Approximately Correct: An AI Podcast from Amii

The episode delivers concrete applications of AI in healthcare (breast cancer screening, liver cancer detection, drug testing via Mendelian randomization) with specific mechanism explanations, though much of the framing remains accessible rather than deeply technical. The guest covers real problems and solutions, but the density peaks around specific examples (Medical Segmentation Decathlon, photon-counting CT, UK Biobank analysis) rather than maintaining insight throughout.

“there's been great clinical trials, randomized clinical trials, showing that AI, uh, screening tools for breast cancer are better than humans”

“we showed that you could replicate. I think it was 27 cardiovascular trials of these different genes turned on and off and targets made for these genes in UK Biobank data”

The episode presents moderately fresh angles - particularly the Mendelian randomization application to drug testing and the emphasis on clinician-centered AI deployment - but relies heavily on established frameworks (precision medicine, radiomics, bias in datasets). The guest's clinician-collaboration philosophy is distinct but not startlingly novel; the core insights around AI augmentation vs. replacement are well-circulated in healthcare AI discourse.

“if humans can't actually look at all the images to begin with, we're not even really talking about replacing humans. Right. We're talking about making sure that all the images get looked at”

“clinicians that don't know how to use AI are the ones that are going to be disadvantaged, but that it's not going to replace them because there is a special sauce to what humans do clinically”

Amber Simpson is a demonstrably serious practitioner: Canada CIFAR AI Chair, former faculty at Memorial Sloan Kettering Cancer Center, actively leading research teams, and publishing work on medical imaging and genomics. She has direct clinical collaborations and has shipped tangible outputs (Medical Segmentation Decathlon dataset), avoiding the trap of being a pure theorist or media-circuit guest. Her framing around clinician engagement and data governance reflects real operational experience.

“my first faculty position was at a big cancer institute in the US So a lot of my experiences around working with cancer data”

“we released the largest data set across 10 different types of organs and tumors. Um, this was called the Medical Segmentation Decathlon”

The episode includes concrete examples: breast cancer screening trials, liver/pancreas imaging work, 27 cardiovascular trial replications via Mendelian randomization in UK Biobank, photon-counting CT as an FDA innovation, the Medical Segmentation Decathlon (2018) challenge, Memorial Sloan Kettering deployment, and Alberta's EPIC rollout. However, specific metrics are sparse - no sensitivity/specificity numbers, ROI figures, or patient outcome deltas. Qualitative specificity is strong, quantitative specificity is weaker.

“There's been great clinical trials, randomized clinical trials, showing that AI, uh, screening tools for breast cancer are better than humans”

“we released the largest data set across 10 different types of organs and tumors. Um, this was called the Medical Segmentation Decathlon”

The hosts ask open-ended, substantive questions and occasionally probe deeper (e.g., generalization beyond Western data, data-sharing barriers, clinician attitudes), but rarely push back or challenge claims directly. Follow-ups are mostly clarifications rather than skeptical interrogations. The conversation feels collegial and collaborative but lacks the tension and rigor of adversarial questioning. A few softball moments (e.g., the liver-as-barometer metaphor) pass without deeper interrogation.

“And is that, like, what you're most excited about right now, those sorts of changes, or is there anything else”

“Yeah, that's a great question. And it's a bit of a difficult one to unpack”