Part of a Series on the Coming Labour Disruption
The Last Fortress: Why Physicians, Teachers, and Therapists Are Not as Safe as They Think
There is a particular kind of professional confidence that deserves its own name. It is not arrogance, and it is not ignorance. It is the quiet, settled conviction — held by physicians, teachers, therapists, social workers, and others in the caring professions — that whatever is coming for the rest of the labour market will stop politely at their door.
They have heard the warnings before. They have watched other sectors absorb disruption. And they have concluded, not unreasonably, that what they do is different: too human, too relational, too morally serious to be handed to a machine.
This article is written for those people — and for the people who love them — with the same honesty that has run through this series from the beginning. The protected professions are not going to be spared. The argument that they are is already contradicted by available evidence, undermined by straightforward economics, and further weakened by the fiscal dynamics described in Part Three. What is coming for physicians, teachers, and therapists is not the same as what came for data entry clerks or travel agents. But it is real, it is already measurable, and most of those bearing it will not see it coming — because it will not arrive as a sudden blow. It will arrive as a slow, largely invisible redistribution of who gets the human version and who gets the clinical substitute.
Before going further, this series owes its readers a clear definition. ACEs — Advanced Cognitive Enablers — are not a rebranding of “critical thinking,” a synonym for emotional intelligence, or a credential that can be earned by completing the right course. They are a cluster of higher-order cognitive capacities that, taken together, allow a person to function well in conditions of genuine complexity, ambiguity, and change.
They include the ability to see patterns and connections that are not obvious — to look at a situation and notice what is actually driving it rather than what appears to be driving it. They include the capacity to generate genuinely new possibilities rather than selecting from a familiar menu of options. They include the ability to hold multiple competing interpretations of a situation in mind simultaneously, without the discomfort of uncertainty forcing a premature conclusion. They include metacognitive awareness — the ability to observe your own thinking, notice where it is running on habit rather than genuine reasoning, and correct course before the habit leads you somewhere wrong. And they include what might be called cognitive courage: the willingness to follow an argument to a conclusion that is uncomfortable, unpopular, or that contradicts something you believed before you started thinking.
None of these capacities develops automatically. As this article discusses in the section on brain development and the cohort AI is arriving for, the neurological substrate for ACEs — the prefrontal cortex maturation that makes genuine abstract reasoning possible — does not complete until the early to mid-twenties, and even then it creates only the hardware. The software must be explicitly built through deliberate practice, honest feedback, and the willingness to be wrong. Most educational institutions, despite decades of rhetoric about critical thinking and higher-order learning, do not actually teach this. Most professional formation does not require it. Most workplaces do not reward it — at least, not until the moment arrives when nothing else is sufficient.
That moment is arriving now.
ACEs are not a guarantee of survival in the coming transition. They are not a magic credential or a protective title. They are the qualities that, in every previous civilisational disruption, allowed individuals and communities to navigate conditions that destroyed others — not because those others lacked intelligence or effort, but because they lacked the cognitive flexibility to see what was actually happening and respond to the world as it was becoming, rather than as it had been.
The “My Work Is Different” Defence
Every conversation about AI and professional work eventually reaches the same point. It comes from intelligent, well-educated people who have thought seriously about their field: But what I do requires genuine human judgment. Genuine relationship. Genuine care.
They are not wrong. They are simply describing something that is no longer, by itself, a sufficient defence.
The defence has two components — a capability argument and an economic argument — and only one of them is still holding.
The capability argument says AI cannot yet do what professionals do at their best. For the most complex, most relational, and most contextually demanding work, this remains largely true. An AI cannot replicate the depth of a twenty-year therapeutic relationship, the embodied judgment of an experienced surgeon mid-procedure, or the sustained attention a gifted teacher gives a struggling child over years of quiet observation. These are real distinctions. They matter.
