WEF - The Future of AI-Enabled Health 2025

Host 1: Okay, just picture this for a second. AI in healthcare.

Host 2: Mhm.

Host 1: Not some, you know, far-off sci-fi thing, but actually here, making a real difference.

Host 2: Mhm.

Host 1: Imagine it helping doctors spot things earlier, making those complicated hospital systems run smoother, or even giving you personalized health advice.

Host 2: And the potential scale of that is just it's huge, isn't it? Healthcare impacts everyone, and AI could genuinely change the game from discovering new drugs right down to how you talk to your GP.

Host 1: Exactly. And that's what we're really digging into today. We've been reading this uh really fascinating white paper from the World Economic Forum. It's called The Future of AI-Enabled Health: Leading the Way.

Host 2: Right.

Host 1: Think of it like their big picture look at how AI innovation could shake up the whole health sector. They're looking at the opportunities, the uh the challenges, the strategy. It's part of a bigger series they're doing on AI across different industries.

Host 2: Precisely. And our goal here, for you listening, is to sort of cut through all the hype you hear about AI in healthcare.

Host 1: Yeah, there's a lot of it.

Host 2: There really is. We want to get to the core insights. Like what are the real ways AI might actually affect your health down the line?

Host 1: That's the million-dollar question, right? Because the promise feels enormous, but the paper makes this really crucial point. Actually getting AI adopted widely in healthcare, it's happening, well, pretty slowly.

Host 2: It really is.

Host 1: So, we're going to unpack why that is, and maybe more importantly, what needs to change to get things moving faster.

Host 2: Uh-huh.

Host 1: Okay, let's get into it.

Host 2: So, first let's just grasp the sheer size of this landscape. Healthcare is massive, so many different players. Uh figure one in the paper really maps out AI's potential reach. You've got pharma companies, hospitals and clinics, insurance payers, medtech firms, even public health agencies.

Host 1: Right. It's not just like robots doing surgery, although that's part of it.

Host 2: Mhm.

Host 1: Take the drug companies, the pharmaceutical firms. AI is already being used to seriously speed up drug discovery. The paper mentions uh Atomwise and Exscientia doing cool stuff there.

Host 2: Yeah, they can sift through huge amounts of data way faster.

Host 1: Exactly. Find potential drug candidates much quicker. Then you have others like Owkin using AI to make clinical trials, you know, more efficient. And for precision medicine tailoring treatments to individuals, you got Verge Analytics, even big names like AstraZeneca using AI.

Host 2: And that's just one slice. Look at providers, doctors, hospitals, the applications are just as broad. You see examples like uh Canary Speech or Cleerly health using voice analysis.

Host 1: Voice analysis, how does that work?

Host 2: Well, the AI looks at acoustic patterns, linguistic stuff to potentially screen patients for health issues.

Host 1: Wow.

Host 2: And think about cutting down on paperwork. Companies like Augmedix, DeepScribe automating documentation, that frees up doctors and nurses.

Host 1: That administrative burden is huge.

Host 2: It is. Then there's medical imaging AI from PathAI or Aidoc helping read X-rays, MRIs, potentially catching things earlier.

Host 1: Faster and maybe more accurate diagnoses.

Host 2: Potentially, yes. And don't forget getting different computer systems to talk to each other, EHR interoperability. Companies like Epic and eClinicalWorks are working on making patient data flow better. The big idea is reducing admin and boosting accuracy.

Host 1: It really touches so many areas. Even the insurance companies, the payers are involved. The paper mentions ConcertAI using predictive models for preventive care.

Host 2: Right, trying to stop problems before they start.

Host 1: And folks like UiPath and H2O.ai using AI to spot fraud in claims. That helps keep costs down, doesn't it?

Host 2: Absolutely. Then you move to medtech, medical technology. They mention generative design for new devices, AI helping optimize supply chains, getting things where they need to be. Companies like Thoughtful.ai, IQVIA are working on that.

Host 1: And it's not just companies, right? Public health agencies, too.

Host 2: Yeah, definitely. InstaDeep working on pandemic early warnings, the CDC looking at AI for resource allocation in crises. It really feels like AI could influence almost every corner of healthcare.

