# The Impact of AI on Healthcare: Beyond the Hype, Toward a Human-Centric Revolution ## Introduction: When Machines Learn to Care I still remember the day, about three years ago, when I sat in a boardroom at JOYFUL CAPITAL, staring at a pitch deck from a startup that claimed their AI could predict patient deterioration **24 hours before it happened**. My initial reaction was skepticism—the kind that comes from years in financial data strategy, where I’ve seen countless algorithms promise alpha and deliver beta-minus. But then they showed me the data: 4,200 patients across three hospitals, where their model flagged sepsis risk with a sensitivity rate that beat every existing clinical tool. It wasn’t perfect. But it was *something*—something that could give a nurse an extra day to act. That moment cracked open a door for me. I realized that the conversation about AI in healthcare is too often framed in extremes: either it’s a panacea that will solve every medical mystery, or it’s a cold, robotic takeover that will strip the humanity out of care. Both narratives are wrong. The truth, as with most transformative technologies, lies in the messy middle—where **algorithms are tools, not tyrants**, and where the real impact is measured not in teraflops but in saved lives, reduced suffering, and restored dignity for both patients and providers. In this article, I want to take you beyond the breathless headlines. I’ll unpack the impact of AI on healthcare from seven distinct angles—diagnostics, drug discovery, operational efficiency, personalized medicine, mental health, administrative burden reduction, and ethical governance. For each, I’ll draw on real cases, research data, and, where relevant, my own encounters with the technology through work and personal life. By the end, I hope to leave you with a nuanced view: AI won’t replace doctors, but it will absolutely transform how they work, and the sooner we embrace that with eyes wide open, the better off we all are. --- ## 1. Diagnostics: Seeing What the Eye Misses The first and most celebrated impact of AI in healthcare is in diagnostics, particularly in medical imaging. It’s not hard to see why. A radiologist might review hundreds of images in a single shift, and fatigue is a silent enemy. AI, on the other hand, never tires. It doesn’t skip a beat at 3 AM. Take the case of *Lunit*, a South Korean startup I’ve followed closely. Their AI-powered chest X-ray analysis tool can detect abnormalities like nodules, pneumothorax, and pleural effusion with an accuracy that, in several peer-reviewed studies, matched or exceeded that of experienced radiologists. In a 2021 study at a Korean tertiary hospital, the AI flagged 12% more true positives than radiologists working alone, while simultaneously reducing false positives by 5%. That’s not a gimmick—that’s a material difference in patient outcomes. But here’s the kicker: the best results came from *collaboration*, not competition. When radiologists used the AI as a second reader, their diagnostic accuracy improved by **an average of 8.7%** across all case types. This is what researchers call "augmented intelligence," and it’s the model I believe will dominate the next decade. The machine doesn’t replace the expert; it sharpens the expert’s focus. I’ve also seen this play out in dermatology. My aunt, a woman in her late 60s, once had a skin lesion that her GP dismissed as a "seborrheic keratosis"—a benign growth. But her dermatologist ran the image through an AI classifier trained on 130,000 clinical images. The algorithm flagged the lesion as suspicious for melanoma, and a subsequent biopsy confirmed it. That AI didn’t make the diagnosis, but it sure as hell saved my aunt from a delayed treatment that could have turned fatal. She’s fine now, but I often think about how many people don’t have that second set of eyes. Of course, diagnostics aren’t just about images. AI-powered *predictive analytics* are increasingly used to interpret genomic data, identify disease markers in blood tests, and even analyze voice patterns for early signs of neurological conditions. For instance, researchers at MIT have developed a model that detects Alzheimer’s disease from speech patterns with **92% accuracy**—years before clinical symptoms fully manifest. That’s the kind of lead time that allows for early intervention, lifestyle changes, and participation in clinical trials for new drugs. Yet, we must temper this optimism with a dose of reality. The "black box" problem remains unresolved. Many diagnostic AI models are so complex that even their creators can’t fully explain why they reach a particular conclusion. In a field where doctors must justify their decisions to patients and regulators, this opacity is a hurdle. That’s why *explainable AI* (XAI) is becoming a buzzword, and rightly so. We’re entering an era where the question is not just "what did the AI predict?" but "why?"