 🤷♂️, it’s a little frightening how good it is
I will say asking Deep Research mode about theory is _not at all_ like a patient asking a question of a standard model. However well Gemini did on this, it doesn't _understand_ what it said. I know when I've been researching things and Google AI has provided an unasked-for answer, that answer is often wrong, in ways that would be consequential for me long term.
One case I particularly remember was when it told me in no uncertain terms to stop exercising. No caveats, no "check with your provider," just stop. I have in fact checked with my providers and every last one has told me to keep it up.
As the comments below the basic science assumptions of the AI response need checking - but isn't that the problem - whereas Prof G did this with the years of experience of reading, critically analysing and researching MS, how will others use the results of asking AI anything. AI could be viewed as another tool and advance of human kind as we moved from pen and paper to computers, from libraries and knowledge being in the hands of the few to the internet where knowledge has been shared. The real question is what we do with that knowledge. The time taken to read the primary research and papers (hours) by an inquisitive mind leads to further questions which can be looked into and researched. AI generates answers in seconds and where is the cultivation of further questions, where is the creation of the desire in a young researcher to do something with the answers they are finding and then wanting to challenge.
A lot of the comments below - state the AI answer was too long to read - and depending on the use maybe it is - but for scientific breakthroughs to continue - creation of the knowledge and information which is then used by future AI answers - does it need to be longer - so the effort into reading, understanding and reflecting on the answer creates a passion for finding out, answering and of course solving more.
AI is a tool that can be used to aid research and maybe even refine how it is written - "ask AI to make the write up better", but just as we all use spell check - it doesn't mean we shouldn't learn basic skills the old way to create the use of AI tools is just that - a tool to aid us - not to replace us.
Overall Prof G - we all still need you to push the boundaries of MS research but also explain it to all PLwMS, HCPs and others. Prof H - you and Gavin both embody the shifting of the paradigms discussed by Kuhn (and you regularly) and ones which are uncomfortable for those who enjoy the status quo - but transformational change and advancements requires us all to embrace the discomfort and move through it.
AI has a role to play in e.g. scanning MRIs but I'm not prepared to ask questions about my health of a machine that scrapes the Internet and reproduces a mean average, independently of the plausibility. I also value human interaction. It's bad enough that theoretically human neuro is an unfeeling cyborg, don't have me talking to robots.
Also that AI podcast the other day was just awful. I work in translation and the lack of localisation (as in, the America-centrism, from the accent to the expressions to the highly cringe interpersonal dynamic) was intolerable. Either record your own audio or stick to text.
I am always surprised that HAM/TSP (Human T-lymphotropic virus type I-associated myelopathy/tropical spastic paraparesis) doesn’t get more attention in these sorts of discussions. As you are doubtless aware TSP/HAM is a chronic demyelinating disease caused by a virus (well a retrovirus) - the Human T-lymphotropic virus type I (HTLV-I.) And you may recall that HIV was called HTLV-3 when first identified some 40 years ago. TSP/HAM is sexually transmitted, unlike MS, but if you put this to one side would it be worth asking Google Gemini Pro 2.5 Deep Research to look at differences/ similarities between these two demyelinating conditions in humans? It might yield something of interest as well as spare the lives of lots of mice 'fed' with cuprizone.
It's a long read, but worth chewing over. I have always thought that MS is a condition that is both genetic and environmental, which leans towards the outside/in model. However, I am not sure. It's puzzling, and that's why I am not. doctor!
I hope that medical diagnostics will not be replaced by AI.
So prof G- how would you rate ai here? Also, did you train Gemini here (via access to your favorite studies in a notebookLM) or was this truly just a q/a to the standard model?
Very many thanks for your important and highly relevant post, and for the opportunity of a personal comment in response to your final question, 'Has Gemini misunderstood the literature? Is there something missing?