Host 1: And the growth they're predicting, it's pretty staggering.

Host 2: It is. Figure two shows this huge market expansion, something like a 43% compound annual growth rate hitting nearly $500 billion by 2032.

Host 1: Wow.

Host 2: And generative AI, the AI that creates stuff like text or images, that's projected to grow even faster within healthcare. An 85% CAGR reaching $22 billion by 2027. That's clearly where a lot of the focus is going.

Host 1: That sounds amazing. But then the paper brings us back down to Earth a bit.

Host 2: Right, the reality check.

Host 1: Despite all that potential, all that growth, figure three shows healthcare's actual maturity in using data and AI. It's lagging behind the global average compared to other industries.

Host 2: Exactly. And there's this statistic about US job ads that really drives it home. Only about one in every 1,850 healthcare job postings asks for AI skills.

Host 1: 1 in 1,800, that's tiny.

Host 2: It is. Compare that to IT, or even finance and insurance, education, they're way ahead. Only construction is lower. It just highlights that, yeah, the promise is there, but actually getting it widely used in healthcare is still pretty early days.

Host 1: So, massive potential, big money flowing in, but slow uptake. Why? The paper dives into this based on talks with what, over 75 experts?

Host 2: Yeah, and they found some really fundamental structural things holding AI back.

Host 1: Like what?

Host 2: Well, political stuff plays a role. Short election cycles mean pressure for quick results, which AI doesn't always deliver immediately.

Host 1: Right.

Host 2: The way national health systems are set up can make it hard to scale solutions everywhere. Plus, you need a lot of validation, long-term data to really prove the public health benefits.

Host 1: Okay, that makes sense.

Host 2: Then there are resource limits. Budgets are always tight, healthcare costs keep rising. It's hard to find big money for new AI projects. And, maybe not surprisingly, there's some resistance to change in medicine. Inertia.

Host 1: Old habits die hard.

Host 2: They do. And finally, you've got old IT systems, existing regulations, payment models. None of it was really designed with AI in mind, so they can be major roadblocks. It's this mix of deep-seated issues.

Host 1: So, building on those structural things, the paper really zeroes in on three main barriers stopping AI from scaling up properly. Figure four lays these out.

Host 2: Yeah. The first one is just how complex AI seems. It can be pretty intimidating for policymakers, for business leaders. Often, there isn't a clear strategy connecting AI to the bigger health goal or political aims.

Host 1: It's not just "let's do AI", it needs a purpose.

Host 2: Exactly. It needs to fit into a clear vision that everyone gets behind.

Host 1: Okay, barrier one: complexity and lack of strategy. What's two?

Host 2: Barrier two is this misalignment between the tech choices being made and those strategic visions. The paper suggests leaders sometimes just leave the tech decisions to the tech experts.

Host 1: Which sounds logical, but...

Host 2: But it can mean missing chances to use AI strategically. Plus, you get different departments doing their own thing with AI, no joined-up thinking. Misaligned incentives, they call it.

Host 1: Silos.

Host 2: Yeah.

Host 1: Always the silos.

Host 2: Pretty much. And the third big barrier is low confidence in AI. You've got public distrust sometimes, skepticism within the industry, and often the rules and governance are all fragmented. People need to trust these systems are safe, fair, and actually work.

Host 1: Okay. So, complexity, misalignment, and lack of trust, those are the big hurdles. But the paper doesn't just stop at the problems, right? It looks forward, too, with these four possible future scenarios for AI in health. Figure six. They stress these aren't predictions, just ways to think about what could happen.

Host 2: Yeah, and these visions really show how disruptive AI could be. First is transformation in well-being. Think widespread sensors, wearables maybe, collecting loads of personal health data.

Host 1: Like your watch telling you more than just your steps.

Host 2: Exactly. Enabling predictive care, personalized wellness plans, shifting the whole focus from treating sickness to promoting wellness.

Host 1: That's a fundamental shift.

Host 2: Huge implications for funding, delivery, the workforce, but big challenges too around data privacy, keeping control, building trust.

