52. Pushing back the boundaries of research with AI
Duration 00:33:47
Spotify
Apple Podcasts
YouTube
Transcript
CORDIScovery podcast- Episode #52 - Pushing back the boundaries of research with AI
This is an AI transcription.
00:00:00:00 - 00:00:39:04
Abigail Acton
This is CORDIScovery. Hello and welcome to this episode of CORDIScovery with me, Abigail Acton. Artificial intelligence. Are we heading for a dystopic future, or one in which labor force is relieved from menial tasks? Machine learning is second guessing us. Is that making our lives easier? Or is it intrusive? Corporates are hoovering up content to train large language models in ways that are hard for users to control.
00:00:39:08 - 00:01:04:12
Abigail Acton
It's clear AI is huge, but good, huge or bad huge? That seems to be up for debate. There's no doubt that AI and related tools are pushing back the boundaries of research. Today, we do have some clarity, at least when it comes to our three projects, all of whom have received EU research and innovation funding. Our three guests are using these tools to develop new technologies and make discoveries that would have previously been impossible.
00:01:04:14 - 00:01:29:13
Abigail Acton
We're looking at the way in which AI is pushing back the boundaries of knowledge to improve the prevention, treatment and rehabilitation of stroke patients. To give a voice to the previously unheard women of historical Ireland, and to develop a small, powerful bio threat detection system to save lives. John Kelleher is the director of the Adapt Research Ireland Center for AI driven Digital content Technology.
00:01:29:15 - 00:01:40:08
Abigail Acton
And professor of computer science at Trinity College Dublin. He focuses on harnessing AI to enhance the understanding and treatment of complex medical conditions. Hello, John. Welcome.
00:01:40:11 - 00:01:44:20
John Kelleher
Hello, Abigail. I'm delighted to be here and very much looking forward to our discussion.
00:01:44:22 - 00:02:01:09
Abigail Acton
Jane Ohlmeyer professor of modern history at Trinity College Dublin and former chair of the Irish Research Council, is an expert on new British and Atlantic histories. She is applying AI to the recovery of the lived experience of ordinary non-elite women in early modern Ireland. Hello, Jane.
00:02:01:11 - 00:02:02:23
Jane Ohlmeyer
Hello. Lovely to be here.
00:02:03:00 - 00:02:09:01
Abigail Acton
Lovely to have you. Béla Mihalik is a senior developer at Ideas Science in Hungary.
00:02:09:03 - 00:02:19:12
Abigail Acton
He specializes in the application of AI and deep learning to develop novel tools that can rapidly screen potential bio threats in the form of pathogens and bacteria. Hi Béla.
00:02:19:14 - 00:02:21:00
Béla Mihalik
Hello. Good to be here.
00:02:21:06 - 00:02:39:13
Abigail Acton
Good to have you. Thank you. John I'm going to turn to you first. The STRATIF-AI project builds on the current use of machine learning applied to stroke patients, to make the response far more customized and time sensitive. John. Healthcare is one of the areas that is most benefiting from AI. Can you tell us a little bit about the goals of the STRATIF-AI project?
00:02:39:15 - 00:03:07:03
John Kelleher
Yes. Abigail, STRATIF-AI Project is interested and focused on integrating heterogeneous data sets to drive insights for patients across their lifetime and with a particular focus on how stroke affects them. So we are trying to make continuous stratification, and that's where the name of the project comes from. We're trying to create AI tools that continuously stratify somebody's stroke risk throughout their lifetime and help them to manage that stroke risk and also have better outcomes after stroke treatment.
00:03:07:05 - 00:03:16:06
Abigail Acton
Okay. Super. How did you identify the patients that you would follow and how did you enroll your participants in this?
00:03:16:08 - 00:03:40:01
John Kelleher
Our ultimate goal would be to have broad coverage across a population rather than, a particular cohort. We want to be able to take for the whole population to develop tools that are suitable for everybody that would be able to take somebody through from, from a very young through their whole life through to further had a stroke under in acute care versus reintegration back into society after that care.
00:03:40:01 - 00:03:45:00
John Kelleher
So the the ultimate ambition is for everybody to be supported and enabled through this research.
00:03:45:02 - 00:03:53:01
Abigail Acton
Okay, fabulous. That sounds really, a really good goal. How are you applying artificial intelligence to this? What is what is artificial intelligence enabling you to do?
00:03:53:03 - 00:04:17:06
John Kelleher
So we're bringing together two different types of artificial intelligence in this in this research, the first type is called the digital twin model. So a digital twin is the idea of we create a computational model that is personalized to an individual. Creating a digital twin, makes the model quite interpretable. Because of the structure of the model. We're able to understand how the different variables that represent the person's health state evolve through time.
00:04:17:08 - 00:04:56:23
John Kelleher
And but it has some challenges in terms of being able to develop those models to scale to large populations and integrate very different types of data. So we're combining those digital twin technologies with what we call machine learning. So machine learning is a way of analyzing very large data sets to extract patterns and interactions between different features. So by bringing these two types of machine learning and digital twin artificial intelligence approaches together, we're trying to make personalized, interpretable systems that can scale and bring together large heterogeneous data sets so that an individual is able to understand their stroke risk throughout their lifetime and manage it right.
00:04:56:24 - 00:05:02:13
Abigail Acton
Excellent. And then also perhaps to to receive advice that would enable them to reduce their risk.
00:05:02:15 - 00:05:30:00
John Kelleher
Oh exactly. Yeah. So we'd like to be able to, for example, will tailor the interventions that are suggested to a person to their stroke risk and help them to understand across their lifetime how that evolving and how modifications in their lifestyles can affect that. So I think one of the really interesting things that this might link to is that actually, the earlier on, somebody makes a change in their life in terms of a stroke risk, the greater the, the effect it has on the cumulative approach over their life course.
00:05:30:05 - 00:05:58:09
John Kelleher
So even a very small change very early on in their life, for example, in terms of modifying their cholesterol early on in life, can have a massive benefit across their lifespan. And the earlier we intervene in somebody's life course, the cumulative benefit is much, much greater. So you can imagine that if we wait till later on in life to intervene in somebody's health trajectory, we have to do a more intense intervention, for example, using drugs.
00:05:58:11 - 00:06:10:03
John Kelleher
So what we'd really like to do is, early on in life, allow people understand the risk and do small changes to their lifestyles that helped them to avoid bigger interventions later in life.
00:06:10:05 - 00:06:26:22
Abigail Acton
And the difference with this, of course, from from current sort of, approaches, from the health care, system where, you know, you're told to eat your five vegetables and fruit a day and, you know, walk however many steps the current thinking is and so on. The difference here is that your advice would be very much tailored to the individual.
00:06:26:22 - 00:06:35:13
Abigail Acton
So you'd be getting a, a much more textured, in-depth insight into the various things that all interplay together. Is that correct?
00:06:35:15 - 00:06:59:01
John Kelleher
Yes. So one of the fascinating things about health and medicine is that while we can get, we can extract information about risk factors at a population level, we are all individuals and we are all different. And how they interact within our own systems are different. So trying to make models that are personalized and tailored to individual to capture risk factors, is what we're trying to do here.
00:06:59:01 - 00:07:32:24
John Kelleher
So, for example, you might consider that some people, have a genetic predisposition to particular risk factors. And there's aspects called epigenetics where our lifestyles turn on and off genes that cause risk factors or that exacerbate or amplify risk factors or reduce risk factors. So trying to integrate, for example, genetic information with lifestyle, nature and nurture, if you like, and give us a much more personalized risk for an individual than if we just consider somebody as somebody in a population similar to everybody else.
00:07:32:24 - 00:07:47:15
Abigail Acton
And of course, the benefit immediately is that with that personalized input, you're far more motivated to to follow the suggestions because you know, it is for I like to know, for John Kelleher or for whoever it's specifically tailored for you. So I think maybe you take it more seriously.
00:07:47:17 - 00:08:08:19
John Kelleher
Yes, I'm actually one of the really exciting aspects of the project is that our partner in Linköping, who's coordinating the project, being a citizen, they have the idea that we'll take this digital twin, and the nice aspect of the digital twin is we can simulate forward in time your health state. And what we are developing is the concept of a digital twin, where there is a graphical interface, a digital twin.