The economic argument says that because AI cannot do these things, the humans who can will remain necessary and will be paid accordingly. This argument is already failing — not because the capability argument is wrong, but because the economic system does not pay for capability in the abstract. It pays for whatever delivers acceptable outcomes at the lowest cost. And the gap between what a good human professional can do at their best and what AI can do well enough to satisfy a health insurer, a government contract, or a stretched school board is narrowing faster than most professionals are willing to look at directly.
The honest question is not whether AI can match the best physician, the best teacher, or the best therapist. The honest question is how it compares to the full range of professionals actually delivering care and instruction at scale, on an average Tuesday, under ordinary resource constraints.
What the Evidence Actually Shows
The comparison most conspicuously absent from professional debates about AI is precisely that one.
In mental health, the data is quietly devastating for those who dismiss the risk. In a peer-reviewed empirical study published in 2025, licensed therapists were asked to distinguish between transcripts of human-to-human therapy and AI-to-human therapy (Tandfonline, 2025). They were accurate only 53.9% of the time — no better than chance — and on average rated the AI transcripts as higher quality than the human ones (Tandfonline, 2025). A first-of-its-kind randomised controlled trial found a Cohen’s d of approximately 0.8 for depression reduction over eight weeks from an AI chatbot — a large effect by clinical standards (PMC, 2025). A separately published clinical trial found a 51% reduction in depression symptoms and a 31% reduction in anxiety from AI-delivered therapy, outcomes comparable to traditional therapy for mild-to-moderate presentations (HeyNoah AI, 2025). A 2025 meta-analysis of 31 RCTs covering nearly 30,000 participants found small-to-moderate effects from AI chatbots in reducing mental distress across adolescents and young adults (JMIR, 2025).
The profession’s difficulty responding to this evidence is compounded by something it has never fully confronted: the near-absence of rigorous outcome standards. Mental health counselling measures success primarily through self-reported symptom scales and client satisfaction — not independent clinical benchmarks that would allow a clean comparison. In the absence of those standards, the claim that a human therapist will always be the better option is not an evidence-based statement. It is a professional preference, and one that the guild has a clear material interest in maintaining.
In medicine, the picture is equally uncomfortable. A 2025 systematic review and meta-analysis of 83 studies found no statistically significant difference between generative AI and non-expert physicians in diagnostic accuracy (PMC, 2025). AI was notably worse than expert physicians but comparable to non-experts — which describes most of the physicians most people actually see (PMC, 2025). In a widely discussed 2025 study, AI operating alone achieved 92% diagnostic accuracy, while physicians using AI scored only 76%, barely above their unaided baseline of 74% (Topol, 2025). Physicians were actively overriding correct AI assessments with their own initial impressions. By 2025, AI companies were attracting 55% of all health technology funding — far above their proportional share — and the capital is not following a bet that physicians will win this comparison (Bessemer Venture Partners, 2026).
In education, AI tutoring systems now achieve effect sizes of 0.4 to 0.8 standard deviations, approaching the effectiveness of small-group human tutoring (EduGenius, 2025). At five to twenty dollars a month against forty to eighty dollars an hour for human tutoring, the economics will make the access decision long before any policy debate concludes (EduGenius, 2025). AI tutors have been shown to improve student outcomes in lower-performing classrooms by nine percentage points — not by replacing good teachers but by filling the gap left by a shortage of them (HTL International School, 2026). That distinction is important: AI is not being introduced to replace the best of human education. It is being deployed to patch a system already failing from under-resourcing and structural neglect.
Where AI still clearly falls short: reading what is genuinely unsaid, holding someone through a real crisis, forming the kind of sustained relationship that changes a person over time, and doing the deep developmental work that depends on one human being present with another over years. That territory is real and worth defending. The argument is about how much of the actual daily work falls inside it versus outside it — and most professionals, if they are honest, know the ratio is not flattering.