As you know, over the past 40 years I have been publishing data on the epidemiology, microbiology, immunology, pathology and most recently, back to the mystery of the incidence of MS recorded on the Orkney islands. I have been highly privileged to be expertly helped by many colleagues to follow this particular route of exploration, and I believe that the evidence accumulated, credibly and specifically points to the involvement of a specific bacterial toxin (sphingomyelinase) as the initiator of the initial primary lesion of MS, in advance of demyelination, the toxin gaining access directly to the CNS via the paranasal mucosa. This is no place to present the evidence that I have published over the years, but towards the end of my career, I feel that I owe colleagues, and patients an appeal to MS investigators to read and assess my work. There IS something missing and I believe that my evidence justifies serious further investigations. Some readers may be interested to look at the course of my investigations as published in the scientific literature. The best source ( they are all available, good old Google!) which is up to date is the references in my paper On the Orkney mystery, 'The risk of Multiple Sclerosis on the Orkney Islands. A review of the search for distinctively Orcadian risks, with a hypothesis for further investigations. Frederick Gay Multiple Sclerosis and Related Disorders 82, (2024). Thank you once again for your remarkable and relevant post. There is something missing, and it needs to be identified.
I have to admit that I got it wrong. I have received this from Google AI and have to apologise for looking in the wrong AI. /Users/derek/Desktop/Fwd ms.pdf
I love this article. It is clear and beautifully constructed. Having SPMS myself, it gives me a way to speak to others about my condition that isn't a lot of assumptions and voodoo. Thank you so much for sharing this.
Thanks so much for this very important and relevant post. I have been a believer in the "inside-out" model of MS for a long time, but there do not seem to be many in the MS community who believe this model. I personally believe that what I refer to as the "conventional MS model" (the "outside-in"/MS as a primary autoimmune disease) was created and exists to benefit the "medical-pharmaceutical-industrial" complex in that it allows and supports the pushing of the immunosuppressive and/or immunomodulatory MS "DMTs" and places the most emphasis on relapses and new conventional MRI activity, such as Gd+-enhancing lesions. I have always disagreed with the USA FDA's labeling indications for the DMTs that state "for the treatment of relapsing forms of MS" (*or in the case of Ocrevus, PPMS), because as I see it, the "DMTs" do NOT "treat MS"-rather, they reduce some manifestations of active inflammation: relapses and new conventional MRI activity. If MS were to be seen, classified, and treated as a primary neurodegenerative disease, at this time, I do not believe there would be any "treatments"-it would be a similar situation to ALS, frontotemporal dementia, Alzheimer's, etc, where the focus would be on any DMTs that might be able to "slow progression," which has not been the primary endpoint of MS DMT trials, except for some trials specifically for progressive forms of MS. It would be MUCH harder and more expensive and time-consuming for the pharma companies to try to come up with "treatments" for MS as a primary neurodegenerative disease vs a primary autoimmune disease.
I would love to hear your thoughts about this and if you feel we will see a shift in the MS community (esp among clinicians, researchers, regulatory bodies, pharma companies, etc) towards the "outside-in" model and away from the "inside-out" model.
Thank you!
I think we will see a shift, but it will be a painful transition. MRI (despite all its virtues), is a half-blind double-edged sword, leading us to neglect the things we cannot see. I fear some newer therapies for smouldering MS may fall on that sword.
I completely agree with your statements here. We know from the research at this point that there is SO MUCH MS disease activity and pathology that cannot be seen on conventional MRI, yet conventional MRI continues to be a major focus and primary endpoint of the MS DMT clinical trials as well as patient monitoring and assessment.
When you stated that "I fear some newer therapies for smouldering MS may fall on that sword," I am wondering if you might be referring to the GEMINI trial of tolebrutinib for RRMS that was deemed a "failure" since tolebrutinib did not outperform Aubagio in terms of ARR, yet we know from research now that relapses are not associated with long-term progression and CDW in MS (the UCSF EPIC study results evidenced this, as one example).
I think one aspect of the transition will also be the role that the revised 2024 McD diagnostic criteria for MS plays during this time in terms of its purported view of the "biological basis" of MS and moving away from the current phenotypes of MS (RIS, CIS, RRMS, SPMS, PPMS).
You might be interested to see this website that Sanofi has created in advance of what seems to be the FDA's expected approval of tolebrutinib here in the USA later this month for non-relapsing SPMS: https://www.rediscoverms.com/en-us/. Sanofi putting this information out is a "double-edged sword" as well, I feel.
A long read which I skimmed.
I suspect Charcot, c.130 years dead, would be turning in his grave.