00:08:08:21 - 00:08:26:10
John Kelleher
And we can not only tell you about how your risk is, but it will evolve through time. But we can also show you graphically to make it much more concrete, to show you, for example, how you would age with this lifestyle versus another lifestyle to try and improve, compliance and interventions to help people and motivate them.
00:08:26:16 - 00:08:39:13
Abigail Acton
Absolutely. And it sounds very motivational. And is this something that would be on someone's telephone app, for example, or would it be like a phone app? You'd just be able to pull it out of your pocket and see the choices that you were making, how they would impact on you in real time. Would that be the ultimate goal?
00:08:39:15 - 00:09:06:12
John Kelleher
So we have two different aspects. Yes. One is for you have a health vault on your phone for you can carry around on your phone your own personalized health, profile and that chronic health record, but also where we'd have a larger graphic display that would be used in an intervention called a health dialog, where you would go to somebody, maybe every ten years or five years, you'd go in and speak to a nurse or a clinician, and we would be able to simulate there and give you a develop a plan with your input.
00:09:06:18 - 00:09:27:09
John Kelleher
Or we could simulate for you a whole body scan, and we're able to work with you and say, well, if you change this and this, if you, for example, reduce your calorie intake, reduce your cholesterol, this is what happens, or that's what happens. And with that insight given to you, have a dialog with you to think about. What's the plan you would like to adopt over the next 5 or 10 years?
00:09:27:10 - 00:09:28:00
Abigail Acton
Fantastic.
00:09:28:02 - 00:09:34:17
John Kelleher
So to have patients and clinicians speaking to each other and co-developing your plans for treatment and support.
00:09:34:17 - 00:09:45:08
Abigail Acton
So collaboration rather than sort of instruction. Fantastic. Do you imagine that digital twins can be used elsewhere in the health care processes? Are there other applications that you can think of as being useful?
00:09:45:10 - 00:10:06:19
John Kelleher
Absolutely. So the the really exciting thing about digital twins is their ability to simulate the evolution of a system. So what that means, for a system here could be the, evolution of a disease within an individual. So, for example, any, any types of health where we have a chronic, long term evolution of that disease within an individual.
00:10:07:00 - 00:10:33:16
John Kelleher
Digital twins can support that. But also if we move away from thinking about precision medicine for an individual, we can also think about, for example, even the logistics of healthcare. So simulating hospital operations and facility management, because the digital twin can simulate either an individual or a physical system. So maybe we can improve the logistics around healthcare, reducing the economic burden of healthcare.
00:10:33:18 - 00:10:51:22
John Kelleher
But there's also the idea of, drug development and medical device simulation. So trying to use these technologies to increase the rapidity in which we can develop new treatments or develop devices and simulate devices and interventions on individuals to check how effective they are before they're deployed.
00:10:52:02 - 00:11:01:02
Abigail Acton
And what about the digital twin of actually a pathogen, so that you could then see the impact of treatment on that pathogen in the form of the digital twin?
00:11:01:04 - 00:11:23:07
John Kelleher
Absolutely. So the challenge with developing digital twins, and one of the things we're really tracking or addressing within the challenge, digital twins have huge capabilities and advantages going forward. The difficulty with them is that in order to implement them, we need a theory, theory or model for how a system, a process, evolves through time, and that can be difficult to implement.
00:11:23:12 - 00:11:43:24
John Kelleher
And that's where the complementarity between the digital twin and machine learning comes together. Because machine learning enables us to model some of the interactions on high level, complex interactions that we might think us have a theoretical knowledge for. And integrating that with the digital twin allows the digital twin to model in an interpretable and comprehend computation feasible manner.
00:11:44:01 - 00:11:52:17
John Kelleher
The areas that we do understand but integrate the patterns that machine learning takes from large data to help the whole system work more comprehensively.
00:11:52:19 - 00:11:59:11
Abigail Acton
Fantastic. It sounds like a wonderful way that AI is pushing back the boundaries of research. Thank you so much, John.
00:11:59:13 - 00:12:21:06
John Kelleher
That's why I think that the integration of the digital twins, the theoretically driven, if you like, approaches versus the data driven machine learning approaches that captured those high level interactions, maybe before we understand them is such a promising approach. This hybrid modeling will allow us to start building systems that integrate both what we understand and what we have.
00:12:21:06 - 00:12:32:23
John Kelleher
what we can model which are different things at some point, integrating them now and then slowly with those initial systems growing the amounts that we understand, to be more interpretable.
00:12:33:00 - 00:12:54:23
Abigail Acton
Fantastic. Thank you John. I'm going to turn to Jane now. The role of women in early modern Ireland was pivotal in a time of societal change, but the stories of the women highlighting their resilience amidst social upheaval and trauma have largely been lost to history. The voices project let them be heard. Horizons are also opening up in the humanities thanks to AI.
00:12:55:02 - 00:13:00:19
Abigail Acton
What was the challenge facing you before such tools became available, and what can you do with them now?
00:13:00:21 - 00:13:22:16
Jane Ohlmeyer
The big challenge was really around access access to the manuscripts themselves and then access to material that was available in digital format but wasn't interoperable. It didn't. These digital silos weren't talking to each other. And it's in those two particular areas that AI really has been very enabling for our work.
00:13:22:16 - 00:13:38:17
Jane Ohlmeyer
And I think that the potential is we are just beginning to explore it. But if we can get it right, AI really will be transformational when it comes to doing historical research, not just about women, but obviously our focus is women.
00:13:38:17 - 00:13:46:05
Abigail Acton
Absolutely I guess. I mean, it seems hard to imagine, even as a journalist before. Before all of these tools became available and so on.
00:13:46:05 - 00:13:53:22
Abigail Acton
But, I mean, how would something like this have been done in the past? Would you even have been able to tackle such a large scale project in the past without such tools?
00:13:53:24 - 00:14:10:12
Jane Ohlmeyer
Well, AI allows us to do this history at scale because it allows us to actually interrogate records, especially, records around taxation or demographic records that it's incredibly time consuming to analyze these or legal records.
00:14:10:18 - 00:14:36:13
Abigail Acton
And what we find is that the women are deeply buried in them. Abigail. So AI is allowing us to identify those women. And once they've been surfaced, if you like. Then obviously, we're analyzing what we're finding. So it really has, has helped us enormously around the identification and access, to non-elite women who, as I say, are very much buried in the records.
00:14:36:15 - 00:14:39:01
Abigail Acton
Wonderful. What sort of things were you finding, Jane?
00:14:39:03 - 00:15:06:06
Jane Ohlmeyer
What we're finding is that actually, women have been hiding in plain sight all along. And that records that we thought were actually of no use, previously, we are finding are of tremendous value. Let me give you an example. We've got things called funeral entries for 17th century Ireland, the equivalent of, death notices today.
00:15:06:08 - 00:15:35:11
Jane Ohlmeyer
And, people had never really bothered with them. And what we're finding is over a third of them actually are by women and about women. So all of a sudden, a whole new body of material has been, if you want, unearthed, thanks to our interactions with AI or Wills testamentary material. Again, we thought that, for a whole variety of reasons, that there were no wills by women or there were very few.
00:15:35:13 - 00:15:53:13
Jane Ohlmeyer
And, we've now uncovered about 500 of them. Again, AI has been extremely helpful in allowing us a) to identify them and then to create digital surrogates that were then able to, analyze and interrogate.
00:15:54:05 - 00:16:09:10
Abigail Acton
I'm just going to cut across here because I've got so many of my this is so interesting. I've got my brain just seething with questions that I want to ask you first and then observation, is it must be a bit like being an archeologist discovering a a hitherto unknown tomb that shines a light on, on the whole kind of culture of that period.
00:16:09:10 - 00:16:12:18
Abigail Acton
And it's absolutely fascinating. Must be very thrilling. Jane is a thrilling?
00:16:12:20 - 00:16:28:04
Jane Ohlmeyer
Oh, it's so exciting, Abigail, I can't tell you because it's like you're going to something, you're not going to find anything. And all of a sudden, you know, it's a third of the records are relevant and interesting, and these are records that we thought many just didn't exist.
