Evis Sala

Illustration: The Aiomic Age. Created with AI.

The conversation

Evis Sala speaks with Karim Galzahr in Tirana about what it takes to put AI to work in healthcare. A radiologist with experience in research, clinical practice and health policy, she was Albania’s Minister of Health and Social Welfare at the time of the interview. The government’s account of her appointment describes her work in Cambridge, New York and Rome.

Both parts of this conversation were first published on YouTube on 27 August 2026. The main interview above concentrates on AI and implementation. The shorter companion conversation appears below.

Disclosure: at 41:28, Galzahr states that he and Sala both have an involvement in a company working on AI-enabled radiology. The company is not named in the recording.

Our reading

Sala’s most useful question comes at 10:26: did the technology change the patient’s pathway through care? She asks whether people reach the right investigation and treatment at the right time, and whether clinicians can combine information across specialties. Our reading is that this should shape the evaluation from the start. Accuracy on a diagnostic task matters; the organisation also needs to establish what happens to the people who enter its service.

That changes the implementation problem. At 34:17, Sala describes innovation and implementation pulling apart. At 38:17, she argues that countries need enough equipment, integration and working clinical services to make further innovation possible. The practical question for a buyer is which part of the care process a new tool will improve, who will use its output and what must change around it. Her account is an argument for investing in those conditions; it is not evidence that a particular programme has already delivered the promised results.

The same concern appears in her account of medical education at 21:13. Sala worries that access to generated answers can change how students learn to reason. She describes teaching from a clinical problem and asking students to work through it. The implication for AI training is concrete: staff need opportunities to explain a decision, identify missing evidence and challenge a recommendation. Knowing how to operate the tool is only part of the job.

What to watch

The test of this approach is what happens across the whole service: whether referrals carry the information clinicians need, whether patients complete appropriate follow-up, and whether the time to suitable care improves. Benefits also need to reach people outside the best-equipped centres. If a system improves isolated task scores while those outcomes remain unchanged, its implementation case remains incomplete.

Chapters in the AI interview

• 00:22 — Why implementation is the subject
• 01:59 — Setting priorities and choosing what to finish
• 07:25 — Clinical knowledge and AI decision support
• 09:45 — Judging AI by the patient pathway
• 14:32 — Coordinating expertise and human judgement
• 18:57 — Radiology and the problem AI is meant to solve
• 21:13 — Medical education and critical thinking
• 29:08 — Albania’s plans for AI and digital health
• 34:17 — The gap between innovation and implementation
• 37:35 — Building the conditions for clinical innovation
• 41:28 — Disclosure and avoiding unnecessary follow-up
• 42:28 — Keeping patients on a coherent pathway
• 46:31 — Collaboration across institutions

Source notes

The radiology prediction recalled at 18:57 dates to Geoffrey Hinton’s 2016 remarks. The recording gives 2015. The examples of biopsy timing in the interview illustrate Sala’s reasoning about prioritisation; they are not trial results or universal treatment intervals.

The screening discussion near 44:52 needs a specific scope. The UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS) found that the screening strategies it tested did not reduce deaths from ovarian and tubal cancer, despite a shift towards earlier-stage diagnoses in one screening group. That result concerns those screening strategies and population; it does not establish that earlier diagnosis has no value across cancers.

The ministerial title describes Sala’s role when the interview was recorded. The government’s current ministerial profile lists Klodiana Spahiu as holding the post since September 2026.

Companion conversation: mental health and digital pressure

In the second part of the same interview, Sala and Galzahr turn to young people’s prospects, mental wellbeing and online abuse. Sala describes a proposed response to digital violence and the connection between health and social welfare. This companion adds context to the wider social consequences of technological change. The initiative is discussed as a proposal at the time of recording.

Watch the companion conversation on YouTube