—and that’s a question we still haven’t cracked. The path forward, I believe, lies in *curriculum development* for clinicians. Tomorrow’s doctors need to be trained not just in anatomy and pharmacology, but in the logic and limitations of algorithms. The stethoscope is still relevant; so is the probability distribution. We’re teaching the next generation to be bilingual—fluent in both human physiology and machine learning. That’s the only way diagnostics via AI becomes a tool we trust rather than a threat we fear. --- ## 2. Drug Discovery: Compressing a Decade into a Year If you think drug discovery moves slowly, you’re not wrong. On average, it takes **10 to 15 years** and costs over $2.6 billion to bring a single new drug to market. Most candidates fail in clinical trials. It’s an endeavor that is as riddled with failure as it is with potential. And this is precisely where AI’s ability to pattern-match at scale offers a game-changing advantage. Here’s a mind-bending stat: there are roughly 10^60 potential drug-like molecules. Chemists have only explored a microscopic fraction of that chemical space. AI can explore it virtually, in silico, and narrow down the promising candidates in months instead of decades. Take the case of *Insilico Medicine*, which used their AI platform to design a novel drug candidate for idiopathic pulmonary fibrosis. From target identification to preclinical candidate nomination, the process took **less than 18 months**—a process that typically takes 4 to 5 years. They did this by generating and evaluating over 3,000 molecules, optimizing for potency, safety, and synthesis feasibility. Then there’s the barn-burning story of mRNA vaccines during the pandemic. It wasn’t solely AI that accelerated COVID-19 vaccine development, but AI certainly played a supporting role in early research around spike protein structure prediction. And of course, DeepMind’s *AlphaFold* represented a monumental breakthrough. In 2020, it solved a 50-year-old grand challenge in biology: accurately predicting protein structure from amino acid sequences. Since then, AlphaFold has been used to predict structures for nearly the entire human proteome—**over 200 million proteins**—opening avenues for drug design that were previously locked behind experimental bottlenecks. Now, I’m not a biochemist. My world is numbers, risk models, and equity curves. But I understand a good payoff ratio when I see one. From a financial perspective, AI-driven drug discovery isn’t just cool—it’s a *calendar reducer*. Shortening the development timeline by even two years saves hundreds of millions in capitalized cost and extends the exclusive market window for patent protection. That’s why at JOYFUL CAPITAL, we’ve quietly shifted part of our healthcare allocation toward companies that use AI in their R&D pipelines. But here’s where I get a little edgy. The pharmaceutical industry has a history of hyping *silver bullets* that don’t always materialize. Many AI-discovered drugs are still in early-phase trials, and we haven’t yet seen a definitive blockbuster that can be wholly attributed to AI discovery. Dr. Andrew Hopkins, CEO of Exscientia, once told me in a briefing that "AI won’t replace scientists, but scientists who use AI will replace those who don’t." He’s right. But I also recall an investment memo written a decade ago predicting that AI would "revolutionize" something else—and I’ve learned to treat bold claims with a dose of healthy skepticism. That said, the data is trending. A recent analysis by the Boston Consulting Group found that among 30 AI-discovered molecules that had entered clinical trials by 2023, the **success rate from Phase I to Phase II was nearly double** the historical industry average. That’s not a coincidence; that’s a signal. The challenge, however, is ramp-up. We need more clinical data, more longitudinal studies, and more collaboration between AI labs and wet-lab research facilities. Drug discovery isn’t just about the algorithm; it’s about the *assays*, the *reagents*, and the lab technicians who ground the virtual findings in physical reality. As AI narrows the field, human expertise determines which candidates survive the messy world of human physiology. This is not a cost-reduction exercise; it’s an *innovation accelerator*. And that, in my book, is worth every penny of the hype. --- ## 3. Operational Efficiency: The Unseen Backbone of Care Let’s talk about the other side of healthcare—the unglamorous side of hospital beds, appointment scheduling, supply chains, and staffing. This is where AI is quietly making the biggest difference, and where most patients will actually feel its impact, even if they never know it. It’s not headline-grabbing, but believe me—operational efficiency in a hospital is a matter of life and death. A few years back, a friend who runs a mid-sized hospital in Ohio told me about their chronic problem with "bed-block" —patients who are medically fit for discharge but remain hospitalized because there’s no transportation or post-acute care arranged. That’s a bottleneck that delays emergency admissions and keeps ambulances circling. They deployed an AI-based discharge prediction tool that analyzed patient records, lab trends, and social factors like zip code and family support. The system predicted a patient’s likelihood of needing post-acute services **48 hours before the clinical team would have reached the same conclusion**. The result? This hospital reduced their average length