Recent thinking has suggested that MS is one disease. Yet this ‘one disease’ has yet to be cracked despite unbelievable advances in imaging, computer power…. What have 40 ECTRIMS annual meetings really achieved? Did the recipients of the Barancik and John Dystel prizes really deserve them given the most basic questions about this one disease have yet to be answered?
MSers wanted one thing from the 4-5 decades of well funded research - to stop the underlying neuro-degeneration which leads to unrelenting disability progression. Some even hoped for a degree of neuro-restoration. Yet the drugs (the MS drugs market is projected to grow to c.$41 billion by 2032)to date have only tackled relapses not the neurodegeneration. Relapses were the low hanging fruit.
The MS Society websites claim that ‘there’s never been a better time to get MS’ and ‘MS research has never been so exciting’. Yet, in 2025, MSers are still moving from crutches to manual wheelchairs to electric wheelchairs. My GP visits two women with MS in their mid-30s who are in care homes. The use of AI in MS research will make little difference to them.
MS is often portrayed as the poster child of neurological diseases. Not difficult given the dire progress relating to MND, Parkinson’s and Huntington’s. But with so many basic questions yet to be answered, the MS research field needs a shake up and a kick up the backside. Too many broken promises (remember NMSS Promise 2010?), too many progressive MS trial failures (MS Smart, MS Stat)…… I hope that AI doesn’t just generate another thousand+ rabbit holes to keep Professors of neuro-immunology, ever expanding research teams, and an army of PhD students busy for the next 25-30 years.
100% agreed
Re: "too many progressive MS trial failures ..."
Not sure if this is correct. Once we learned how to do progressive trials, we have had some success, i.e. ocrelizumab, siponimod, tolebrutinib, ...
I will be presenting the O'Hand study at this year's ECTRIMS, which is based on insights from our PROMISE 2010 work. It may not all be bad news or no news at all. ;-)
CUPID trial, Lamotrigine trial… There’s a long list of progressive MS trials that failed and others which simply never reported.
BTK inhibitors look the most likely to deliver some success in the near future.
Enjoy Barcelona.
I strongly favour the inside-out view. Otherwise it is difficult to explain evident abnormalities brewing in pre-lesional “normal appearing” white matter (NAWM), that are measurable up-to 2 years before a Gad enhancing (or new T2) lesion appears.
I worry there is a false dichotomy here, between inflammation vs. neurodegeneration. Plenty of what is coming from inside-out (and secondarily propagating Gad lesions) could still be inflammatory. Smouldering inflammation (and not just microglia) compartmentalised within the CNS could rationally be a major driving factor for the mitochondrial dysfunction and eventual axonal degeneration seen in MS. Indeed, cortical lesions (that often co-localise with meningeal B-cell follicles) independently correlate with metrics of axon loss in the NAWM.
Neurodegeneration risks becoming an umbrella term, that incorrectly swallows up all the smouldering inflammation which: 1) we cannot see on conventional MRI and 2) fails to respond adequately to the current crop of anti-inflammatory therapies.
The risk of inaccurate terminology is that we may prematurely divert attention to neuroprotection (which I agree is important), before fully addressing the unmet need of smouldering inflammation (compartmentalised within the blood brain barrier).
If your house was on fire, you’d probably wait for the fire brigade to fully put all the flames out, before you start to redecorate and install new furniture.
Best regards
Nij
You definitely still have a role in the brave new world of AI! But not all of your peers will. The difference is how you centre patients. Whether NHS, academic or AI generated information and data, you’re making knowledge accessible to support informed choice, respectfulness and dignity - and through that helping patients feel less isolated and hopeless. That’s irreplaceable. Those who provide a poor service are on borrowed time. AI will change the horizon on patient experience and options. People will always prefer people - but where human medics let them down, AI-assisted healthcare will increasingly offer an alternative. It won’t offer the quality that you do but how many of your peers can either.
Gavin,
Thank you for the opportunity to comment on the back and forth we have been having with others about the potential role of AI, and this example. My, and others', comments in the to- and fro- were that the level of thinking and interpretation required were mainly the domain of the human brain, and that i was unclear if AI could provide this; although AI could facilitate this. You commented that you expected AI to achieve the levels of thinking and interpretation.