00:16:28:06 - 00:16:48:03
Jane Ohlmeyer
Because, in 1922, the archive in Ireland was destroyed. So, I mean, it's one of those ones where, all the material is dispersed, in other words, transcripts that are scattered all over the place that we are, thanks to the technology, are able to reconstitute, I think, in a very important and innovative way.
00:16:48:08 - 00:16:52:03
Abigail Acton
Okay. So you said that the material apparently had been lost.
00:16:52:05 - 00:16:59:09
Abigail Acton
So so where did you find it? The scattered, scattered transcripts of duplicate copies found in places, hitherto not looked at.
00:16:59:09 - 00:17:20:14
Jane Ohlmeyer
There are some wills in the National Archives in Dublin, that people really hadn't appreciated had survived the fire. So that's obvious. Right. But what we're also finding is transcripts that are in private collections or in publications, that were published before 1922, which is when the big fire took place.
00:17:20:16 - 00:17:32:03
Jane Ohlmeyer
So they're in hundreds of different locations and they're it's just they're so disparate. And this is a way of taking all these digital windfalls and bringing them together.
00:17:32:05 - 00:17:44:19
Abigail Acton
I love digital windfalls. And you. Well, it must be because wills are such a window onto a society in a culture, for example, obviously immediately comes to mind that women had property, to, to bequeath and, and all sorts of things.
00:17:44:19 - 00:17:48:00
Abigail Acton
Tell me some of the insights you're gaining from looking at such documents.
00:17:48:00 - 00:18:15:13
Jane Ohlmeyer
Well, on the wills front, it tells us a lot about material culture and the property that the jewelry, the household goods. So you've got that side of it, but also it tells you about who women hold most dear, their intimate networks, and how women tend to bequeath particularly moveable property, as well as sometimes property at land and, and, and fixed assets to other women.
00:18:15:15 - 00:18:36:16
Jane Ohlmeyer
So we're just really beginning to recover all of this, and you have to remember, in the 17th century, a woman didn't have her own legal identity unless she was a widow. So, and a married woman could only make, well if her husband allowed her to do that. So, again, it's giving us insights that we didn't think were possible to find.
00:18:36:18 - 00:18:37:18
Jane Ohlmeyer
Yeah. It's been great.
00:18:37:23 - 00:18:52:00
Abigail Acton
Yeah, I know it sounds. It sounds thrilling. Have you any, stories or examples that you've dug out that surprised you or that you feel well, maybe I can use the old cliche, a bit of a eureka moment about.
00:18:52:02 - 00:18:57:06
Jane Ohlmeyer
We're having lots of eureka moments. It’s really exciting.
00:18:57:06 - 00:19:24:11
Jane Ohlmeyer
And we're just we're only two years into the project, Abigail. So when we speak in a couple of years time, you know, so my colleague, who's doing a lot of work on the Irish Court of Chancery, which is the Court of Equity, where large numbers of women are plaintiffs and defendants. And again, it's so fascinating to see how during usually moments of personal crisis, these women are coming out of the shadows and they're going to court often to protect their own interests.
00:19:24:11 - 00:19:50:08
Jane Ohlmeyer
It might be that they've been given some, money upon marriage, and a male relative is trying to steal their money or steal their land or steal their property, or they're going to court to protect the interests of their children, or, their reputation. We have one very, very juicy account of a young woman who has entered into a relationship with a man.
00:19:50:10 - 00:20:12:11
Jane Ohlmeyer
And she's had a child out of wedlock, and she then marries. And the husband she marries is actually taking her former lover to court for damaging the reputation of his wife. So, again, these are the sorts of stories that are usually lost from history. Yes. And it gives you, as I say, a very personal insight into people's lives.
00:20:12:13 - 00:20:20:10
Abigail Acton
Absolutely wonderful. Well, thank you so much for explaining all that to us and setting it out so clearly. Does anyone have any questions for Jane?
00:20:20:12 - 00:20:42:06
John Kelleher
Actually, I will would, Jane. I'm fascinated by the VOICES project, and I'm really excited, by the way, you're using new technologies to try and understand the past in new ways, particularly technology that's evolving very rapidly. So I'd really like you to, if you could, to think about or imagine how these AI systems are evolving. It could be designed in the future to improve and support historical research.
00:20:42:06 - 00:20:47:23
John Kelleher
And what capabilities do you hope it will have, and how do you think it will enable historic research in the future?
00:20:48:00 - 00:21:17:18
Jane Ohlmeyer
Oh great question, John. I want a AI system that is transparent. I want the computer scientists, never mind the historians, to understand the algorithms that underlie it. I want an AI system that is ethically, bounded, and, that code of ethics, then protects the intellectual property of everybody who is contributing, to it.
00:21:17:20 - 00:21:48:07
Jane Ohlmeyer
But the third thing I want is an AI system that is environmentally responsible. I'm so conscious that the current system is so greedy of power and water and especially generative AI. So as we look to the future, those ethical and environmental issues for me are hugely important. And then I'll feel much more comfortable about being able to trust, the AI system that we come up with.
00:21:48:07 - 00:21:54:21
Jane Ohlmeyer
Because just at the moment, trust is my fundamental objection, or concern with AI.
00:21:54:23 - 00:22:12:15
Abigail Acton
Yes. I think we can all absolutely identify with that. Perfect. Thank you. Béla, I'm turning to you now. Terrorism can take many forms, including biological ones. Bioweapons pose a threat mainly due to detection difficulties. But that's something the HoloZcan project is tackling head on.
00:22:12:18 - 00:22:21:05
Abigail Acton
So the project's been using AI to develop equipment to help first responders in an emergency. What problem were you hoping to solve Béla?
00:22:21:07 - 00:22:54:06
Béla Mihalik
Yes. The problem, is mainly discovered by the ENCIRCLE project of the European Union, ENCIRCLE project defined the gaps, and, in the gaps they identified, there is no good onfield tool for biological detection. And we wanted to, to, to create a device which can work in on field and also possible to used in many scenarios, not just in one scenario.
00:22:54:06 - 00:23:13:12
Béla Mihalik
So it's a big challenge, of course. It's, it's over the state of the art. But, we started, to think about how we can apply microscopy, especially holographic microscopy technology, to fill a gap a bit, even if we not solve the whole problem.
00:23:13:14 - 00:23:22:11
Abigail Acton
Oh, geez. You made a great start on this on this goal. Béla what are the current ways of detecting what might be present in a biological attack?
00:23:22:13 - 00:23:50:18
Béla Mihalik
Currently, the most accepted way to detect is use PCR methods. And also, the fluorescent labeling methods, DNA sequencing, all of them needs laboratory preparation. It's a long, long method. And, for example, PCR software from false positive results. It means, it can signal, it can see it's a danger even if there is not.
00:23:50:18 - 00:24:20:20
Béla Mihalik
And it's the threat is very high above 20%, 30% for current, detectable pathogens. And in second, seeing the problem of epigenetic segments is, is currently not handled. It means, dangerous fragments can be detected from DNA, but it is not present, on, protein level. It means it's not really that harmful.
00:24:20:22 - 00:24:29:08
Abigail Acton
What you mean is that the current systems could pick up fragments of DNA? Yeah, that are not accurate in some way. What's the problem with the picking up of the fragments?
00:24:29:10 - 00:25:01:24
Béla Mihalik
The problem is that, we cannot find dangerous fragments which can coding dangerous proteins, but because of epigenetic information, the cells are not expressing, not creating these dangerous, proteins. And it's everywhere around us in the environment, this code is everywhere, but it's not expressed because of epigenetic blocks. So this is why we cannot use, sequencing in each case I see.
00:25:01:24 - 00:25:21:02
Abigail Acton
So the sequencing might be picking up something that's around us anyway. And not posing a problem. Yes, but it's identifying it and then it's confusing the end result because then the responders don't know what they're looking at specifically. Yes, I see what you mean. Okay. So how did HoloZcan aim to help all of this? And how did it use AI?
00:25:21:04 - 00:25:55:18
Béla Mihalik
In HoloZcan the focus on microscopic technology or holographic microscopic technology via holography, and it's a good question because, it's, let's say self-calibratable. It's set of calibrated, self-calibrated pictures. It means, we can numerically analyze the picture so we can get values from it under different conditions. And holography gives a very stable base for this numerical analyzes.