The Smug Card and What It Hides
The professionals most inclined to dismiss this evidence are, by a well-documented pattern, the ones with least reason for confidence. This is not a provocation. It is the Dunning-Kruger dynamic applied at professional scale — and it is particularly acute in fields where the absence of rigorous external standards makes it easy to assume you are performing well simply because no one has told you otherwise.
When intelligent people read about higher-order thinking and its rarity, a predictable reflex appears: Of course I already think this way. My training demanded it. The physician who performs differential diagnosis believes this makes them a strong abstract reasoner. The therapist who holds a client’s pain without flinching believes this makes them relationally irreplaceable. The teacher who builds a classroom culture believes this is self-evidently beyond automation.
Sometimes they are right. Often, the confidence outpaces the reality — not because these people lack talent, but because almost no professional formation anywhere actually teaches the deeper capacities it claims to produce. Research on whether universities succeed in teaching critical thinking is discouraging: most institutions teach domain knowledge and procedural competence while describing it as something richer (Harvard Kennedy School, 2024; Frontiers in Education, 2026). ACEs are rare not because only special people can develop them, but because very few people have ever genuinely been required to (Thomson Reuters, 2026).
The professionals who will remain genuinely irreplaceable are not the ones who most loudly insist their work is different. They are the ones who have actually tested that assumption, know where their thinking runs deep and where it follows well-worn grooves, and can say precisely — not rhetorically — what they offer that AI cannot. That is a much smaller group than the population currently playing the “my work is special” card.
The impact of AI on the protected professions will not be felt only by professionals. It will be felt by the people they serve — and the collision of those two losses is the most human and painful part of this argument.
What is coming has a name in economics and sociology: frustrated preference. It describes what happens when what people want and what the system delivers diverge not because of ideology or policy failure, but because the economic logic of the system overrides individual preference at scale.
The teacher still wants to teach. They did not choose this work for the salary. They chose it because they believed — correctly — that the relationship between a skilled adult and a curious child is one of the most meaningful forms of human contact available. The parent still wants their child taught by that teacher, not out of sentimentality but because they can see, watching their child day to day, what a gifted human presence does in a classroom that a screen cannot. The system will fail them both. Not out of cruelty. Out of arithmetic.
It is not a single group going wanting. It is two groups, simultaneously, whose preferences align perfectly with each other and are still overridden by system economics.
The therapist who entered the profession to sit with people in their darkest moments will find their caseload restructured around a government contract with an AI platform at a fraction of the per-session cost. The patient who wants continuity of care with a physician who knows their history will receive a twelve-minute AI-assisted triage followed by a prescription generated by a model. The retired teacher sitting on a school board, watching class sizes climb toward forty and AI terminals replace the colleagues she worked beside for thirty years, will understand what has happened. The thirty-year-old who has never experienced anything different will not.
The cruelest aspect of this pattern is the vocation. The professionals most likely to be restructured out are, overwhelmingly, those who chose their field because it felt humanly essential and economically durable. They did not go in for the money. They went in because they believed what they did was irreducibly human. The loss, when it comes, is therefore not merely economic. It is the loss of a calling. Vocation, unlike a job, does not have a market substitute.
And there is a third party in this dynamic that receives almost no attention: the students, patients, and clients who will receive AI-primary care and never know what they missed.
A child who goes through school with AI as the primary instructor, and occasional human contact as the supplement, will not experience their education as a deprivation. They will have no reference point for what a deeply engaged human teacher offers over years of relationship. The loss is real. It is consequential. And it is entirely invisible to those bearing it — which is perhaps the most troubling part of all.
The child sitting in front of the screen will not know what was taken from them, because they will have nothing to compare it to.
Everything above involves capability and preference. What actually determines the outcome is money.
Public healthcare budgets across the developed world are already under structural pressure before AI displacement has meaningfully contracted the tax base. Ontario’s healthcare spending is growing at 0.5% in real per-capita terms before contracting the following year, and the province’s finance minister has described the trajectory as unsustainable (Policy Alternatives, 2026; LinkedIn, 2026). The NHS faces the same arithmetic. Every developed country’s public health system is caught between rising demand — driven partly by an aging population and partly by the secondary health consequences of economic disruption — and a tax base being eroded by the same AI displacement described in Part Three.
