The Role of Leadership in the Advancement of AI – Dr. Dimitrios Kalogeropoulos and Dalith Steiger Interview - Thinking Beyond Necessity
To have an accurate picture, allow me to set some context first. AI has been around and supporting the US health industry, and other industries, for more than 30 years, and the underlying inference techniques are pretty much the same. For example, I developed a scaling environment for integration into routine clinical practice of a Bayesian digital endotyping system for ICUs some 25 years ago.
What has changed however, are the fundamentals in the field, in terms of the exponential increase of computational power and the democratization of this power; as well as in terms of the applications of AI which are now imaginable as a result.
What has been achieved is we can now dream in more detail – because AI is a knowledge activity and because the amount of knowledge AI can handle, has increased together with the speed with which AI can elicit, model, use and share new knowledge of the world.
New knowledge is manifested in the transition from empirical reasoning, or clinical correspondence reasoningand expert systems, toward learning AI, clinical coherence and systemsorprecision medicine, rendering an ability for sensing, detecting, predicting and translating disease trajectories and genomics.
This acceleration in knowledge-processing over the past few years, is impressive, and has started a revolution which spans the entire health industry. It is in this acceleration that the pharma industry sourced all its might to deliver Covid vaccines with record breaking speed and unprecedented precision. And this is also the springboard on which leading innovators have been fundamentally shifting entire industries and their traditional boundaries.
But since you did mention global view, it is important to keep in mind that the speed with which knowledge is produced varies widely, even across the street. This originates in, leads to, and exacerbates information asymmetries, which further humper progress in scaling and spreading digital and health innovation in an equitable and sustainable manner. Genomic sequencing is much cheaper and so are AI services. But the data used are not and this is a problem.
With the health industry being a care-standards industry, this asymmetry means adoption and delivery are not up to speed. Global innovators with access to knowledge, push for more adoption – local systems are not ready. In fact, the whole industry is currently in disarray.
Let me give you an example of this thing called the innovator’s predicament. To deliver equity as an ethical principle in AI we must know what equity looks like. But how can we, if the data we draw upon to elicit the knowledge is not inclusive? A recent review of RWE studies showed the largest cohorts today are 30% inclusive at best. This means our research and innovation are not representative of our equitable care goal and thus drive further underrepresentation in equity knowledge models and misrepresentation in model development – to bias and error. Moreover, this is a very carbon intensive, resource demanding, and expensive vicious circle.
In short, the demand for more and better AI and innovation cannot be met by supply, as that acceleration runs way beyond the pace and current capacity of the innovation ecosystems to evolve.
It feels as if someone gave us lots of money and we don’t know how to spend it.
So, what do we do? – We are discussing about the ethics of AI in healthcare. We are regulating the spending, and as a result prescriptive instead of descriptive regulation is being introduced on both sides of the Atlantic, which, amidst the disarray, is both celebrated and vilified, for attempting to decelerate progress until we can better prepare our health systems.
My preferred take home message on the current state in the industry is this. Massive investments are made in AI-based health innovation at the moment, presenting global health policy makers with a unique opportunity to capitalize on those investments, to get the AI into the value production process, where it belongs, and to make AI work for the economy instead of the other way around.
Look. Right now, according to global-reach institutions such as the Stanford Institute for Human-Centered AI and the Council of Europe, AI deployment in care remains nascent, despite the explosion in reported R&D activity, and investment. In plain words, AI is still very much an experiment. This, because AI, medical knowledge, and health innovation are evolving in an environment which is starved of oxygen. They are starved of good data.
What we should be doing, instead of attempting to regulate innovation, would therefore be to we to accelerate it even further, to remove asymmetry and solve the data problem. Put fire out with fire.
But for this we need to enable innovation ecosystems that foster collaboration. We also need to save for margins to go around for everyone. We need to establish circular economies in health innovation through a new data industry in order to make innovation sustainable. This either means further consolidation, which no one wants, or connecting innovation and recycling data with new policies and standards.
This is where AI has its big opportunity to shine. To truly reemerge, by assisting with the data problem and to become a tool for the new data industry – to help create, capture and recycle the value in data. Of course, we need to help a little too, but let’s look at what AI can do better.
A very promising domain is federated learning, where instead of moving the data around, AI reaches out to connect to care data, in order to build critical mass and to build, test and evaluate reliable ML models.