00:25:55:20 - 00:26:30:14
Béla Mihalik
And so we can see exact morphological properties and other optic, properties like, spectral information. And also refractive information, of the small objects and wide range of information, we can even not identify but classify, the particles. And, also we can see the complex morphologies around, in the sample and around, the, the objects.
00:26:30:18 - 00:26:52:00
Béla Mihalik
And from this, morphological complexity, with using of artificial intelligence, if we can figure out what the situation is, is it intentional? It is a manmade situation, or it's some environment or situation. So AI can help a lot, to identify the the real source.
00:26:52:02 - 00:27:21:22
Abigail Acton
Okay. Excellent. So instead of taking little tiny samples to try and identify what the material is, you've developed a system that actually is visualizing, okay, the material and then through the visualization what's happening then is AI somehow cross-referencing what the thing looks like, what the what the sample looks like or the elements within the sample look like. And then is that cross-referencing across another kind of database of similar images and deciding what it is or what happens next?
00:27:21:22 - 00:27:22:07
Abigail Acton
In fact.
00:27:22:11 - 00:27:52:06
Béla Mihalik
Yes, it started in many lines. First, of course, we use the laboratory, samples, very clean, samples, of bacteria and other pathogens. And, we created, database, and also trained, artificial intelligence to identify these different type of bacteria. And it works very well on laboratory samples. We can see it works more than 99% of accuracy.
00:27:52:10 - 00:28:01:01
Abigail Acton
I think you said something about a generation of 70,000 synthetic holograms. Yeah, that reflect real biological particles. 70,000. That's huge.
00:28:01:03 - 00:28:27:16
Béla Mihalik
Yes. We created a simulator tool because, in laboratory, we we couldn't to achieve all the samples. There are some very dangerous BSL four level, BSL four level that means, it is deadly. And we cannot cure, so we we are not allowed to, to handle this type of material. So we created simulated, samples how this looks like.
00:28:27:18 - 00:28:48:18
Béla Mihalik
And, also we could create the type of, objects, but we can never measure, but in a certain time period. But it can appear in reality. So we we use single simulation or this digital twins, let's say, the extended, the training, range of the, of the artificial intelligence.
00:28:48:23 - 00:28:52:18
Abigail Acton
So it's basically synthetic holograms like a reference library.
00:28:52:20 - 00:28:53:23
Béla Mihalik
Yes. Absolutely.
00:28:54:03 - 00:29:09:06
Abigail Acton
Lovely. So then a first responder takes this device to a situation where they suspect that there's been some sort of bio threat, and what, they take a sample of the air or the sample of the soil, they put it into the canister. Or how does that work?
00:29:09:08 - 00:29:20:18
Béla Mihalik
Actually focused only for airborne pathogens. So we use air sample collectors, Coriolis based air sample collectors and also impactors.
00:29:20:18 - 00:29:22:04
Abigail Acton
So what's an impactor?
00:29:22:06 - 00:29:45:10
Jane Ohlmeyer
An impactor is a device, when we create that air flow with ventilators and we put obstacle, in the line of airflow, and on this, obstacle, the the small objects will stick, and they pick it up for if it's, microscopic slide, and we directly can put it under the microscope and analyze.
00:29:45:14 - 00:29:48:03
Abigail Acton
And that can be done in situ in real time.
00:29:48:07 - 00:29:57:00
Béla Mihalik
Yes, it can be done in real in situation and in real time. And also we can integrate together an impactor and the microscope.
00:29:57:04 - 00:30:07:12
Abigail Acton
Right. So basically you're able to collect data in the field that will then allow first responders to see immediately what it is that has has contaminated the area. Is that correct?
00:30:07:14 - 00:30:24:14
Béla Mihalik
Yes. Is that correct? And, we can analyze the composition of air, which is the most important information for the first responders. And with this composition, we can infer, is it a real threat? It can be a real threat or it's some environmental variation.
00:30:24:16 - 00:30:37:21
Abigail Acton
Okay, but that's fantastic. So it rules out all these, false positives, which may be caused chaos. And in fact, there's nothing dangerous there. And and so on. Super. That's excellent. Thank you very much. Thanks, Béla. Well explained. John, would you have a question for Béla?
00:30:37:23 - 00:31:04:24
John Kelleher
I believe yes, actually, I'm fascinated by this work. I'm really interested in the aspect previous synthetic data. So my own work, and in healthcare we often use synthetic data to protect people's privacy. But here I see you're using synthetic data because real data will be so dangerous to handle. And I was wondering if the fact that, are the challenges you face in creating a system to synthesize data, you can't handle the real data at all.
00:31:05:01 - 00:31:18:08
John Kelleher
You know, I can understand that at health care, we we can take a sample of patient data and synthesize from that to create, to protect privacy. But I presume for your work, there's even more challenges involved in creating a good synthesis system.
00:31:18:10 - 00:31:47:21
Béla Mihalik
Yes. It's a real challenge, as you said. First of all, we had to calculate a microscopic level of light propagation and calculate holograms. It means we have to calculate interferences. It means, we used different type of mathematical models. One type of mathematical model is the step by step, let's say a step by step FDDD simulation.
00:31:47:21 - 00:32:16:00
Béla Mihalik
And then other type is, analytic. It uses some rules of physics which are haps, to handle image planes, as one, data information and propagate the plane by plane. So we use different mathematical methods and compare them together. This was a challenge to, to build a model and also to validate it with different simulation and with the reality.
00:32:16:02 - 00:32:20:02
Béla Mihalik
So this was the highest change, challenge, I think.
00:32:20:04 - 00:32:44:10
Abigail Acton
You see, it takes somebody using AI for their own work to spot that. That might have been a a difficult thing to achieve. Thank you. John. Yes, a very pertinent question. And thank you to you, Béla. Excellent. Well, thank you very much to all of you. This has been really fascinating. And it's so nice to hear about the application of AI in a way that feels, that it contributes to our well-being rather than some sort of scary thing coming down the line that's going to take all our jobs.
00:32:44:10 - 00:32:49:06
Abigail Acton
So thank you very much, all of you, for being with me today. It was very, very interesting.
00:32:49:08 - 00:32:53:06
John Kelleher
Goodbye, Abigail. Béla, Jane, it's been a pleasure talking to you all. I've really enjoyed it.
00:32:53:10 - 00:32:56:05
Jane Ohlmeyer
It's been an absolute pleasure. Thank you for having me.
00:32:56:07 - 00:32:57:17
Béla Mihalik
Thank you, thank you, thank you.
00:32:57:18 - 00:33:18:01
Abigail Acton
Thanks very much for joining us. Bye bye. If you've enjoyed this podcast, follow us on Spotify and Apple Podcasts or wherever you get your podcast, and check out the homepage on the Cordis website. Subscribe to make sure that the hottest research and EU funded science isn't passing you by. And if you're enjoying listening, spread the word.
00:33:18:03 - 00:33:38:03
Abigail Acton
We've talked about the world's largest collection of viruses, using VR in conflict resolution, and where the crickets are going to be replacing the beef in your burger any time soon. In our last 51 episodes, there'll be something there to tweak your curiosity. So come and check out the research that's revealing what makes our world tick. We're always happy to hear from you.
00:33:38:07 - 00:33:47:20
Abigail Acton
Drop us a line editorial@cordis.europa.eu. Until next time.
This is an AI transcription.
00:00:00:00 - 00:00:39:04
Abigail Acton
This is CORDIScovery. Hello and welcome to this episode of CORDIScovery with me, Abigail Acton. Artificial intelligence. Are we heading for a dystopic future, or one in which labor force is relieved from menial tasks? Machine learning is second guessing us. Is that making our lives easier? Or is it intrusive? Corporates are hoovering up content to train large language models in ways that are hard for users to control.
00:00:39:08 - 00:01:04:12
Abigail Acton
It's clear AI is huge, but good, huge or bad huge? That seems to be up for debate. There's no doubt that AI and related tools are pushing back the boundaries of research. Today, we do have some clarity, at least when it comes to our three projects, all of whom have received EU research and innovation funding. Our three guests are using these tools to develop new technologies and make discoveries that would have previously been impossible.