Under that pressure, AI is not a choice between human care and its substitute. It becomes the mechanism by which governments deliver the minimum viable version of care to the maximum number of people. The mental health system is first in line for this transformation — precisely because it already has the least rigorous outcome standards, the longest waiting times, and the largest gap between what people need and what is currently available. An AI platform that delivers structured CBT protocols to ten thousand people simultaneously at a fraction of the per-session cost of a human clinician will be funded, not because anyone believes it matches a skilled therapist, but because it is radically better than a two-year waiting list.
Education follows the same logic. When any provincial or state government faces a choice between maintaining thirty teachers in a school and deploying AI-primary learning systems supervised by fewer staff, the pressure from treasury will eventually prevail — particularly when outcome data are ambiguous enough to allow a minister to claim that “learning results are maintained.” The teacher still wants to teach. The system will find ways, gradual and institutional, to need fewer of them.
The result of this process is not the wholesale replacement of human professionals across the board. It is the emergence of two parallel systems that deliver, under the same names, fundamentally different experiences.
The first track is the augmented human. A consultation with a physician who reviewed AI diagnostic summaries before you arrived and who has thirty minutes instead of twelve. A therapist with a manageable caseload using AI-generated session tracking to focus her full attention on the relationship. A classroom of twenty students, with AI handling personalised practice and assessment, and a teacher who actually knows every child in the room. This track will exist. It will be excellent. It will be expensive. It will be available to those who can pay privately, access elite institutions, or live in well-resourced jurisdictions.
The second track is the clinical substitute. An AI triage system generating a differential and a prescription. A structured AI therapy protocol that reduces symptom scores on a validated scale. A classroom of forty-two supervised by an AI learning platform and a teaching assistant whose primary function is behaviour management. This track will also exist. It will be described as providing evidence-based care. It will be what is available to the majority.
The people most likely to need mental health support, educational guidance, and regular medical care in the coming decade are precisely those who will have been displaced by AI in the labour market — people arriving at the health and education systems with less money than before, at exactly the moment those systems are being reorganised around cheaper delivery.
Economics, not preference, will decide which track people land on. The irony is not subtle.
Guilds, Inheritance, and Who Gets In
One pattern that will not be new, but will be sharply renewed, is the concentration of surviving professional roles in families, networks, and credentialing systems that function as guilds.
When roles become scarce and valuable, those who hold them erect barriers. This is among the oldest patterns in economic life. Medieval craft guilds controlled entry through apprenticeships that privileged the children of members. Modern professional bodies already behave as soft guilds: licensing requirements, controlled numbers, accreditation systems, and informal networks that route the best opportunities to the socially connected.
As AI compresses the available space for human professional work, surviving high-status roles will harden in exactly this direction. The children of physicians, lawyers, senior academics, and elite consultants will receive the cues, coaching, introductions, and quiet preferences that matter far more than formal qualification scores. The credential will remain officially open to anyone who qualifies. The pathway to the credential will become progressively less navigable for those without the right starting position.
This is not a conspiracy. It is a predictable response to scarcity. When there are fewer positions, those who hold them protect what they have and pass it to those they love. AI does not create this pattern. By compressing the available space, it sharpens it.
There is a developmental dimension to this argument that rarely enters the conversation about AI and the professions, and it matters considerably for how we think about who will be most affected.
The cohort who will bear the greatest weight of these changes — people currently under thirty — is the cohort already shaped by fifteen years of social media exposure, bubble-wrap parenting, and an educational culture that optimised for anxiety reduction rather than resilience building.