This is better than building a cement factory every time a building goes up – something the industry does with data farms, a practice which leads to a staggering 80% waste and carbon emissions. Then there is synthetic data. Great use of the tech, but why not use the real thing?
Federated AI can also benefit a lot from semantic guidance. Blockchain concepts are very effective for that.
Another thing AI is very good at, is knowledge discovery and validation – by fitting our phenotypic placeholders with endotyping knowledge using ML to look for patterns in existing big data sources and EHRs.
Once these data asset classes are ready, they can be recycled to add new improved uses, more knowledge and more value by supporting further health innovation and clinical decisions along the continuity-of-care journey. This is the great promise of AI – to support a circular economy in healthcare and to provide the foundations for reducing carbon emission practices; not cause them.
Then, once we have successfully delivered primary uses of AI for better data, we can move on to better address secondary uses of AI, with which we are currently evidently unsuccessful – in scaling innovative uses toward digital clinics and self-driving telehealth.
What we are now doing is backward.
Same as with the GDPR, which the EU is now attempting to make productive with the European Health Data-Space Regulation. And a little before that, with the FAIR principles for data and open science.
We basically keep coming at it the wrong way.
We have already established scaling is an issue, as much for AI as for any digital-health-powered innovation which is considered a medical device.
Nonetheless, we are seeing acceleration in AI for the workforce of the future – supporting healthcare staff with clinical administrative tasks which cause fatigue and produce resource over or under-utilization. To deliver this kind of value, innovators must reimagine data-capture together with the care intervention and new business models. And that helps a lot. We are seeing some great startups at the Edison Accelerator in this cohort.
Translational research with AI-assisted biomarker discovery will be a big thing too.
As far as challenges go – we have no scaling policies in place to enable collaborative ecosystems, and no normative frameworks nor industry standards, to enable those policies and to connect.
To illustrate this, imagine the world without the IEEE 802.11 Wi-Fi Standard. Which brings us to industry standards. On this front, we are literally staring at a black hole sucking data into oblivion, somewhere between the popular FHIR standard for interoperable messaging and clinical identifier systems such as ICD-11.
From the policy perspective, successful normative frameworks shall surface once we regulate against information asymmetry, introducing concepts such as data asset liquidity and value-recycling on top of portability (or the GDPR) and interoperability, or information sharing. Right now, we are still struggling to define the term interoperability.
The EHR was meant to deliver that, together with a scaling environment for AI; but it failed miserably, after gobbling down a few tens or hundreds of billions.
Then there is best practice. Some twenty years ago we were concerned with scaling up health service innovations. WHOs ExpandNet offered practical guidance for this. And while innovation is still about raising the standard of care, leaps are now necessary. And with inequitable access to data and the mounting cost of data, we keep fueling information asymmetry and struggling to resist an unintended degrowth spiral. At least this is how it feels.
Policies such as regulatory sandboxes can, and do help with that, but are nowhere near enough. Scaling AI shall continue being an issue until information sharing itself becomes a standard of care, until we start recycling data to eliminate information asymmetries, to enable collaborative ecosystems, and until we reach an ability to deliver multi-stakeholder shared innovation goals; until we establish circular economies in health innovation and collective intelligence.
And for this we need to change our approach to investment – stop pursuing point returns and value creation and enable public-social-private partnerships that engage in co-producing increasing returns to scale and scope with value recycling.
Our strategic, policy and design-thinking need change too. We must adopt a systems investment approach to portfolio management to connect the innovation dots and to avoid waste, to demonstrate the value of recycling, and to accelerate adoption and innovation ahead of the current choking pace. Catch up we must.
Another thing about scaling innovation in health is that margins are diminishing, and value capture is hard to come by, let alone shared equitably. Without access to data as assets – moreover assets which can be recycled, little value is available in the ecosystem to go around. This is the innovation oxygen starvation predicament. We need innovation for value-margins, but this relies on existing value-margins.
If I were to make a prediction, I believe, we are about to enter a long and complex trough of disillusionment for innovation scaling, during which superstructure development will take place to tip the scales to a focused, pre-emptive & pro-active market strategy with the public sector as a partner.
Telehealth-delivered AI will play an important role in such a future. To streamline connected remote-care channels, and to thus funnel data into longitudinal structures which are ideally designed to fill the gaps in the care continuum.
The goal is to keep listening into disease trajectories and patient pathways assisted by AI, until value is captured from patient outcomes.