00:01:04:14 - 00:01:29:13
Abigail Acton
We're looking at the way in which AI is pushing back the boundaries of knowledge to improve the prevention, treatment and rehabilitation of stroke patients. To give a voice to the previously unheard women of historical Ireland, and to develop a small, powerful bio threat detection system to save lives. John Kelleher is the director of the Adapt Research Ireland Center for AI driven Digital content Technology.
00:01:29:15 - 00:01:40:08
Abigail Acton
And professor of computer science at Trinity College Dublin. He focuses on harnessing AI to enhance the understanding and treatment of complex medical conditions. Hello, John. Welcome.
00:01:40:11 - 00:01:44:20
John Kelleher
Hello, Abigail. I'm delighted to be here and very much looking forward to our discussion.
00:01:44:22 - 00:02:01:09
Abigail Acton
Jane Ohlmeyer professor of modern history at Trinity College Dublin and former chair of the Irish Research Council, is an expert on new British and Atlantic histories. She is applying AI to the recovery of the lived experience of ordinary non-elite women in early modern Ireland. Hello, Jane.
00:02:01:11 - 00:02:02:23
Jane Ohlmeyer
Hello. Lovely to be here.
00:02:03:00 - 00:02:09:01
Abigail Acton
Lovely to have you. Béla Mihalik is a senior developer at Ideas Science in Hungary.
00:02:09:03 - 00:02:19:12
Abigail Acton
He specializes in the application of AI and deep learning to develop novel tools that can rapidly screen potential bio threats in the form of pathogens and bacteria. Hi Béla.
00:02:19:14 - 00:02:21:00
Béla Mihalik
Hello. Good to be here.
00:02:21:06 - 00:02:39:13
Abigail Acton
Good to have you. Thank you. John I'm going to turn to you first. The STRATIF-AI project builds on the current use of machine learning applied to stroke patients, to make the response far more customized and time sensitive. John. Healthcare is one of the areas that is most benefiting from AI. Can you tell us a little bit about the goals of the STRATIF-AI project?
00:02:39:15 - 00:03:07:03
John Kelleher
Yes. Abigail, STRATIF-AI Project is interested and focused on integrating heterogeneous data sets to drive insights for patients across their lifetime and with a particular focus on how stroke affects them. So we are trying to make continuous stratification, and that's where the name of the project comes from. We're trying to create AI tools that continuously stratify somebody's stroke risk throughout their lifetime and help them to manage that stroke risk and also have better outcomes after stroke treatment.
00:03:07:05 - 00:03:16:06
Abigail Acton
Okay. Super. How did you identify the patients that you would follow and how did you enroll your participants in this?
00:03:16:08 - 00:03:40:01
John Kelleher
Our ultimate goal would be to have broad coverage across a population rather than, a particular cohort. We want to be able to take for the whole population to develop tools that are suitable for everybody that would be able to take somebody through from, from a very young through their whole life through to further had a stroke under in acute care versus reintegration back into society after that care.
00:03:40:01 - 00:03:45:00
John Kelleher
So the the ultimate ambition is for everybody to be supported and enabled through this research.
00:03:45:02 - 00:03:53:01
Abigail Acton
Okay, fabulous. That sounds really, a really good goal. How are you applying artificial intelligence to this? What is what is artificial intelligence enabling you to do?
00:03:53:03 - 00:04:17:06
John Kelleher
So we're bringing together two different types of artificial intelligence in this in this research, the first type is called the digital twin model. So a digital twin is the idea of we create a computational model that is personalized to an individual. Creating a digital twin, makes the model quite interpretable. Because of the structure of the model. We're able to understand how the different variables that represent the person's health state evolve through time.
00:04:17:08 - 00:04:56:23
John Kelleher
And but it has some challenges in terms of being able to develop those models to scale to large populations and integrate very different types of data. So we're combining those digital twin technologies with what we call machine learning. So machine learning is a way of analyzing very large data sets to extract patterns and interactions between different features. So by bringing these two types of machine learning and digital twin artificial intelligence approaches together, we're trying to make personalized, interpretable systems that can scale and bring together large heterogeneous data sets so that an individual is able to understand their stroke risk throughout their lifetime and manage it right.
00:04:56:24 - 00:05:02:13
Abigail Acton
Excellent. And then also perhaps to to receive advice that would enable them to reduce their risk.
00:05:02:15 - 00:05:30:00
John Kelleher
Oh exactly. Yeah. So we'd like to be able to, for example, will tailor the interventions that are suggested to a person to their stroke risk and help them to understand across their lifetime how that evolving and how modifications in their lifestyles can affect that. So I think one of the really interesting things that this might link to is that actually, the earlier on, somebody makes a change in their life in terms of a stroke risk, the greater the, the effect it has on the cumulative approach over their life course.
00:05:30:05 - 00:05:58:09
John Kelleher
So even a very small change very early on in their life, for example, in terms of modifying their cholesterol early on in life, can have a massive benefit across their lifespan. And the earlier we intervene in somebody's life course, the cumulative benefit is much, much greater. So you can imagine that if we wait till later on in life to intervene in somebody's health trajectory, we have to do a more intense intervention, for example, using drugs.
00:05:58:11 - 00:06:10:03
John Kelleher
So what we'd really like to do is, early on in life, allow people understand the risk and do small changes to their lifestyles that helped them to avoid bigger interventions later in life.
00:06:10:05 - 00:06:26:22
Abigail Acton
And the difference with this, of course, from from current sort of, approaches, from the health care, system where, you know, you're told to eat your five vegetables and fruit a day and, you know, walk however many steps the current thinking is and so on. The difference here is that your advice would be very much tailored to the individual.
00:06:26:22 - 00:06:35:13
Abigail Acton
So you'd be getting a, a much more textured, in-depth insight into the various things that all interplay together. Is that correct?
00:06:35:15 - 00:06:59:01
John Kelleher
Yes. So one of the fascinating things about health and medicine is that while we can get, we can extract information about risk factors at a population level, we are all individuals and we are all different. And how they interact within our own systems are different. So trying to make models that are personalized and tailored to individual to capture risk factors, is what we're trying to do here.
00:06:59:01 - 00:07:32:24
John Kelleher
So, for example, you might consider that some people, have a genetic predisposition to particular risk factors. And there's aspects called epigenetics where our lifestyles turn on and off genes that cause risk factors or that exacerbate or amplify risk factors or reduce risk factors. So trying to integrate, for example, genetic information with lifestyle, nature and nurture, if you like, and give us a much more personalized risk for an individual than if we just consider somebody as somebody in a population similar to everybody else.
00:07:32:24 - 00:07:47:15
Abigail Acton
And of course, the benefit immediately is that with that personalized input, you're far more motivated to to follow the suggestions because you know, it is for I like to know, for John Kelleher or for whoever it's specifically tailored for you. So I think maybe you take it more seriously.
00:07:47:17 - 00:08:08:19
John Kelleher
Yes, I'm actually one of the really exciting aspects of the project is that our partner in Linköping, who's coordinating the project, being a citizen, they have the idea that we'll take this digital twin, and the nice aspect of the digital twin is we can simulate forward in time your health state. And what we are developing is the concept of a digital twin, where there is a graphical interface, a digital twin.
00:08:08:21 - 00:08:26:10
John Kelleher
And we can not only tell you about how your risk is, but it will evolve through time. But we can also show you graphically to make it much more concrete, to show you, for example, how you would age with this lifestyle versus another lifestyle to try and improve, compliance and interventions to help people and motivate them.
00:08:26:16 - 00:08:39:13
Abigail Acton
Absolutely. And it sounds very motivational. And is this something that would be on someone's telephone app, for example, or would it be like a phone app? You'd just be able to pull it out of your pocket and see the choices that you were making, how they would impact on you in real time. Would that be the ultimate goal?
00:08:39:15 - 00:09:06:12
John Kelleher
So we have two different aspects. Yes. One is for you have a health vault on your phone for you can carry around on your phone your own personalized health, profile and that chronic health record, but also where we'd have a larger graphic display that would be used in an intervention called a health dialog, where you would go to somebody, maybe every ten years or five years, you'd go in and speak to a nurse or a clinician, and we would be able to simulate there and give you a develop a plan with your input.