My thoughts are:
This is very nice summary to read. Is it accurate? To be absolutely honest, that's hard for me to evaluate as some of this basic science is out-with of my expertise and that would be the role of those people with the expertise to debate. but let's assume it is.
Additional important issues for me are:
This, i presume, is a synthesis of available literature. I see no QC. To what extent are the various components substantiated and accurately interpreted, rather than the interpretations of others?
Whilst the discrepancies for the inside-out and outside-in hypotheses are mentioned, what alternative hypotheses are generated from them by AI, or support for that, rather than to say that, at the end day, MS may be a combination of the inside-out and outside-in hypothesis. Feels like a bit of a cop out.
There is mention that MS doesn't meet criteria as a canonical autoimmune disease. Given concerns raised over the years about the suitability of the MS diagnostic criteria, and their developments, how suitable are these canonical autoimmune disease criteria. Again, above my level of expertise but others should comment, critique and challenge.
AI can highlight discrepancies, but to what extent can AI help us understand the nuances meaning of discrepancies from theory and expectation.
Can AI help us determine exactly what work is needed to clarify the specific work required to progress out understanding. For example, in the section titled "Implications for research and therapeutics" my understanding is that AI has trotted out the list of suggestions of others. Nothing new. Perhpas my expectations are unreal
So, as of yet, whilst i can see a very clear role of AI for summarizing and synthesizing available information, whether it be written or numeric, it is the critique and interpretation of that information, and the thinking about markedly alterative ideas, that is required for scientific development. Here the human brains of unique individuals become irreplaceable.
Given my last viewpoint, is AI, when set in a Kuhnian frame of reference, a winning combination for speeding up scientific developments?
To understand what i mean by this question, and why i think it is relevant, I need to explain a bit about the work of Thomas Kuhn.
Thomas Kuhn wrote a challenging-to-read book, The Structure of Scientific Revolutions (1962), and subsequently wrote and thought extensively and deeply about the issues contained therein. In his book, Kuhn proposed that major developments in science come from paradigm shifts in thinking; where a paradigm shift represents a very substantial, often radical, change in opinions or theories, rather than just a small improvement in understandings. Kuhn argued that paradigm shifts are a natural part of the cycle of scientific development.
Kuhn's cycle, in simple terms is this: we have an idea about something scientific - an hypothesis - we gain information over time that supports and refutes the hypothesis. Over time, evidence against our original hypothesis mounts and new hypotheses are proposed. Tensions rise and tempers fray in the scientific field between proponents of the new and original hypotheses. The new hypotheses are typically ignored, dismissed and/or ridiculed by developers and supporters of the original hypotheses until finally a new hypothesis is accepted because the evidence for it is overwhelmingly greater than for the original hypothesis. Then, the original hypothesis generators said they knew this all along and it was their idea in the first place. And the cycle continues.
Kuhn's work was very controversial when it landed. Not because of the idea of a cycle of development with paradigm shifts, as he gave multiple exemplars from physics, but because of his ideas of some of the related issues. There were three issues proposed by Kuhn that scientists found particularly jarring:
First, Kuhn said that most scientific research adds nothing to the field in terms of development. This, Kuhn said, was because most research was not aimed at solving the difficult problems in a field. Unsurprisingly, as Kuhn suggested that this comment applied to most research in a field, it was not well received (although many quietly agreed in terms of their own work!).
Second, Kuhn suggested that people responsible for paradigm shifts were from out-with of the mainstream scientists of the field of interest. This was not well received for obvious reasons.
Third, and the most jarring, was that Kuhn argued that a major cause of hindrance to the scientific development of a field was the scientists working within that field. He cited the propagation of old ideas, doing the same thing over and over again, resistance to new ideas and new personnel, lack of inclusivity, and failure to undertake problem-solving coordinated targeted work. In saying this Kuhn had released a very large lead balloon.
Whilst Kuhn's work was controversial there are things the MS field might reflect on, and that AI could help facilitate; hence my question. There has been a gradual build up over time of evidence against the original ideas of the biology of MS (outside-in hypothesis). A new hypothesis, the inside-out, has been proposed and increasing, but incomplete, evidence for it provided. This sounds uncannily like a Kuhnian cycle moving toward a paradigm shift. But is the inside out hypothesis the paradigm shift required or is their another alternative? Can AI help facilitate our speed round the natural Kuhnian cycles we are in? What coordinated, targeted, problem-solving research in MS is required? What work out-with of the field of MS can help us tackle these issues. Can AI help the MS field maximize inclusivity to fast-track our work and ultimately improve the opportunities for people with MS.