Prefrontal cortex myelination — the neurological substrate for genuine abstract reasoning, complex judgment, and the emotional regulation that real ACE development requires — is not complete in females until approximately eighteen to twenty-two and in males until approximately twenty to twenty-six (PMC, 2013). Adolescence is when the hardware for complex, abstract thinking comes online. But unlike Piaget’s earlier developmental transitions, where new cognitive capacity emerges more or less automatically as the brain matures, the arrival of this hardware does not produce the corresponding software by itself. The capacities the newly matured brain is capable of must be explicitly taught — and the research on whether educational institutions actually do this is not encouraging (Frontiers in Education, 2026; Harvard Kennedy School, 2024).
The young people entering this labour market have grown up in an environment that has, in many measurable ways, worked against this development. Platforms designed around immediate reward, low tolerance for disagreement, and algorithmic smoothing of discomfort have interacted with still-maturing self-regulation systems. Parenting norms have shifted toward risk-avoidance and protection. The unsupervised time, real-world conflict, boredom, and problem-solving that previous generations navigated as a matter of course have been progressively reduced. The lower resilience, higher anxiety, and reduced tolerance for ambiguity that experienced educators and employers now consistently report in younger cohorts are not character failings (NPR, 2026; EdWeek, 2025). They are predictable developmental outcomes of a specific kind of environment, arriving for a generation that is about to need resilience more than any generation in recent memory.
AI is not dropping into a psychologically robust, well-prepared population. It is arriving for a cohort already carrying a significant developmental load — at exactly the moment the labour market will most reward the deep capacities that load has made harder to build.
There is a group this article has not yet named directly, and they may be the ones for whom the argument is most urgent: the people who are, right now, in the final years of professional formation — the medical student about to complete their residency, the education graduate doing their practicum, the counselling student accumulating supervised hours toward certification.
They have done everything right. They identified a field that felt meaningful, stable, and humanly essential. They made the sacrifices that professional training demands — the years, the debt, the deferred income, the sustained effort. They are arriving at the threshold of the profession they chose at exactly the moment when the floor of that profession is shifting beneath it.
Unlike the established professional with twenty years of relationships, reputation, and embedded institutional knowledge, the new entrant has none of those buffers. They are competing for a shrinking number of entry-level positions against both AI systems doing the routine work and experienced professionals displaced from higher rungs who are now competing for the same junior roles. The established teacher with a permanent contract and a known record in a community is difficult to remove even as budgets tighten. The newly certified teacher entering a system that is quietly reducing its permanent headcount faces a much colder arithmetic.
The cruelty here is particular. The established professional has at least had the career they signed up for. The new entrant is discovering, as they cross the threshold, that the profession they trained for is already narrowing behind them. The therapist who completes her masters in counselling in 2026 enters a mental health system that is actively funding AI platform contracts precisely because human practitioners cannot be afforded at the scale required. The physician completing residency enters a healthcare system that is deploying AI diagnostics not to augment its physicians but to reduce the number it needs to hire. The teacher emerging from their education degree enters a school system where permanent positions are scarce and the fastest-growing role is “AI learning facilitator” — a title that pays less, carries no professional status, and requires primarily that its holder supervise students working with software.
None of this means these people should not have trained as they did. It means they need to understand, clearly and without comfort, what they are entering — and they need to build, from the first day of their practice, the kind of capability that cannot be automated, institutionally defunded, or restructured away. The ACEs this series has been building toward are not a luxury for the professionally comfortable. For the ones just arriving, they are the most urgent item on the agenda.
The institutions that trained them will not say this. They have every incentive to keep enrolments up and every disincentive to tell applicants that the profession they are about to enter looks different from the one described in the brochure. The honest version of what these new professionals need to hear is that the credential is necessary but no longer sufficient; that the relationships and trust they build from their first day of practice matter more than they have ever mattered before; and that the professional who survives the coming decade will be the one who is known, trusted, and irreplaceable to specific people — not the one who holds the strongest institutional position in a system whose institutions are under structural pressure from every direction.