00:09:06:18 - 00:09:27:09
John Kelleher
Or we could simulate for you a whole body scan, and we're able to work with you and say, well, if you change this and this, if you, for example, reduce your calorie intake, reduce your cholesterol, this is what happens, or that's what happens. And with that insight given to you, have a dialog with you to think about. What's the plan you would like to adopt over the next 5 or 10 years?
00:09:27:10 - 00:09:28:00
Abigail Acton
Fantastic.
00:09:28:02 - 00:09:34:17
John Kelleher
So to have patients and clinicians speaking to each other and co-developing your plans for treatment and support.
00:09:34:17 - 00:09:45:08
Abigail Acton
So collaboration rather than sort of instruction. Fantastic. Do you imagine that digital twins can be used elsewhere in the health care processes? Are there other applications that you can think of as being useful?
00:09:45:10 - 00:10:06:19
John Kelleher
Absolutely. So the the really exciting thing about digital twins is their ability to simulate the evolution of a system. So what that means, for a system here could be the, evolution of a disease within an individual. So, for example, any, any types of health where we have a chronic, long term evolution of that disease within an individual.
00:10:07:00 - 00:10:33:16
John Kelleher
Digital twins can support that. But also if we move away from thinking about precision medicine for an individual, we can also think about, for example, even the logistics of healthcare. So simulating hospital operations and facility management, because the digital twin can simulate either an individual or a physical system. So maybe we can improve the logistics around healthcare, reducing the economic burden of healthcare.
00:10:33:18 - 00:10:51:22
John Kelleher
But there's also the idea of, drug development and medical device simulation. So trying to use these technologies to increase the rapidity in which we can develop new treatments or develop devices and simulate devices and interventions on individuals to check how effective they are before they're deployed.
00:10:52:02 - 00:11:01:02
Abigail Acton
And what about the digital twin of actually a pathogen, so that you could then see the impact of treatment on that pathogen in the form of the digital twin?
00:11:01:04 - 00:11:23:07
John Kelleher
Absolutely. So the challenge with developing digital twins, and one of the things we're really tracking or addressing within the challenge, digital twins have huge capabilities and advantages going forward. The difficulty with them is that in order to implement them, we need a theory, theory or model for how a system, a process, evolves through time, and that can be difficult to implement.
00:11:23:12 - 00:11:43:24
John Kelleher
And that's where the complementarity between the digital twin and machine learning comes together. Because machine learning enables us to model some of the interactions on high level, complex interactions that we might think us have a theoretical knowledge for. And integrating that with the digital twin allows the digital twin to model in an interpretable and comprehend computation feasible manner.
00:11:44:01 - 00:11:52:17
John Kelleher
The areas that we do understand but integrate the patterns that machine learning takes from large data to help the whole system work more comprehensively.
00:11:52:19 - 00:11:59:11
Abigail Acton
Fantastic. It sounds like a wonderful way that AI is pushing back the boundaries of research. Thank you so much, John.
00:11:59:13 - 00:12:21:06
John Kelleher
That's why I think that the integration of the digital twins, the theoretically driven, if you like, approaches versus the data driven machine learning approaches that captured those high level interactions, maybe before we understand them is such a promising approach. This hybrid modeling will allow us to start building systems that integrate both what we understand and what we have.
00:12:21:06 - 00:12:32:23
John Kelleher
what we can model which are different things at some point, integrating them now and then slowly with those initial systems growing the amounts that we understand, to be more interpretable.
00:12:33:00 - 00:12:54:23
Abigail Acton
Fantastic. Thank you John. I'm going to turn to Jane now. The role of women in early modern Ireland was pivotal in a time of societal change, but the stories of the women highlighting their resilience amidst social upheaval and trauma have largely been lost to history. The voices project let them be heard. Horizons are also opening up in the humanities thanks to AI.
00:12:55:02 - 00:13:00:19
Abigail Acton
What was the challenge facing you before such tools became available, and what can you do with them now?
00:13:00:21 - 00:13:22:16
Jane Ohlmeyer
The big challenge was really around access access to the manuscripts themselves and then access to material that was available in digital format but wasn't interoperable. It didn't. These digital silos weren't talking to each other. And it's in those two particular areas that AI really has been very enabling for our work.
00:13:22:16 - 00:13:38:17
Jane Ohlmeyer
And I think that the potential is we are just beginning to explore it. But if we can get it right, AI really will be transformational when it comes to doing historical research, not just about women, but obviously our focus is women.
00:13:38:17 - 00:13:46:05
Abigail Acton
Absolutely I guess. I mean, it seems hard to imagine, even as a journalist before. Before all of these tools became available and so on.
00:13:46:05 - 00:13:53:22
Abigail Acton
But, I mean, how would something like this have been done in the past? Would you even have been able to tackle such a large scale project in the past without such tools?
00:13:53:24 - 00:14:10:12
Jane Ohlmeyer
Well, AI allows us to do this history at scale because it allows us to actually interrogate records, especially, records around taxation or demographic records that it's incredibly time consuming to analyze these or legal records.
00:14:10:18 - 00:14:36:13
Abigail Acton
And what we find is that the women are deeply buried in them. Abigail. So AI is allowing us to identify those women. And once they've been surfaced, if you like. Then obviously, we're analyzing what we're finding. So it really has, has helped us enormously around the identification and access, to non-elite women who, as I say, are very much buried in the records.
00:14:36:15 - 00:14:39:01
Abigail Acton
Wonderful. What sort of things were you finding, Jane?
00:14:39:03 - 00:15:06:06
Jane Ohlmeyer
What we're finding is that actually, women have been hiding in plain sight all along. And that records that we thought were actually of no use, previously, we are finding are of tremendous value. Let me give you an example. We've got things called funeral entries for 17th century Ireland, the equivalent of, death notices today.
00:15:06:08 - 00:15:35:11
Jane Ohlmeyer
And, people had never really bothered with them. And what we're finding is over a third of them actually are by women and about women. So all of a sudden, a whole new body of material has been, if you want, unearthed, thanks to our interactions with AI or Wills testamentary material. Again, we thought that, for a whole variety of reasons, that there were no wills by women or there were very few.
00:15:35:13 - 00:15:53:13
Jane Ohlmeyer
And, we've now uncovered about 500 of them. Again, AI has been extremely helpful in allowing us a) to identify them and then to create digital surrogates that were then able to, analyze and interrogate.
00:15:54:05 - 00:16:09:10
Abigail Acton
I'm just going to cut across here because I've got so many of my this is so interesting. I've got my brain just seething with questions that I want to ask you first and then observation, is it must be a bit like being an archeologist discovering a a hitherto unknown tomb that shines a light on, on the whole kind of culture of that period.
00:16:09:10 - 00:16:12:18
Abigail Acton
And it's absolutely fascinating. Must be very thrilling. Jane is a thrilling?
00:16:12:20 - 00:16:28:04
Jane Ohlmeyer
Oh, it's so exciting, Abigail, I can't tell you because it's like you're going to something, you're not going to find anything. And all of a sudden, you know, it's a third of the records are relevant and interesting, and these are records that we thought many just didn't exist.
00:16:28:06 - 00:16:48:03
Jane Ohlmeyer
Because, in 1922, the archive in Ireland was destroyed. So, I mean, it's one of those ones where, all the material is dispersed, in other words, transcripts that are scattered all over the place that we are, thanks to the technology, are able to reconstitute, I think, in a very important and innovative way.
00:16:48:08 - 00:16:52:03
Abigail Acton
Okay. So you said that the material apparently had been lost.
00:16:52:05 - 00:16:59:09
Abigail Acton
So so where did you find it? The scattered, scattered transcripts of duplicate copies found in places, hitherto not looked at.
00:16:59:09 - 00:17:20:14
Jane Ohlmeyer
There are some wills in the National Archives in Dublin, that people really hadn't appreciated had survived the fire. So that's obvious. Right. But what we're also finding is transcripts that are in private collections or in publications, that were published before 1922, which is when the big fire took place.
00:17:20:16 - 00:17:32:03
Jane Ohlmeyer
So they're in hundreds of different locations and they're it's just they're so disparate. And this is a way of taking all these digital windfalls and bringing them together.