Finally, but importantly and very relevant, Kuhn also wrote on the importance of measurement in science; The function of measurement in modern physical science (1961). He highlighted not just the importance of measurement - i.e., its quality not just its use - but that the role was to highlight anomalies (discrepancies from expectations). This aligns with his cycle, where the testing of hypotheses identifies the anomalies that drive reconsideration of our theories, and modification of existing hypotheses and/or the generation of new hypothesis. If we took note of this we would pay much more attention to our measurements in MS research and the anomilies they highlight.
So, in short, AI to help facilitate progress. But the hard yards of thinking and interpretation require the human brain. Therefore, to maximise our ability to solve the main problems in MS we need AI, our clinicians and researchers to think more, think harder, think deeper, i believe, invoke Kuhn's work better our measurements.
As I have responded well by treating my MS as EBV driven for twenty years, and taken an antiviral as a consequence, I'm sure it is an 'inside out' pathology.
What is missing, in majority of articles, is a description of how that infection actually changes things.
My thinking is the damage comes from purinergic receptors being inappropriately opened, allowing an influx of calcium that causes cell death. That destruction is what the immune system reacts to, creating the impression of 'outside in' activity.
What drives the receptors to be opened is too much adenosine created when EBER2 outcompetes adenosine deaminase (not damages, just outcompetes it ) as described by two Israeli researchers nearly thirty years ago https://www.researchgate.net/publication/51313862_Does_EBV_RNA_modulate_ADA_mRNA_translation
Too much adenosine will create lassitude and down regulate T cells response ( per Penders observation for CD8+ T cells in 2017) .
Adenosine deaminase ends up elevated in CSF of MS patients
https://www.sciencedirect.com/science/article/abs/pii/S2211034817300226#:~:text=Polachini%20CR%20and%20colleagues%20showed,a%20biomarker%20for%20clinical%20diagnosis.
and adenosine shows up as lowered in their serum https://www.sciencedirect.com/science/article/abs/pii/S0306452214000694#:~:text=Results%20showed%20that%20AChE%20in,neurological%20dysfunction%20of%20RRMS%20patients.
Consequently, ATP is mucked up.
I think adenosine explains your smouldering MS.
Let’s be honest it’s going to affect everyone/everything in the coming years , there’s no turning back so might as well embrace it 🤷♂️, it’s a little frightening how good it is
I'm not wading through that.
I will say asking Deep Research mode about theory is _not at all_ like a patient asking a question of a standard model. However well Gemini did on this, it doesn't _understand_ what it said. I know when I've been researching things and Google AI has provided an unasked-for answer, that answer is often wrong, in ways that would be consequential for me long term.
One case I particularly remember was when it told me in no uncertain terms to stop exercising. No caveats, no "check with your provider," just stop. I have in fact checked with my providers and every last one has told me to keep it up.
Google AI mode has very little to do with gemini (or openai) Deep Research
Gemini Deep Research is Google's LLM.
As the comments below the basic science assumptions of the AI response need checking - but isn't that the problem - whereas Prof G did this with the years of experience of reading, critically analysing and researching MS, how will others use the results of asking AI anything. AI could be viewed as another tool and advance of human kind as we moved from pen and paper to computers, from libraries and knowledge being in the hands of the few to the internet where knowledge has been shared. The real question is what we do with that knowledge. The time taken to read the primary research and papers (hours) by an inquisitive mind leads to further questions which can be looked into and researched. AI generates answers in seconds and where is the cultivation of further questions, where is the creation of the desire in a young researcher to do something with the answers they are finding and then wanting to challenge.
A lot of the comments below - state the AI answer was too long to read - and depending on the use maybe it is - but for scientific breakthroughs to continue - creation of the knowledge and information which is then used by future AI answers - does it need to be longer - so the effort into reading, understanding and reflecting on the answer creates a passion for finding out, answering and of course solving more.
