Peron / MIT Sloan Management Review — AI for Interoperability in Health Care (2026-05-31)
On today’s episode, Philips’s chief medical officer Carla Goulart Peron shares how artificial intelligence is reshaping health care — not by replacing clinicians but by expanding access, improving diagnostics, and freeing doctors to focus more time on patients. Drawing on her experience practicing medicine in Brazil’s strained public health system, she explains how technologies like AI-assisted imaging and remote collaboration can bridge critical gaps in care. Carla also explores the challenges of trust, bias, interoperability, and women’s health data in the next era of AI-enabled medicine. She offers a grounded, global perspective on how technology can make health care more human.
(Channel description, MIT Sloan Management Review YouTube — Me, Myself, and AI podcast Season 13.)
A ~31-minute long-form podcast episode published 31 May 2026 by MIT Sloan Management Review on its YouTube channel as part of the Me, Myself, and AI podcast (Season 13). Host Sam Ransbotham (Professor of Analytics at Boston College; MIT SMR research lead since 2014). Guest Dr. Carla Goulart Peron — Chief Medical Officer at Philips; physician by training; before Philips, VP and CMO for surgical innovations and robotics at Medtronic. Peron trained and practiced medicine in São Paulo, Brazil, where she worked in the public universal-healthcare system in the morning and in private hospitals in the afternoon — sometimes the same day — giving her a resource-scarcity + resource-abundance same-day comparative vantage on healthcare that she carries into her current Philips role.
This is the wiki’s first dedicated healthcare-AI source + the first Me, Myself, and AI podcast ingest (the prior MIT SMR sources are all Ransbotham-and-others’ written-research outputs — Augmented Learners (May 2026), Chatterjee on agent harness (May 2026), Kiron & Schrage on compound benefits (May 2026) — none from the podcast). Sits adjacent to Luminai (Apr 2026) at startup-altitude on hospital ops — together the two sources form the wiki’s first two-altitude healthcare-AI cluster: Peron at clinical-practice altitude, Luminai at hospital-administrative-operations altitude.
TL;DR
Six substantive contributions.
1. The Philips SmartArt cardiac-MRI worked example — recent FDA clearance for an AI-driven one-click automation that “plans all setups that drives how the cardiac imaging needs to be captured … in 30 seconds”. Peron explains the structural lift: a cardiac-MRI exam normally requires a technician who knows “exactly how he or she should be positioning you into the MR table, in which angle, if you are tall or short, if you are someone that is big or small, if it’s a kid or if it’s a female or a male — there’s so many different data points that a technician needs to understand in order to capture the right level of imaging.” The before/after: ~15 minutes of expert-technician setup → 30 seconds of one-click AI setup. Three operational consequences: (a) ~4× throughput on an expensive machine that otherwise sits underused; (b) reduced dependency on highly-trained technicians; (c) elimination of the patient-callback-because-views-missed loop (“you’ll go through the exam, send the images to the radiologist, and they will say — Oh, you need to call this patient back because we are missing one or two views. And with something like that, this doesn’t happen.”). The wiki’s first concrete vendor-side AI-imaging-automation worked example with explicit before/after timing.
2. The radiology-displacement-that-didn’t-happen historical anchor + the radiologist-training-gap follow-up question. Peron: “A decade ago people were saying — Oh gosh, we’re never going to have radiologists again because the machines are going to do everything. And that narrative has really not played out at all.” Her positive reframe: AI takes the low-value radiology work — reviewing imprecisely-captured images, doing reports, reassessing normal images — and frees radiologists for the unique-judgment work. The unresolved follow-up question (raised at a recent radiologist-society meeting Peron attended, no answer): “What if we could have AI defining all normal images and then radiologists will be looking at only abnormal? … How are we going to get the radiologist trained in what is abnormal if they are not going to be seeing normal?” The wiki’s first clean clinical-domain instance of the deskilling concern: if AI handles the exemplar-of-normal training-data class, downstream specialists lose the calibration-against-normal they previously developed through volume practice. Convergent with the broader ai-deskilling concept page’s task-composition-shift-within-retained-jobs mechanism, sharpened by being applied to a profession with a clean normal-vs-abnormal binary that maps directly to a training-pipeline gap.