Host 1: Okay, that's vision one. What's next?

Host 2: Vision two is pretty radical: eight billion doctors.

Host 1: Whoa.

Host 2: The idea is everyone could have a personalized AI doctor on their phone or device giving real-time advice, maybe overcoming barriers like geography or cost.

Host 1: An AI doctor in your pocket.

Host 2: Kind of. But it would totally change healthcare roles, need completely new regulations, and the challenges are massive: making sure everyone gets fair access, upskilling professionals and the public to use and trust these things.

Host 1: Right. Vision three?

Host 2: Third is AI-powered operational excellence. This is more behind the scenes. Digital twins of hospitals to test things out.

Host 1: Virtual hospitals.

Host 2: Yeah, essentially. Predictive analytics for patient flow, ambient listening in clinics to automate notes, AI writing documents, basically streamlining everything.

Host 1: Reducing that admin burden again.

Host 2: Exactly. Big efficiency gains, lower costs, but challenges around equitable access to the tech, regulations keeping up, clinical risks if we rely too much on AI, and cybersecurity, of course.

Host 1: And the last one?

Host 2: The final vision is health leapfrog, particularly for low- and middle-income countries.

Host 1: Leapfrog?

Host 2: Yeah. The idea is they could potentially skip some traditional stages of healthcare development by adopting AI directly, using it to overcome infrastructure gaps, create new ways for patients to get care, new partnerships.

Host 1: So, using AI to jump ahead.

Host 2: Potentially. It would need things like digital public infrastructure, new business models. Hurdles include making it sustainable, ensuring fair access, managing integration and security risks in places with fewer resources.

Host 1: It's fascinating stuff. But the paper stresses that it's not the tech holding these visions back, right?

Host 2: No, not primarily. It's those systemic structural constraints we talked about earlier. The tech is moving fast, but the systems around it need to adapt.

Host 1: Absolutely. Okay, let's go deeper into those three core challenges, then. The paper really unpacks them, starting with that complexity issue deterring policymakers and leaders. It's not just tech complexity.

Host 2: No, it's also about proving the value in a way that clicks with political agendas. It's hard to get political and financial backing when there's pressure for quick wins.

Host 1: And AI in health is often a long game.

Host 2: Exactly. High upfront costs, benefits might take time. It's a tough sell. The paper says the basic value of AI in health isn't always clearly understood, so leaders hesitate.

Host 1: It's that gap between the hype and the reality on the ground. Like that example they give: tech increases patient demand, but the system can't cope, so innovation stalls.

Host 2: Right. It shows you need clear, measurable goals for AI, focused on validating data, making sure it's diverse, thinking end-to-end about optimization.

Host 1: This is where those public-private partnerships, PPPs, come in?

Host 2: Yeah, they're seen as critical. They can bridge that gap, show the practical value. Policymakers get cost-benefit insights, private sector understands the rules and priorities better. And we need new ways to evaluate AI's impact, like tracking real-world evidence.

Host 1: And funding it. The paper suggests a two-pronged approach.

Host 2: That's right. Short term, you need public-private investment to get through that tricky early phase, the valley of death.

Host 1: Ah, yes, the valley of death.

Host 2: Managing change, gathering evidence, proving value. It means investing in training, skills, and doing multiple projects together, not just isolated pilots, to really change the system.

Host 1: They mention huge potential savings, like $150 billion annually in the US by 2026 from Accenture.

Host 2: They do. But it needs new ways of buying this stuff, new procurement processes.

Host 1: Okay, that's short-term. Long-term?

Host 2: Long term means market-based financing for sustained investment and scaling. That means new cost models: who pays for what at each stage—research, development, scaling, especially for early innovators.

Host 1: And the potential returns are big.

Host 2: The paper cites studies suggesting maybe 10-15% annual ROI over 5 years, potentially 10% savings in total healthcare spending. But it relies on good reimbursement models, good datasets, solid health technology assessments, HTAs.

Host 1: Evaluating the value of the tech.