00:17:32:05 - 00:17:44:19
Abigail Acton
I love digital windfalls. And you. Well, it must be because wills are such a window onto a society in a culture, for example, obviously immediately comes to mind that women had property, to, to bequeath and, and all sorts of things.
00:17:44:19 - 00:17:48:00
Abigail Acton
Tell me some of the insights you're gaining from looking at such documents.
00:17:48:00 - 00:18:15:13
Jane Ohlmeyer
Well, on the wills front, it tells us a lot about material culture and the property that the jewelry, the household goods. So you've got that side of it, but also it tells you about who women hold most dear, their intimate networks, and how women tend to bequeath particularly moveable property, as well as sometimes property at land and, and, and fixed assets to other women.
00:18:15:15 - 00:18:36:16
Jane Ohlmeyer
So we're just really beginning to recover all of this, and you have to remember, in the 17th century, a woman didn't have her own legal identity unless she was a widow. So, and a married woman could only make, well if her husband allowed her to do that. So, again, it's giving us insights that we didn't think were possible to find.
00:18:36:18 - 00:18:37:18
Jane Ohlmeyer
Yeah. It's been great.
00:18:37:23 - 00:18:52:00
Abigail Acton
Yeah, I know it sounds. It sounds thrilling. Have you any, stories or examples that you've dug out that surprised you or that you feel well, maybe I can use the old cliche, a bit of a eureka moment about.
00:18:52:02 - 00:18:57:06
Jane Ohlmeyer
We're having lots of eureka moments. It’s really exciting.
00:18:57:06 - 00:19:24:11
Jane Ohlmeyer
And we're just we're only two years into the project, Abigail. So when we speak in a couple of years time, you know, so my colleague, who's doing a lot of work on the Irish Court of Chancery, which is the Court of Equity, where large numbers of women are plaintiffs and defendants. And again, it's so fascinating to see how during usually moments of personal crisis, these women are coming out of the shadows and they're going to court often to protect their own interests.
00:19:24:11 - 00:19:50:08
Jane Ohlmeyer
It might be that they've been given some, money upon marriage, and a male relative is trying to steal their money or steal their land or steal their property, or they're going to court to protect the interests of their children, or, their reputation. We have one very, very juicy account of a young woman who has entered into a relationship with a man.
00:19:50:10 - 00:20:12:11
Jane Ohlmeyer
And she's had a child out of wedlock, and she then marries. And the husband she marries is actually taking her former lover to court for damaging the reputation of his wife. So, again, these are the sorts of stories that are usually lost from history. Yes. And it gives you, as I say, a very personal insight into people's lives.
00:20:12:13 - 00:20:20:10
Abigail Acton
Absolutely wonderful. Well, thank you so much for explaining all that to us and setting it out so clearly. Does anyone have any questions for Jane?
00:20:20:12 - 00:20:42:06
John Kelleher
Actually, I will would, Jane. I'm fascinated by the VOICES project, and I'm really excited, by the way, you're using new technologies to try and understand the past in new ways, particularly technology that's evolving very rapidly. So I'd really like you to, if you could, to think about or imagine how these AI systems are evolving. It could be designed in the future to improve and support historical research.
00:20:42:06 - 00:20:47:23
John Kelleher
And what capabilities do you hope it will have, and how do you think it will enable historic research in the future?
00:20:48:00 - 00:21:17:18
Jane Ohlmeyer
Oh great question, John. I want a AI system that is transparent. I want the computer scientists, never mind the historians, to understand the algorithms that underlie it. I want an AI system that is ethically, bounded, and, that code of ethics, then protects the intellectual property of everybody who is contributing, to it.
00:21:17:20 - 00:21:48:07
Jane Ohlmeyer
But the third thing I want is an AI system that is environmentally responsible. I'm so conscious that the current system is so greedy of power and water and especially generative AI. So as we look to the future, those ethical and environmental issues for me are hugely important. And then I'll feel much more comfortable about being able to trust, the AI system that we come up with.
00:21:48:07 - 00:21:54:21
Jane Ohlmeyer
Because just at the moment, trust is my fundamental objection, or concern with AI.
00:21:54:23 - 00:22:12:15
Abigail Acton
Yes. I think we can all absolutely identify with that. Perfect. Thank you. Béla, I'm turning to you now. Terrorism can take many forms, including biological ones. Bioweapons pose a threat mainly due to detection difficulties. But that's something the HoloZcan project is tackling head on.
00:22:12:18 - 00:22:21:05
Abigail Acton
So the project's been using AI to develop equipment to help first responders in an emergency. What problem were you hoping to solve Béla?
00:22:21:07 - 00:22:54:06
Béla Mihalik
Yes. The problem, is mainly discovered by the ENCIRCLE project of the European Union, ENCIRCLE project defined the gaps, and, in the gaps they identified, there is no good onfield tool for biological detection. And we wanted to, to, to create a device which can work in on field and also possible to used in many scenarios, not just in one scenario.
00:22:54:06 - 00:23:13:12
Béla Mihalik
So it's a big challenge, of course. It's, it's over the state of the art. But, we started, to think about how we can apply microscopy, especially holographic microscopy technology, to fill a gap a bit, even if we not solve the whole problem.
00:23:13:14 - 00:23:22:11
Abigail Acton
Oh, geez. You made a great start on this on this goal. Béla what are the current ways of detecting what might be present in a biological attack?
00:23:22:13 - 00:23:50:18
Béla Mihalik
Currently, the most accepted way to detect is use PCR methods. And also, the fluorescent labeling methods, DNA sequencing, all of them needs laboratory preparation. It's a long, long method. And, for example, PCR software from false positive results. It means, it can signal, it can see it's a danger even if there is not.
00:23:50:18 - 00:24:20:20
Béla Mihalik
And it's the threat is very high above 20%, 30% for current, detectable pathogens. And in second, seeing the problem of epigenetic segments is, is currently not handled. It means, dangerous fragments can be detected from DNA, but it is not present, on, protein level. It means it's not really that harmful.
00:24:20:22 - 00:24:29:08
Abigail Acton
What you mean is that the current systems could pick up fragments of DNA? Yeah, that are not accurate in some way. What's the problem with the picking up of the fragments?
00:24:29:10 - 00:25:01:24
Béla Mihalik
The problem is that, we cannot find dangerous fragments which can coding dangerous proteins, but because of epigenetic information, the cells are not expressing, not creating these dangerous, proteins. And it's everywhere around us in the environment, this code is everywhere, but it's not expressed because of epigenetic blocks. So this is why we cannot use, sequencing in each case I see.
00:25:01:24 - 00:25:21:02
Abigail Acton
So the sequencing might be picking up something that's around us anyway. And not posing a problem. Yes, but it's identifying it and then it's confusing the end result because then the responders don't know what they're looking at specifically. Yes, I see what you mean. Okay. So how did HoloZcan aim to help all of this? And how did it use AI?
00:25:21:04 - 00:25:55:18
Béla Mihalik
In HoloZcan the focus on microscopic technology or holographic microscopic technology via holography, and it's a good question because, it's, let's say self-calibratable. It's set of calibrated, self-calibrated pictures. It means, we can numerically analyze the picture so we can get values from it under different conditions. And holography gives a very stable base for this numerical analyzes.
00:25:55:20 - 00:26:30:14
Béla Mihalik
And so we can see exact morphological properties and other optic, properties like, spectral information. And also refractive information, of the small objects and wide range of information, we can even not identify but classify, the particles. And, also we can see the complex morphologies around, in the sample and around, the, the objects.
00:26:30:18 - 00:26:52:00
Béla Mihalik
And from this, morphological complexity, with using of artificial intelligence, if we can figure out what the situation is, is it intentional? It is a manmade situation, or it's some environment or situation. So AI can help a lot, to identify the the real source.
00:26:52:02 - 00:27:21:22
Abigail Acton
Okay. Excellent. So instead of taking little tiny samples to try and identify what the material is, you've developed a system that actually is visualizing, okay, the material and then through the visualization what's happening then is AI somehow cross-referencing what the thing looks like, what the what the sample looks like or the elements within the sample look like. And then is that cross-referencing across another kind of database of similar images and deciding what it is or what happens next?