3. The Philips Future Health Index trust gap. Peron cites Philips’s own annual survey: 79% of healthcare professionals optimistic about AI vs ~50% of patients worried about reduced face-to-face time. The reconciliation argument she offers: clinician concern is data-driven (bias in training data, validation, can-I-trust-fully-AI, how-much-do-I-need-to-review); patient concern is experiential (will-I-see-my-doctor-less; the AI-buzzword-misperception). Peron’s mechanism for closing the gap: AI is freeing physician time for “the empathy piece, to the touch, to that one-on-one I-to-eye”. The empirical wrinkle she names: “There are some studies already out there that show if you are talking to a real physician or to an AI version of that physician, sometimes the AI can learn how to be more empathetic than the physician.” + the research finding “individuals who trust AI are twice as likely to use it regularly” — both are reported as known findings without specific paper citations.
4. The patient word critique + precise-medicine reframe. Peron’s etymological/operational reading: “It’s almost like you stay there, be patient, and wait until somebody tells you what to do — while now with AI enabling interoperability, data points, you are giving more visibility to the overall health and also what are the options that those patients may have in front of them.” The frame she introduces: AI shifts patients from the be-patient-and-wait default to the precise-medicine default — “if patients will be able to be offered one, two, three potential treatment pathways with pros and cons and the ability to choose … AI may enable those patients to kind of make more informed decisions at least.” This is the wiki’s first articulation of AI-as-patient-agency-extender in healthcare.
5. The nine-German-women postpartum-blood-loss anchor + the gender-bias-in-clinical-protocols worked example. Peron’s two-years-ago WEF anecdote: she was on a women’s-health panel discussing the standard for normal postpartum blood loss; an audience member asked “How was that established?” Peron paused — she had never asked. The answer: the global standard was set based on nine German women and exported worldwide. India recalibrated to ~300mL given smaller average body sizes. Peron’s frame: “I think technology, that interconnectivity, that not only the fact that we’re going into automation, but that now AI can analyse such a big data set so quickly, can really improve the way we are practicing medicine.” Ransbotham extends the point: “I can also imagine a very simple job for agents would be — hey, go through all of our clinical practices in every area and find the root study for that and assess how that plays out. I would feel like your nine-person-in-Germany sample should rise pretty quickly to the top of that list.” The wiki’s clearest single worked example of AI-as-clinical-protocol-bias-auditor at population scale. Direct convergence with the broader responsible-ai concern about gender-bias in clinical training data — Peron supplies the operational what-an-agent-could-do prescription.
The same theme extends to women’s cardiac health: cardiac causes are highest mortality among women; women have longer waiting times for diagnostic because symptoms present differently; most cardiac protocols were designed based on male-only trial data. Peron names two specific physiological differences AI algorithms can incorporate — heart position (slightly different in female vs male chest, affects MR-imaging capture) and rhythm patterns (the female cardiac rhythm has a slightly-different pattern that algorithms need to calibrate against). A second clean AI-corrects-embedded-clinical-bias worked example.
6. Interoperability as Peron’s single-pick global AI capability + three named barriers. When Ransbotham asks “If you could pick one AI capability to deploy globally that would make the biggest difference, what would you push out to the world?” Peron’s answer: “Interoperability. That is going to change completely the way we practice medicine because today we’re very much closed or restricted to the health care system that you are operating. The ability to see the patients longitudinally without those barriers I personally believe is going to change outcomes significantly.” Three named barriers to interoperability she surfaces: (a) data quality + standardization from the get-go (“if you don’t have some level of standardization, it’s very difficult to think about interoperability — that’s the first piece which is science”); (b) reimbursement systems (current code-coverage and incentive structures don’t reward tech that reduces length-of-stay — “if that’s not the incentive from the healthcare system perspective, this is not going to happen”); (c) regulation must evolve (“the type of regulation that brought us here is not going to take us to the future because the future is very different than the one we are playing today”).