What the Protected Professions Should Actually Do
This article has not been written to demoralise. It has been written because the professionals who most need to prepare are, by the very logic of their conviction that they are safe, the least likely to be doing so.
The honest path forward has three parts.
The first is to look clearly at what portion of your actual daily work is AI-accessible — not the best version of what you do in your best moments with your most complex cases, but the average version. The routine consultations. The standard lesson plans. The structured therapy protocols. That portion is larger than most professionals are willing to estimate, and the resistance to estimating it honestly is itself a signal worth paying attention to.
The second is to develop — explicitly and deliberately — the capacities that remain genuinely irreplaceable: the ability to read what is unsaid, to hold complexity without collapsing it into a protocol, to stay present with a person in real distress without reaching for the nearest script. These are ACEs in their most specific professional form. They are not developed by doing the same work for twenty years on autopilot. They are developed by deliberately practising the thinking that the work requires, past the point where accumulated habit takes over.
The third is to stop waiting for the institution to protect you. In most plausible futures, institutional protection is partial at best. The professionals who navigate what is coming will not be those with the most secure positions or the strongest credentials. They will be those who have built, genuinely and independently, the kind of capability that another person — a parent, a patient, a student — would seek out even if no institution was pointing the way.
The teacher still wants to teach. There will still be parents who will find that teacher, wherever she is. The question is whether she has built enough — in herself, not in her position — to be worth finding.
Four Scenarios for the Protected Professions
As with every group examined in this series, the future of physicians, teachers, and therapists is not fixed. It will be shaped by the speed of disruption, the capacity of institutions to adapt, and the choices made by governments, professional bodies, and individuals in the months and years immediately ahead. The four scenarios introduced in Part Four of this series frame the range of what is plausible.
Scenario One: Managed Transition is the least likely but worth naming clearly, because it describes what a deliberate and honest response would actually require. Governments fund public education and healthcare at levels that preserve human professional presence. AI is integrated as a tool under professional oversight rather than as a cost-cutting substitute. Professional formation is redesigned to explicitly develop the ACEs that distinguish genuinely irreplaceable human practice from routinised competence. Outcome standards in mental health are finally made rigorous enough to evaluate AI and human care on equal terms. The teacher who wants to teach, and the parent who wants a teacher, get what they both need — not universally, and not without significant transition costs, but as a systemic commitment rather than a premium luxury.
Scenario Two: Bifurcated Society is more likely, and is already partly visible. Two parallel systems operate under the same names. The premium track — available to those who can pay privately or access elite institutions — delivers human professionals augmented by AI. The universal track delivers AI-primary care and instruction, supervised by reduced human staff. Professional guilds harden. Surviving high-status roles cluster in families and networks. The children of those already inside the fortress receive the coaching, connections, and quiet preferences that make access to those roles look, from the outside, like meritocracy. The majority of professionals displaced from public systems discover that the private alternative has no room for them. Some find their way into informal and community-based arrangements — a partial barter economy of care, trust, and local knowledge — but only those who have something genuinely concrete to offer.
Scenario Three: Political Rupture is, given current trajectories, the scenario this series judges most likely to actually occur. Institutions fail faster than they can be reformed. Governments fund AI systems at scale because no other option is fiscally credible. The distinction between “AI-primary care” and “human care” stops being debated once the institutions that would have debated it have lost legitimacy. The teacher who wants to teach, and the parent who wants a teacher, join a long list of people whose preferences the system has simply stopped being able to honour — not from indifference, but from arithmetic that got away from everyone. In this scenario, the professional’s vocation is the first casualty, years before the credential disappears.
