Meta: course=ai-clinical · module=1 · lesson=1.1 · ~70 min · keywords: clinical AI, TB CAD, diagnostic gap, augmentation, WHO recommendation, leapfrog, retinal screening, capability ladder Objectives:
- Name the AI application families with the strongest evidence for African clinical use and rank them by maturity.
- Explain why African AI economics make "augmentation, not replacement" the design posture.
- Separate evidence-graded capability from hype using the book's capability ladder.
The first African digital leapfrog was infrastructural — past the landline, straight to mobile. The second is cognitive: past the analog scarcity of expertise toward systems in which AI extends the reach of every scarce specialist. As a clinician you stand exactly where that leapfrog lands. But this course is written against hype as much as for ambition. AI arrives carrying both real, evidence-graded capability and documented risk — algorithmic bias trained on other populations, governance gaps, and the temptation to bolt advanced analytics onto unfinished foundations of data quality, workforce, and trust (Ch7 §7.2).
The most consequential near-term applications attack Africa's diagnostic gap directly, and the flagship evidence base is tuberculosis. Computer-aided detection (CAD) software reading digital chest X-rays now performs comparably to human readers for triage, and the WHO formally recommends CAD as an alternative to human interpretation for TB screening and triage in people aged fifteen and over — the first WHO endorsement of AI replacing a human diagnostic task, confirmed by independent evaluation in high-burden settings (Qin et al., 2021). The operational meaning is profound: a district hospital without a radiologist can screen at radiologist-level quality.
The same pattern — narrow task, scarce specialist, image- or signal-based input — defines the credible frontier: diabetic retinopathy screening at primary care, cervical cancer screening support, dermatology triage, ultrasound guidance for non-specialist operators, and ECG interpretation. Beyond imaging, clinical decision support is being upgraded from static protocols to data-driven guidance; the WHO SMART guidelines initiative makes machine-readable national protocols deployable into digital systems at software speed. Large language models extend the horizon to conversational support — and the WHO's 2024 guidance on large multi-modal models catalogues the failure modes (fabrication, bias, automation over-reliance) that make governed deployment, not consumer enthusiasm, the right pathway.
The economics of African AI differ fundamentally from rich-world debates about automation displacing professionals. Where the baseline is one radiologist for millions of people, AI does not threaten the workforce — it is workforce, capacity that otherwise would not exist. The design consequence is augmentation architecture: AI as triage and second reader within clinical governance, with escalation to humans, audit trails, and a new workforce competency — critical supervision of algorithmic colleagues. The systems that succeed make the frontline worker more capable and more confident.
The honest constraint is the data foundation. Models trained on incomplete or biased routine data automate the blind spots of the system that produced them. So the book insists on a capability ladder, not a procurement: data quality and use culture first, statistical capacity second, machine learning third. A critical adopter reads any AI claim against this ladder and against the evidence posture — WHO-recommended and deployable, strong-elsewhere-but-validate-locally, or early-and-pilot-only.
Key terms:
- Computer-aided detection (CAD) — software that reads a medical image (e.g. chest X-ray) to flag or triage abnormality; WHO-recommended for TB screening in people aged 15+.
- Augmentation architecture — deploying AI as triage or second reader inside clinical governance, with human escalation, rather than as a replacement for clinicians.
- Capability ladder — the sequenced build of data quality, then statistical capacity, then machine learning; ML bolted onto poor data automates existing blind spots.
- SMART guidelines — WHO's machine-readable clinical recommendations that let national protocols deploy into digital systems at software speed.
- Evidence posture — a label for how ready an AI use is: WHO-recommended, strong-elsewhere-validate-locally, or early-pilot-only.
Knowledge check: Q: What is the significance of WHO recommending TB CAD for chest X-ray triage? A: It is the first WHO endorsement of AI performing a human diagnostic task — letting a district hospital without a radiologist screen at radiologist-level quality. It is deployable now, under governance.
Q: Why is "augmentation, not replacement" the right frame for African AI rather than a debate about job loss? A: Where the baseline is roughly one radiologist for millions of people, AI is not displacing workforce — it is creating capacity that would not otherwise exist, so the goal is to make the frontline worker more capable, not redundant.
Q: A vendor claims its ML model will "transform" your clinic's outcomes. What does the capability ladder tell you to check first? A: Whether the data foundations beneath it (quality, completeness, registries) are sound. ML trained on incomplete or biased routine data automates the system's blind spots; data quality must come first.
Summary: The credible AI frontier is specific — narrow tasks where specialists are scarcest, led by WHO-recommended TB CAD. In African economics AI is augmentation, not replacement, and the critical adopter reads every claim against the capability ladder and its evidence posture.