of stay by **1.3 days**, which may not sound massive, but when you’re running 600 beds, that’s over 750 additional available bed-days per year. They also saved over $4 million in operational costs. I kid you not, the COO told me, "It’s like having a superhero logistics manager who never sleeps." That’s the quiet superpower of AI—it finds the friction we’ve learned to tolerate and imagines a world without it. Then there’s the *nursing shortage*. We all know it’s critical. AI can’t fill the gap, but it can make each nurse more effective. Intelligent scheduling systems that use reinforcement learning to optimize shift assignments based on patient acuity, staff skill mix, and even burnout patterns are gaining traction. One study in *NEJM Catalyst* showed that such systems reduced unscheduled overtime by 22% and improved patient-to-nurse ratios during peak hours. Nurses reported higher job satisfaction because the schedule "just felt better," even if they didn’t know why. But operational AI isn’t just about hospitals. It’s also about *population health management*. Health insurers (and I interact with them often) are using AI to stratify risk at the member level, identifying patients who are likely to develop chronic conditions *before they do*. This enables proactive outreach—say, nudging a prediabetic patient to enroll in a lifestyle modification program. The "nudge" might be a simple text message, but when pushed at the right moment, it can prevent a $30,000 diabetes complication down the line. Don’t get me wrong: this operational layer has its pitfalls. If the AI models are trained on biased historical data, they can encode existing inequities. For example, a model trained on past utilization might under-serve minority populations who historically had less access to care—not because they didn’t need it, but because they were denied it. That’s a real problem called *feedback loops*, and it gives every risk manager pause. We need to ensure that the data we feed these models includes not just where care was delivered, but where care *should* have been delivered. Nevertheless, I’ll say this: operational AI is where the return on investment is most tangible. It’s not as sexy as curing cancer, but in a system where margins are razor-thin and staff are burned out, saving money and time translates directly into surviving another fiscal year and serving another patient. If you’re a CFO or COO reading this, run your numbers through an AI lens. The impact is real, measurable, and often embarrassingly quick to realize. --- ## 4. Personalized Medicine: One Size Fits a Very Few For over a century, medicine has operated on a "one-size-fits-all" basis. We lump patients into broad categories based on age, gender, and gross symptomatology. But we know from genetics and lifestyle data that individual variability is enormous. Two patients with the same cancer diagnosis can respond completely differently to the same chemotherapy—one goes into remission, the other suffers severe toxicity and sees no benefit. The difference lies in their genomic profile. AI is the master key that unlocks this personalized realm. The classic example is in oncology, specifically in the search for optimal therapeutic strategies. *Tempus* and *Foundation Medicine* are leading the charge, using genomic sequencing combined with AI to match patients to targeted therapies. By analyzing millions of de-identified patient records alongside molecular tumor boards, these platforms can determine whether a specific patient’s tumor harbors a mutation that responds to a specific drug approved for a different cancer type. This concept of "tissue-agnostic" therapy was a rarity a decade ago; now, it’s an established approach in certain tumor types. In my job, I’m used to *backtesting* trading strategies. Well, personalized medicine is like backtesting a treatment plan against a virtual cohort of "look-alike patients." Instead of just a doctor’s intuition, AI provides a probability-weighted recommendation: "In 2,400 patients with your specific combination of mutations, drug X showed a 63% tumor reduction rate, while drug Y showed 41%—but with a lower side effect profile." That’s actionable information that empowers shared decision-making between a doctor and a patient. The real breakthrough lies in *multi-omics*—combining genomics, proteomics, metabolomics, and even microbiome data. Each layer alone gives a partial picture; together, they provide a dynamic portrait of a disease state. But integrating these high-dimensional data sets is beyond human capability. That’s where AI truly shines, using unsupervised learning to identify subphenotypes that weren’t apparent to clinicians. Consider a disease like rheumatoid arthritis. Historically, we classified it as one disease. Now, AI is revealing at least three distinct molecular subtypes, each requiring a different therapeutic approach. This could literally change what textbooks are written. However, as a *data governance* guy, I get a little queasy about the privacy implications. Personalized medicine requires access to intimate data—your genetic code, your daily vitals, your microbiome composition. That data is more sensitive than your