Host 2: Exactly. And health economics and outcomes research, HEOR. They even call for HTA standards to align globally. And crucially, making sure investment is equitable and supports local AI development, which bodies like G20 and WHO emphasize.

Host 1: Okay, moving to the second big challenge: fragmented coordination. It's not enough for individual hospitals or companies to do AI, is it?

Host 2: No, you need a cohesive ecosystem. Large-scale AI needs coordination: national governments setting a vision, but also local champions driving it on the ground.

Host 1: A top-down and bottom-up approach?

Host 2: Kind of. AI can redesign systems, but governments need to lead with a clear vision and roadmap. And the change needs to be demand-driven. Listen to patients, communities, not just payers. Public campaigns are needed, too, to get people on board with changes like digital front doors.

Host 1: And those local champions are key, like that India TB example.

Host 2: Yeah, the AI for TB initiative. Using AI apps locally boosted early detection by 16%. That's significant. But the paper says that local push needs a pull from policymakers demanding these solutions. That coordination is vital.

Host 1: Another example was the UK's COVID imaging database.

Host 2: Right, the NCCID. The NHS built this huge data repository, enabling collaboration between hospitals, universities, companies. It led to AI tools for faster diagnosis, better ICU resource allocation. Shows the power of shared data and decentralized work.

Host 1: So, it's about balancing short-term pressures with long-term goals. And leaders need to get involved in the tech decisions, not just leave it to the experts.

Host 2: Absolutely. There's a real danger in deferring those choices. It can lead to fragmented, inefficient systems.

Host 1: Like how things often evolved historically.

Host 2: Exactly. Separate tenders, lowest-cost wins leads to disjointed systems. You need a clear, long-term architectural vision, and that vision needs to see health as part of broader digital public infrastructure, DPI, strategies.

Host 1: So, health tech isn't built in isolation.

Host 2: Right. Avoid reinventing the wheel. Ensure fair access, because the paper notes that poor DPI is often why digital health access is unequal. Using comprehensive datasets, including social factors, SDOH, helps tackle bias. Promoting local AI production, integrating old systems, it's all part of that coordinated picture.

Host 1: And that equity point keeps coming up. Table two in the paper talks about strategies for data ownership and access to make it fairer.

Host 2: Yes, equity has to be central. Examples like the UK's NHS centralizing data or Israel's system enabling their rapid vaccine rollout show how good data management and integration are key.

Host 1: Then there's the digital literacy issue, even among leaders.

Host 2: Big hurdle. AI is still new for many decision-makers. Lack of understanding can fuel fear, lead to bad regulations, make people hesitant.

Host 1: So, upskilling is crucial.

Host 2: Definitely. Not just on AI strategy, but basic digital principles, defining what leaders need to know, equipping everyone—leaders, users—to properly evaluate AI tools, track KPIs, understand the value, and educating patients, professionals, caregivers, too, to build trust and show AI complements, not replaces, people.

Host 1: Which brings us neatly to barrier number three: low confidence and fragmented regulation. Trust is everything here, isn't it?

Host 2: It really is. You need AI to be transparent, accountable. Medicine is naturally cautious, so you need clear, adaptable regulations that keep up with the tech.

Host 1: One size fits all doesn't work for AI.

Host 2: Not really. And you need harmonized data-sharing rules across borders for collaboration. Plus, the WHO stresses keeping humans in the loop. Human oversight is vital for trust and safety.

Host 1: Don't treat the AI like a person either.

Host 2: Right. Avoid anthropomorphizing. Keep accountability clear. From a regulatory view, the paper sees big short-term challenges: rules varying wildly between countries, lack of clarity stifling innovation, rigid rules not fitting software or GenAI, and trying to regulate without clear data-sharing rules.

Host 1: That gap between regulatory leaders and laggards is widening, too.

Host 2: It is. Public engagement, political will like that US executive order on AI are needed. And supporting emerging economies with their regulations is key.

Host 1: The paper suggests regulation is often too slow, waiting for AI to be finished.

Host 2: Yeah, which is impractical. They suggest using a mix of tools. Guidance and standards are more flexible than strict laws. Invest in those early. And we need to speed up validation, maybe delegate parts of it under supervision to non-profits, providers, the private sector. Collaboration with industry is crucial for practical, agile rules.