AI is a tool that can be used to aid research and maybe even refine how it is written - "ask AI to make the write up better", but just as we all use spell check - it doesn't mean we shouldn't learn basic skills the old way to create the use of AI tools is just that - a tool to aid us - not to replace us.
Overall Prof G - we all still need you to push the boundaries of MS research but also explain it to all PLwMS, HCPs and others. Prof H - you and Gavin both embody the shifting of the paradigms discussed by Kuhn (and you regularly) and ones which are uncomfortable for those who enjoy the status quo - but transformational change and advancements requires us all to embrace the discomfort and move through it.
AI has a role to play in e.g. scanning MRIs but I'm not prepared to ask questions about my health of a machine that scrapes the Internet and reproduces a mean average, independently of the plausibility. I also value human interaction. It's bad enough that theoretically human neuro is an unfeeling cyborg, don't have me talking to robots.
Also that AI podcast the other day was just awful. I work in translation and the lack of localisation (as in, the America-centrism, from the accent to the expressions to the highly cringe interpersonal dynamic) was intolerable. Either record your own audio or stick to text.
I use deep research (Google or openai, Claude is not worth it) quite often, for MS and other stuff. Occasionally even for work.
Not perfect but much more efficient than dealing with google scholar
I am always surprised that HAM/TSP (Human T-lymphotropic virus type I-associated myelopathy/tropical spastic paraparesis) doesn’t get more attention in these sorts of discussions. As you are doubtless aware TSP/HAM is a chronic demyelinating disease caused by a virus (well a retrovirus) - the Human T-lymphotropic virus type I (HTLV-I.) And you may recall that HIV was called HTLV-3 when first identified some 40 years ago. TSP/HAM is sexually transmitted, unlike MS, but if you put this to one side would it be worth asking Google Gemini Pro 2.5 Deep Research to look at differences/ similarities between these two demyelinating conditions in humans? It might yield something of interest as well as spare the lives of lots of mice 'fed' with cuprizone.
It's a long read, but worth chewing over. I have always thought that MS is a condition that is both genetic and environmental, which leans towards the outside/in model. However, I am not sure. It's puzzling, and that's why I am not. doctor!
I hope that medical diagnostics will not be replaced by AI.
So prof G- how would you rate ai here? Also, did you train Gemini here (via access to your favorite studies in a notebookLM) or was this truly just a q/a to the standard model?
Standard model using its Deep Research mode.
Incredible.
Very many thanks for your important and highly relevant post, and for the opportunity of a personal comment in response to your final question, 'Has Gemini misunderstood the literature? Is there something missing?
As you know, over the past 40 years I have been publishing data on the epidemiology, microbiology, immunology, pathology and most recently, back to the mystery of the incidence of MS recorded on the Orkney islands. I have been highly privileged to be expertly helped by many colleagues to follow this particular route of exploration, and I believe that the evidence accumulated, credibly and specifically points to the involvement of a specific bacterial toxin (sphingomyelinase) as the initiator of the initial primary lesion of MS, in advance of demyelination, the toxin gaining access directly to the CNS via the paranasal mucosa. This is no place to present the evidence that I have published over the years, but towards the end of my career, I feel that I owe colleagues, and patients an appeal to MS investigators to read and assess my work. There IS something missing and I believe that my evidence justifies serious further investigations. Some readers may be interested to look at the course of my investigations as published in the scientific literature. The best source ( they are all available, good old Google!) which is up to date is the references in my paper On the Orkney mystery, 'The risk of Multiple Sclerosis on the Orkney Islands. A review of the search for distinctively Orcadian risks, with a hypothesis for further investigations. Frederick Gay Multiple Sclerosis and Related Disorders 82, (2024). Thank you once again for your remarkable and relevant post. There is something missing, and it needs to be identified.
I agree with you. The AI left out the possibility of a systemic or locally produced toxin as part of the cascade that leads to the focal MS lesion.
I have to admit that I got it wrong. I have received this from Google AI and have to apologise for looking in the wrong AI. /Users/derek/Desktop/Fwd ms.pdf
I love this article. It is clear and beautifully constructed. Having SPMS myself, it gives me a way to speak to others about my condition that isn't a lot of assumptions and voodoo. Thank you so much for sharing this.