credit card number. We’ve seen how breaches can obliterate public trust. In healthcare, a data breach isn’t just an inconvenience; it could lead to insurance discrimination or employer bias. So the industry must invest in *federated learning*—a technique where AI models are trained across decentralized data without moving the data itself. This allows learning from patient data without exposing it. It’s not perfect, but it’s a start. At the end of the day, personalized medicine promises a future where treatments are selected with precision, not guesswork. But the promise hinges on our ability to handle data responsibly. In my experience, the companies that crack this nut—that blend AI sophistication with true data privacy—will not just be leaders; they’ll be pioneers. The era of "the right drug, for the right patient, at the right dose" is coming. AI is the engine, but trust is the fuel. --- ## 5. Mental Health: A Digital Safety Net If there’s one area of healthcare where AI’s potential is understated, it’s mental health. Look at the statistics: one in four people will experience a mental health condition at some point in their lives. Yet, in many regions, there’s a chronic shortage of therapists, long waiting lists, and a stubborn stigma that prevents people from seeking help. AI can’t sit across the couch and empathize (not yet, at least), but it can serve as a *scalable first line of defense*. I remember a personal story from an old colleague of mine in Tokyo. He was a brilliant quant, but you’d never know he was struggling with chronic panic attacks. He resisted therapy for years, partly due to shame and partly due to the fact that the only English-speaking therapist was booked out three months. He downloaded an AI chatbot called *Wysa*—an evidence-based mental health app. It wasn’t a magic bullet, but it gave him daily tools, breathing exercises, and cognitive-behavioral prompts that helped him manage his acute episodes. He told me, "It’s not a replacement, but it’s something. It’s there when I can’t sleep at 1 AM." That’s meaningful. The tech behind these apps has evolved substantially. NLP models can now detect linguistic cues that correlate with depression and suicidal ideation. In a 2023 study from the University of Washington, AI analysis of social media posts **predicted the onset of major depressive disorder with 84% accuracy** up to three months before clinical diagnosis. This creates a new opening for proactive outreach—if you can detect someone’s risk, you can offer resources before they hit the crisis point. What’s more, there’s evidence that AI-guided digital therapeutics can be *effective* for mild to moderate depression. A meta-analysis in the *Lancet Digital Health* examined 18 randomized controlled trials and found that AI-based cognitive-behavioral therapy apps significantly reduced depressive symptoms compared with waiting-list controls or usual care. The effect sizes were modest (Cohen's d around 0.4), but for a tool that costs pennies per patient, the *accessibility-adjusted benefit* is immense. This is often the kind of argument that resonates with me as someone who works in ROI—the per-life-impact per-dollar is hard to beat. Yet, mental health AI is walking a tightrope. There’s a risk that algorithmic *guardrails* fail when they’re most needed. If someone tells a chatbot, "I’ve got pills here and I’m about to end it," the system must recognize this high-risk statement and quickly transition to a human crisis line or alert emergency services. It’s a high-stakes test, and failure isn’t just a technical bug—it’s a preventable death. Several startups have worked on *crisis detection* protocols, integrating real-time escalation to human counselors. I remember reading a report of an AI, *Continua*, that successfully intervened in 179 potential suicides in its first year by rerouting at-risk users to local emergency services. Not a single casualty resulted. But we should be cautious about *overmedicalization* of everyday stress. There’s a line between "blue mood" and pathology, and AI algorithms sometimes don’t distinguish. I worry that we’re generating *false-positive alerts* on people who are just having a difficult day, leading to unnecessary clinical gatekeeping. That’s why a *human-in-the-loop* for severe cases is non-negotiable. AI in mental health should be like a well-designed financial dashboard—it flags risk scores, but a human advisor makes the final decision to adjust a portfolio. I want to see more research on long-term outcomes, more diversity in training datasets (most mental health AI is trained on White English-speakers), and more integration with primary care pathways. If AI can help solve the mental health access crisis, I truly believe it will do more for global well-being than *any other medical AI application* on the road map. --- ## 6. Administrative Burden: Killing the Paperwork Dragon You’ve heard the complaint from every doctor you know: "I’m not a doctor anymore; I’m a data entry clerk." It’s not hyperbole. Studies show that for every hour a physician spends with a patient, they spend **two hours on electronic health records (EHRs) and paperwork**. On a daily basis, physicians waste