Leaders must shake the habit to predict the future, and instead design the future, leading from the future. And this requires a new breed of leader. A generation of leaders who will commit to this global change, to step out of the comfort zone of ex-post reporting and disconnected, empirical decisions, and to embrace AI for a sustainable, learning, agile and anticipatory health innovation ecosystem.
New leaders must embrace AI-thinking, to deliver new efficiencies with AI. And there, lies the irony of the whole thing. The fact that we shall serve AI before it serves us. I believe it is this very idea that scares regulators.
Regarding policy for change, there is good news.
The UN Coalition for Digital Environmental Sustainability (CODES) just launched an action plan calling for innovations that catalyze the transition to a green, digital and just economy for sustainable development.
The WHO followed the UK’s lead while presiding COP26 and just launched an Alliance on Transformative Action on Climate and Health,also adopting new models of greener care.
Same time, the EU with its Twinning Green and Digital Transitions strategic foresight report,calls for the adoption of circular economy models in all sectors.
How does that sound for leading from the future?
As far as an agenda goes, I could propose a couple of items:
Firstly, we must recognize existing global data governance proposals and project lessons and adopt them. Also,
Finally – agree to strengthen the role of medical liability frameworks in data governance.
Albeit current EU and U.S. regulation, serves well the established device safety regime, when it comes to AI as a medical device, clinical robustness analyses are a persistent no-go. This because emphasis must shift to medical liability and the provider’s responsibility, in fair measure. And this again is about leadership, governance, and the data production factor as a regulated spin-off market.
I think that’s it for today. Thank you once again for the invitation. It has been a pleasure and a privilege to be here with you today.
Dr. Dimitrios Kalogeropoulos, Independent Adviser, Global Health Innovation Expert: World Health Organization Digital Health and Innovation World Bank, Health, Nutrition and Population Global Practice UNICEF Digital Health Centre of Excellence European Commission, Health and Social Sector GE Healthcare, Wayra UK Edison Accelerator Judge and Mentor Health Executive in Residence at the UCL Global Business School for Health IEEE SA HLS Global Practice, Transforming the Telehealth Paradigm Programme Chair, IEEE SA Data Innovation Ecosystem, Healthcare & Life Sciences Global Practice
Dr Dimitrios Kalogeropoulos is an independent global health innovation adviser, industry leader and expert with the World Bank Health, Nutrition and Population Global Practice, WHO’s Digital Health and Innovation, the European Commission, and UNICEF’s Digital Health Centre of Excellence. He has a PhD in Artificial Intelligence in Medicine, and a track record of thirty years, including twenty years in international development, specializing in digital medical device evaluation and regulation, the development of national and global health data spaces and data recycling, including policy and regulation, and in systems investment strategy for the delivery of value-driven care, sustainable development projects and the enablement of connected innovation to accelerate transformational growth and the adoption of circular economy practices.
Dimitrios is ex-IBM and has served in the board of start-ups in Europe, Asia and the US, is judge and mentor in the Edison™ Accelerator global innovation ecosystem powered by GE Healthcare and Wayra UK and Health Executive in Residence at the UCL Global Business School for Health. He is the founder and Chair of the IEEE Standards Association Healthcare and Life Sciences Practice, Telehealth System Data Asset Recycling for Value-based Care incubator, a collaborative data innovation ecosystem for sustainable global health innovation, founding officer and chapter organizer of the IEEE Engineering in Medicine and Biology Society Greece Chapter, of which he also served as Chair, an adviser in the IEEE SA Healthcare and Life Sciences Practice, Transform the Telehealth Paradigm Programme, and member of the IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems.
In partnership with the World Bank, the WHO and the European Commission, Dimitrios has had an instrumental role in designing and building collaborative innovation and data recycling ecosystems through standards, regulation and policy in several countries, from East and Central Asia to Central America, and in designing and delivering Greece’s resilience and recovery capacity during the Covid-19 global health emergency. Further through his participation in Greece’s national policy think-tank working group on AI, Dimitrios shaped the country’s contribution to the EU AI Act and the establishment of national innovation hubs for AI. His policy landmark achievement has been his contribution to the World Bank and WHO China Joint Study Partnership to develop a policy and agenda for Deepening Health Reform in China: Building High-Quality and Value-Based Service Delivery, now implemented as Healthy China 2030.