Recurring Peron frame: “AI is here to add, not to take over”
The phrase recurs explicitly in the radiology-displacement segment. The context that makes the frame work differently from a generic optimist-framing: healthcare has a structural undersupply of clinicians and imaging capacity globally — “I personally believe that there is such a big gap out there in access to care that as we incorporate technology we’ll be able to do more with the same not with less, right?” — and “we cannot fool ourselves that reality also exists in the US and in Europe and in Asia. We have deserts, right? We have areas where people don’t have any access to care.” The reframe matters because the binding economic constraint is supply, not labor cost.
Linked entities and concepts
Already-promoted entities referenced: Sam Ransbotham (host; source-count bump +1), MIT Sloan Management Review (channel/publisher; source-count bump +1).
Concept pages this source informs (Process step 6 targets): automation-vs-augmentation (Peron’s AI-adds-not-takes-over + radiology-displacement-historical-anchor + SmartArt 15-min-to-30-sec worked example), ai-deskilling (the radiologist-trained-on-normal question is a clean clinical instance), responsible-ai (the nine-German-women postpartum-blood-loss anchor + women’s-cardiac-health gap + clinical-protocol-bias as an audit task an agent could perform).
Dangling (single-source mentions, deferred per the second-source promotion rule): Carla Goulart Peron (the subject of the interview; first appearance — the entire content of this source is her practitioner thinking; the spirit of the rule is preserved by the substantial body coverage on this page), Philips (referenced multiple times; not yet an entity page — likely candidate for promotion when a second healthcare-AI source mentions Philips), Medtronic (Peron’s pre-Philips employer; one-mention), SmartArt (Philips product; first wiki mention; deferred), Future Health Index (Philips’s annual survey; first mention), Me, Myself, and AI (the podcast; first ingest under MIT SMR author: from this show specifically — note that the channel-as-author convention means the show-name is not in author:), Bernard Hampton (named in the episode’s closing as the next-week guest; one-mention forward-reference), Josh from CVS (Ransbotham’s prior guest reference; one-mention).
W&W cells touched (10 cells, broader than typical for a single interview — reflects the cross-cutting nature of the healthcare-AI question across the W&W process model):
digital-sensing/digital-mindset-crafting— Peron’s Brazil-public-private-same-day vantage as the cognitive substrate for her healthcare-AI strategy at Philips.digital-sensing/digital-scenario-planning— the if-AI-flags-only-abnormal, how-do-radiologists-train-on-normal? unresolved question is scenario-style.digital-seizing/balancing-digital-portfolios— Philips’s spread across imaging + interventional + ICU monitoring + remote monitoring + AI across all = portfolio-balancing across the health-tech adjacencies.digital-seizing/rapid-prototyping— the SmartArt iteration from product-team work through to FDA-clearance is the rapid-prototyping cell at vendor altitude.digital-transforming/improving-digital-maturity— Philips’s AI-supporting-all-the-areas-of-care-we-have organisational framing is the digital-maturity reading.digital-transforming/navigating-innovation-ecosystems— Peron’s role description (ensuring engineer-developed innovations in partnership with hospitals and physicians meet regulatory and reimbursement requirements) is the canonical innovation-ecosystem-navigation cell.strategic-renewal/business-model— the interoperability-as-the-single-pick + reimbursement-as-barrier framing is a business-model-renewal claim about how healthcare value-flow should work.strategic-renewal/collaborative-approach— engineers + hospitals + physicians + regulators as the named collaboration loop.contextual/external-triggers— the FDA-clearance event for SmartArt + India’s recalibration of postpartum-blood-loss-standard are external-trigger anchors.contextual/internal-enablers— Peron’s CMO-vantage “if you don’t have some level of standardization, it’s very difficult to think about interoperability” names data-standardization as the internal-enabler that interoperability rests on.
Roles override (roles: frontmatter explicit, overriding the W&W defaults): cto, cdo, cmo, transformation-lead, product-manager, rd-director. Healthcare-domain ingest with Chief Medical Officer as the canonical role; cmo here means Chief Medical Officer (not Chief Marketing Officer — the wiki’s role-vocabulary cmo slug is overloaded in clinical contexts and the body resolves it). The override is justified because the strategic-renewal/collaborative-approach default (cio, cdo, rd-director) misses the CMO/CTO/transformation-lead cluster that actually owns clinical-AI strategy in healthcare-vendor settings.