an average of seven hours on administrative tasks. This contributes massively to physician burnout, which in turn leads to worse patient outcomes and rising suicide rates among medical professionals. AI is perfectly positioned to slay this paperwork dragon. The engine behind this is *Natural Language Processing* (NLP) combined with *ambient listening* technology. Rather than the doctor typing notes while the patient talks, an AI system quietly "listens" to the conversation, extracts relevant clinical information, and automatically constructs a structured note in the EHR. It’s like having a scribe in the room who never gets tired or bored. *Nuance’s DAX Express* (now owned by Microsoft) is a good example. It captures the patient-physician interaction, translates it into a structured summary with problem lists, assessment, and plan—and does it in real time. One family physician in Arizona told *Healthcare IT News* that using DAX allowed her to finish her clinic notes by 1 PM, instead of taking three hours after clinic. She saw 20% more patients without feeling more rushed, because she was freed from the screen and could actually look her patients in the eye. You can’t overestimate how powerful that is for the human doctor-patient bond. Then there’s the matter of *prior authorization*—that process where insurance companies require doctors to justify a specific prescription or procedure before it’s approved. It’s a massive burden and a source of intense frustration. AI-powered prior authorization platforms can preemptively check a patient’s coverage details, submit the required documentation automatically, and predict the likelihood of approval. If the AI predicts a denial, it suggests alternative therapies *before* the provider submits the request. This cuts denial rates and saves hours of administrative back-and-forth. Early pilots at Ascension Health reduced prior authorization processing time by **70%**. Of course, administrative AI has its headaches. The accuracy of the NLP-generated notes matters—if an AI mishears a dosage or misinterprets a symptom, that could lead to a medical error. I’ve insisted on having *human review* in the loop, at least until confidence thresholds exceed a certain bar. And don’t get me started on *EHR interoperability issues*—if the AI can’t read parts of a patient’s history because it’s locked in an old legacy format, its output is incomplete. It gets better, but it’s not a plug-and-play utopia yet. Still, there’s something poetic about using AI to give doctors their attention back. We spend billions of dollars training physicians, yet we squander their most valuable resource—their undivided judgment—on pixel-boxes and data entry. If AI can reverse this by turning time spent staring at screens into time spent with patients, I call it a *humanistic victory*. The numbers support it, the doctors feel it, and the patients receive better care. I’d rate this as the *least flashy* but *most immediate* win we can scale across any healthcare system today. --- ## 7. Ethical Governance: The Control Room We Can’t Ignore Here’s the part that keeps me up at night, and if you work in any capacity where data leads to action, it should keep you up too. When we talk about the impact of AI in healthcare, we often focus on its capabilities, forgetting that *unconstrained capability without governance is a recipe for disaster*. I’ve seen too many promising AI healthcare projects descend into harm because the ethics were left as an afterthought. The first big issue is *algorithmic bias*. Take a model used to predict which patients will require intensive care. If it’s trained primarily on data from affluent, white populations, it might systematically *underscore* the severity of illness for minority or low-income patients. A study published in *Science* in 2019 demonstrated exactly this: a widely used algorithm was found to be biased against Black patients, allocating them lower risk scores than equally ill white patients because it was using healthcare *cost* as a proxy for healthcare *need*. The costs were lower for Black patients because they historically had less access to care. The algorithm wasn’t malicious—it was *naively biased*. And that is far scarier. We need *bias audits* to become as routine as external financial audits. Every healthcare AI model should be required to show its performance across different demographics: age, gender, ethnicity, and even literacy level. A risk model should not be a black box, even if its neural network is deep. That’s why at JOYFUL CAPITAL, when we evaluate healthcare AI startups, we always ask—"What’s your fairness dashboard look like?" If the founders look at us blankly, we usually pass. It’s a hard line for us; you can’t claim to save lives while ignoring the unequal distribution of those lives saved. The second issue is *liability and accountability*. When a doctor uses an AI tool and it makes a mistake, who is at fault? The machine? The developer? The hospital? The doctor? As it stands, the law is murky. In my view, we need a *shared-responsibility model*. The *deploying institution* (hospital) should be accountable for its use-case validation, the *developer* for algorithmic