Scenario Four: Something We Have Not Named is the most historically grounded position and the least describable. Every civilisational transition has produced arrangements unimaginable from within the prior order. It is genuinely possible that new forms of human professional practice emerge — embedded in communities rather than institutions, organised around trust and relationship rather than insurance codes and government contracts, and valued in ways the current economy has no mechanism to price. A teacher known by name in a neighbourhood. A physician trusted by a community and compensated in whatever medium survives the transition. A therapist who sits with people in genuine distress because that is what she does, not because a reimbursement code has been approved. Whether these arrangements are better or worse than what preceded them will depend, almost entirely, on the ACEs and social capital of the people building them.
In every scenario, the professionals who fare best are not those who held the most secure institutional positions or accumulated the most impressive credentials. They are those who built, genuinely and independently, the capacity to be useful to another human being in ways no model can fully replicate — and who did so before the institution stopped pointing the way.
References
Bessemer Venture Partners. (2026, January 19). State of health AI 2026. https://www.bvp.com/atlas/state-of-health-ai-2026
EduGenius. (2025, August 14). How AI tutors compare to human tutors. https://www.edugenius.app/blog/ai-tutors-compare-human-tutors-research
Frontiers in Education. (2026, February 9). From implicit to explicit: Overcoming common barriers to critical thinking instruction. https://www.frontiersin.org/journals/education/articles/10.3389/feduc.2026.1689765/full
Harvard Kennedy School. (2024, July 7). Professors say they teach critical thinking. But is that what students are learning? https://www.hks.harvard.edu/faculty-research/policy-topics/education-training-labor/professors-say-they-teach-critical-thinking
HeyNoah AI. (2025, September 30). AI therapy vs human therapy: When each approach works best. https://heynoah.ai/blog/ai-therapy-vs-human-therapy-when-each-approach-works-best
HTL International School. (2026, February 3). Can AI replace teachers? The truth about machine vs. human learning in 2026. https://www.htlinternationalschool.com/2026/02/04/can-ai-replace-teachers-the-truth-about-machine-vs-human-learning-in-2026/
JMIR Mental Health. (2025, November 25). The effectiveness of AI chatbots in alleviating mental health symptoms. https://www.jmir.org/2025/1/e79850
LinkedIn. (2026, February 16). Ontario’s unsustainable healthcare spending: AI solutions for fiscal sustainability. https://www.linkedin.com/posts/liam-gill_ontario-finance-minister-says-rate-of-health-care-activity-7429530567169277952-w0Ym
NPR. (2026, January 14). The risks of AI in schools outweigh the benefits, report says. https://www.npr.org/2026/01/14/nx-s1-5674741/ai-schools-education
Policy Alternatives. (2026, March 26). Ontario is underfunding public health care — and private health care is the beneficiary. https://www.policyalternatives.ca/news-research/ontario-is-underfunding-public-health-care-and-private-health-care-is-the-benefi
PMC — National Library of Medicine. (2013). Maturation of the adolescent brain. https://pmc.ncbi.nlm.nih.gov/articles/PMC3621648/
PMC — National Library of Medicine. (2025a, March 21). A systematic review and meta-analysis of diagnostic accuracy: LLMs vs. clinical professionals. https://pmc.ncbi.nlm.nih.gov/articles/PMC11929846/
PMC — National Library of Medicine. (2025b). When can AI psychotherapy be considered comparable to human psychotherapy? https://pmc.ncbi.nlm.nih.gov/articles/PMC12504474/
Tandfonline. (2025). Human-human vs human-AI therapy: An empirical study. https://www.tandfonline.com/doi/full/10.1080/10447318.2024.2385001
Thomson Reuters. (2026, February 8). 2026 AI in professional services report. https://www.thomsonreuters.com/en/reports/2026-ai-in-professional-services-report
Topol, E. (2025, February 1). When doctors with AI are outperformed by AI alone. https://erictopol.substack.com/p/when-doctors-with-ai-are-outperformed
EdWeek. (2025, October 7). Rising use of AI in schools comes with big downsides for students. https://www.edweek.org/technology/rising-use-of-ai-in-schools-comes-with-big-downsides-for-students/2025/10