00:27:21:22 - 00:27:22:07
Abigail Acton
In fact.
00:27:22:11 - 00:27:52:06
Béla Mihalik
Yes, it started in many lines. First, of course, we use the laboratory, samples, very clean, samples, of bacteria and other pathogens. And, we created, database, and also trained, artificial intelligence to identify these different type of bacteria. And it works very well on laboratory samples. We can see it works more than 99% of accuracy.
00:27:52:10 - 00:28:01:01
Abigail Acton
I think you said something about a generation of 70,000 synthetic holograms. Yeah, that reflect real biological particles. 70,000. That's huge.
00:28:01:03 - 00:28:27:16
Béla Mihalik
Yes. We created a simulator tool because, in laboratory, we we couldn't to achieve all the samples. There are some very dangerous BSL four level, BSL four level that means, it is deadly. And we cannot cure, so we we are not allowed to, to handle this type of material. So we created simulated, samples how this looks like.
00:28:27:18 - 00:28:48:18
Béla Mihalik
And, also we could create the type of, objects, but we can never measure, but in a certain time period. But it can appear in reality. So we we use single simulation or this digital twins, let's say, the extended, the training, range of the, of the artificial intelligence.
00:28:48:23 - 00:28:52:18
Abigail Acton
So it's basically synthetic holograms like a reference library.
00:28:52:20 - 00:28:53:23
Béla Mihalik
Yes. Absolutely.
00:28:54:03 - 00:29:09:06
Abigail Acton
Lovely. So then a first responder takes this device to a situation where they suspect that there's been some sort of bio threat, and what, they take a sample of the air or the sample of the soil, they put it into the canister. Or how does that work?
00:29:09:08 - 00:29:20:18
Béla Mihalik
Actually focused only for airborne pathogens. So we use air sample collectors, Coriolis based air sample collectors and also impactors.
00:29:20:18 - 00:29:22:04
Abigail Acton
So what's an impactor?
00:29:22:06 - 00:29:45:10
Jane Ohlmeyer
An impactor is a device, when we create that air flow with ventilators and we put obstacle, in the line of airflow, and on this, obstacle, the the small objects will stick, and they pick it up for if it's, microscopic slide, and we directly can put it under the microscope and analyze.
00:29:45:14 - 00:29:48:03
Abigail Acton
And that can be done in situ in real time.
00:29:48:07 - 00:29:57:00
Béla Mihalik
Yes, it can be done in real in situation and in real time. And also we can integrate together an impactor and the microscope.
00:29:57:04 - 00:30:07:12
Abigail Acton
Right. So basically you're able to collect data in the field that will then allow first responders to see immediately what it is that has has contaminated the area. Is that correct?
00:30:07:14 - 00:30:24:14
Béla Mihalik
Yes. Is that correct? And, we can analyze the composition of air, which is the most important information for the first responders. And with this composition, we can infer, is it a real threat? It can be a real threat or it's some environmental variation.
00:30:24:16 - 00:30:37:21
Abigail Acton
Okay, but that's fantastic. So it rules out all these, false positives, which may be caused chaos. And in fact, there's nothing dangerous there. And and so on. Super. That's excellent. Thank you very much. Thanks, Béla. Well explained. John, would you have a question for Béla?
00:30:37:23 - 00:31:04:24
John Kelleher
I believe yes, actually, I'm fascinated by this work. I'm really interested in the aspect previous synthetic data. So my own work, and in healthcare we often use synthetic data to protect people's privacy. But here I see you're using synthetic data because real data will be so dangerous to handle. And I was wondering if the fact that, are the challenges you face in creating a system to synthesize data, you can't handle the real data at all.
00:31:05:01 - 00:31:18:08
John Kelleher
You know, I can understand that at health care, we we can take a sample of patient data and synthesize from that to create, to protect privacy. But I presume for your work, there's even more challenges involved in creating a good synthesis system.
00:31:18:10 - 00:31:47:21
Béla Mihalik
Yes. It's a real challenge, as you said. First of all, we had to calculate a microscopic level of light propagation and calculate holograms. It means we have to calculate interferences. It means, we used different type of mathematical models. One type of mathematical model is the step by step, let's say a step by step FDDD simulation.
00:31:47:21 - 00:32:16:00
Béla Mihalik
And then other type is, analytic. It uses some rules of physics which are haps, to handle image planes, as one, data information and propagate the plane by plane. So we use different mathematical methods and compare them together. This was a challenge to, to build a model and also to validate it with different simulation and with the reality.
00:32:16:02 - 00:32:20:02
Béla Mihalik
So this was the highest change, challenge, I think.
00:32:20:04 - 00:32:44:10
Abigail Acton
You see, it takes somebody using AI for their own work to spot that. That might have been a a difficult thing to achieve. Thank you. John. Yes, a very pertinent question. And thank you to you, Béla. Excellent. Well, thank you very much to all of you. This has been really fascinating. And it's so nice to hear about the application of AI in a way that feels, that it contributes to our well-being rather than some sort of scary thing coming down the line that's going to take all our jobs.
00:32:44:10 - 00:32:49:06
Abigail Acton
So thank you very much, all of you, for being with me today. It was very, very interesting.
00:32:49:08 - 00:32:53:06
John Kelleher
Goodbye, Abigail. Béla, Jane, it's been a pleasure talking to you all. I've really enjoyed it.
00:32:53:10 - 00:32:56:05
Jane Ohlmeyer
It's been an absolute pleasure. Thank you for having me.
00:32:56:07 - 00:32:57:17
Béla Mihalik
Thank you, thank you, thank you.
00:32:57:18 - 00:33:18:01
Abigail Acton
Thanks very much for joining us. Bye bye. If you've enjoyed this podcast, follow us on Spotify and Apple Podcasts or wherever you get your podcast, and check out the homepage on the Cordis website. Subscribe to make sure that the hottest research and EU funded science isn't passing you by. And if you're enjoying listening, spread the word.
00:33:18:03 - 00:33:38:03
Abigail Acton
We've talked about the world's largest collection of viruses, using VR in conflict resolution, and where the crickets are going to be replacing the beef in your burger any time soon. In our last 51 episodes, there'll be something there to tweak your curiosity. So come and check out the research that's revealing what makes our world tick. We're always happy to hear from you.
00:33:38:07 - 00:33:47:20
Abigail Acton
Drop us a line editorial@cordis.europa.eu. Until next time.
AI is huge, but ‘good’ huge or ‘bad’ huge seems to be up for debate
Artificial intelligence: are we heading for a dystopic future, or one in which a labour force is relieved from menial tasks? Machine learning is second-guessing us; is this making our lives easier, or is it intrusive? Corporates are hoovering up content to train large language models in ways that are hard for users to control. But there is no doubt that AI, and related tools, are pushing back the boundaries of research. Today we do have some clarity, at least when it comes to our three projects, all of which have received EU research and innovation funding. Our three guests are using these tools to develop new technologies and make discoveries that would previously have been impossible. We’re looking at the way in which AI is pushing back the boundaries of knowledge to improve the prevention, treatment and rehabilitation of stroke patients, to give a voice to the previously unheard women of historical Ireland and to develop a small, powerful biothreat detection system to save lives. John Kelleher(opens in new window) is the director of the ADAPT Research Ireland Centre for AI-Driven Digital Content Technology and professor of Computer Science at Trinity College Dublin. He focusses on harnessing AI to enhance the understanding and treatment of complex medical conditions. John coordinated the STRATIF-AI project. Jane Ohlmeyer(opens in new window), professor of Modern History at Trinity College Dublin and chair of the Irish Research Council, is an expert on New British and Atlantic Histories. She is applying AI to the recovery of the lived experiences of ‘ordinary’, non-elite women in early modern Ireland which she explored in the VOICES project. Béla Mihalik(opens in new window) is a senior developer at Ideas Science(opens in new window), in Hungary. He specialises in the application of AI and deep learning to develop novel tools that can rapidly screen potential biothreats in the form of pathogens and bacteria. HoloZcan has developed technology to help first responders.
Happy to hear from you!
If you have any feedback, we’re always happy to hear from you! Send us any comments, questions or suggestions to: editorial@cordis.europa.eu
Countries
Hungary, Ireland, Sweden