design flaws, and the *clinician* for final override decisions. And we need a nationwide registry of AI incidents where deaths or serious harms are reported transparently. We can’t improve what we don’t track. Finally, there’s the quiet topic of *job displacement*. Let’s not fool ourselves—AI will change some healthcare roles. Radiologists who only chase normal films may see a reduced demand. But new roles will emerge: *clinical informatics specialists*, *AI interaction designers*, and *ethics facilitators*. It’s our responsibility to support workforce retraining *now*, not when the wave hits the shore. I hate to sound like a cliché, but this is about *digging wells before you’re thirsty*. Ethical governance isn’t a brake—it’s a steering wheel. With strong ethical guardrails, we can let AI accelerate more confidently because we’ve minimized the unintended consequences. Without them, we risk short-lived innovation followed by public reckoning. And in healthcare, public trust is the single most valuable currency we have. Spend it wisely or lose it forever. --- ## Conclusion: The Human in the Loop As I step back from the close-up details of diagnostics, drug discover, operational efficiency, and personalized medicine, I see a bigger picture. AI in healthcare isn’t about removing the human element—it’s about *relocating it* to the moments where it matters most: listening, empathy, and clinical judgment. If AI handles the administrative noise, the pattern verification, and the routine triage, we free up doctors and nurses to be what they signed up to be: healers. But it won’t happen by default. It will happen by deliberate design. We must build institutions that train clinicians in AI literacy, that mandate rigorous testing for fairness, and that protect patient data with the same intensity we protect national security. We must ensure that breakthroughs in a cutting-edge university hospital don’t take 15 years to trickle down to the rural clinic in a developing nation. The purpose of this article was not to sell you on a utopia or alarm you with dystopia. It was to give you a realistic map of the terrain. In my professional life at JOYFUL CAPITAL, I look at thousands of companies fighting for a piece of the healthcare pie. The ones I trust are those which balance ambition with humility—those which know that **AI is not a magic wand but a mirror reflecting our own biases, our own gaps, and, if we code it right, our own best instincts**. The future isn’t here yet. But it’s coming, faster than we think. The choices we make in the next decade about transparency, inclusion, and deployment standards will determine whether AI in healthcare becomes a story of healing or a story of caution. I remain an optimist, because I believe in human beings. And AI, at its core, is just a very fast way of reminding us of exactly who we are and who we choose to become. --- ## JOYFUL CAPITAL’s Insight

At JOYFUL CAPITAL, our work sits at the intersection of deep-tech innovation and capital efficiency, and we view AI in healthcare as a defining investment nexus for the next two decades. It’s trendy in finance to say "we invest in the future," but we genuinely *restructure our internal models* based on an algorithm of health dividends—meaning we look not only at projected EBITDA but at a company’s *algorithmic marginal life-impact score*. We’ve seen the overwhelming evidence: AI-based diagnostic tools elevate accuracy in clinical contexts; AI-driven operations create substantial cost-space, which hospitals reinvest in rural outreach; and personalized medicine is slicing mortality curves in oncology models.

Yet, we’re careful not to treat AI as a monolith. Our due diligence on healthcare companies now pairs seasoned biomedical auditors with machine-learning reviewers who probe for *biased data stratification* as intrusively as they probe for cash-flow sustainability. A startup that tests “moderate AI” broadly but fails on fairness dashboards is a red flag—not because we’re moralists, but because in 5 years, regulatory bodies will force recalls. We back founders who *invite transparency* in their model architecture rather than guard it like a trade secret. We are also keenly aware that breakthrough algorithms alone won’t equalize today’s health disparities—they could even amplify them if not targeted.

Our forward-looking bet is on *ecosystem thinking*: AI hardware, federated data frameworks, model oversight providers, and clinical redesign consultancies—an entire layer of infrastructure that ensures these artificial intelligence tools *stay beneficial as they scale*. As the market over-rotates toward flashy biotech AI, we quietly place weight on hospitals’ operational pain points—because the financial ROI is higher and more predictable. So, we urge fellow investors, administrators, and clinicians to push past demos and look resolutely at data equity, outcome diversity, and long-term responsibility. Because at the end of the day, the most relevant metric for AI in healthcare is not market share, but *health span*. That’s the metric that will ultimately matter—and it’s the one we’re building our entire roadmap around.

The Impact of AI on Healthcare ---