Host 1: Especially for generative AI, which changes as it's used.

Host 2: Exactly. Needs adaptive rules, more focus on post-market surveillance, watching how it performs in the real world. And data protectionism, while important, can sometimes block access to diverse data needed for unbiased AI and local testing. We need common data standards like FAIR principles and techniques like federated learning to share insights without sharing raw data.

Host 1: Building trust across this whole complicated system is tough.

Host 2: Very tough. Global studies show trust levels vary. Misinformation doesn't help. It needs effort from regulators and businesses: transparency, integrity, human oversight, clear accountability. Businesses need to show competence, reliability, shared values.

Host 1: Figure seven shows that difference in consumer attitudes globally, right? High concern in places like the US and UK, less in China and India.

Host 2: Exactly. Shows you need different approaches to build trust in different places.

Host 1: Okay, so we've covered the context, the visions, the big challenges. Now, the strategy part. The paper offers six calls to action to actually create value. Says those three challenges are linked in a vicious cycle.

Host 2: Yeah, and breaking it means changing how leaders approach AI and digital health.

Host 1: So, call to action one?

Host 2: From dreaming of breakthroughs to delivering near-term benefits. Basically, start with practical operational AI like supply chains, things with clear, short-term ROI.

Host 1: Build momentum.

Host 2: Exactly. Learn from other industries. Start small, be agile, track results. Plus, these operational uses can really help reduce caregiver workload.

Host 1: Makes sense. Number two?

Host 2: From private sector progressing independently to public-private ecosystems. Especially for medical AI, we need public and private sectors to agree on priorities, understand risks, figure out how to share the value created.

Host 1: Like they do in defense, maybe.

Host 2: Something like that. It's vital for big clinical changes and avoids investing in things that won't get reimbursed or scaled later.

Host 1: Okay. Third call?

Host 2: From fighting on infrastructures to winning on services. Stop competing over building the basic plumbing. Prioritize shared foundations, especially digital public infrastructure, DPI.

Host 1: Let the private sector compete on the services built on top.

Host 2: Precisely. Public leaders see DPI as crucial investment, encourage blended financing, more private investment in infrastructure.

Host 1: Number four is aimed squarely at leaders, isn't it?

Host 2: Yes. From leaders with good intentions to leaders who make responsible technical decisions. Health leaders, CEOs, clinicians need to skill up, engage with the tech stuff, ask tough questions.

Host 1: Be healthily skeptical.

Host 2: Exactly. Like ensuring interoperability is actually a requirement, not just a wish. Build capacity, get AI into medical training. Understanding AI potential, limits, risks has to be a core skill for leaders.

Host 1: Fifth call to action?

Host 2: From waiting for guidelines to proactively building trust. Don't just wait for regulators, they might lag. Use flexible risk management, boost post-market surveillance to spot problems early, be transparent, maybe set up AI ethics committees, keep existing standards up to date.

Host 1: Be proactive on trust. And the final one, number six?

Host 2: From dispersed data to deliberate integration. Data access is key for trust and performance. The call is for datasets that are globally connected, but locally controlled.

Host 1: Including wider medical and socioeconomic data.

Host 2: Yes. It keeps local ownership, but allows global collaboration, helps address local needs, and it supports that first point about near-term benefits by having common data models.

Host 1: So, summing up, the paper sees massive potential, but...

Host 2: But realizing it depends on policymakers making AI a priority, aligning tech with strategy, and getting shared regulations right to build trust.

Host 1: And embracing those six shifts is key to speeding things up for a healthier, more equitable world.

Host 2: And it all comes back to new kinds of cooperation between public and private leaders, building a sustainable AI ecosystem, avoiding fragmentation, finding people-centered ways to work together. PPPs are highlighted again as a key mechanism.

Host 1: Okay. So, thinking about everything we've discussed: personalized AI health assistants, all this data, what's the most pressing thing you think needs addressing to make sure AI truly serves people's well-being in healthcare? That's something for you all to ponder.