Artificial intelligence delivers public benefit when a specific tool does a specific job and an outside body has checked it. Two documented examples are computer-aided detection software for tuberculosis screening, which the World Health Organization (WHO) endorsed in guidelines dated March 22, 2021, and language tools built by Canada's National Research Council (NRC) with Indigenous communities. In both, the benefit is narrow, the limits are written down, and people stay in the loop.
Key takeaways
- WHO's 2021 screening guidelines list computer-aided detection (CAD) software among recommended tools for reading chest X-rays in tuberculosis screening.
- WHO's operational handbook limits the use to people aged 15 and older and asks programs to calibrate the software and keep human readers available.
- The NRC's Indigenous Languages Technology project builds tools such as verb conjugators, read-along software, and predictive text with named community partners.
- Several NRC tools are still in testing or research, and the NRC itself lists machine translation for Inuktitut as research work.
- A claimed social benefit is easier to trust when it names the task, the population, the checker, and the known limits.
Tuberculosis screening: software that reads chest X-rays
Tuberculosis screening often relies on chest X-rays, and many high-burden settings have few trained readers. WHO's guideline "Module 2: screening: systematic screening for tuberculosis disease," published March 22, 2021, contains 17 new or updated recommendations. Among the tools it recommends is computer-aided detection software, which analyzes a digital chest X-ray and returns a score that suggests how likely the image is to show disease.
WHO's operational handbook for the same module sets the practical terms. The software may be used in place of a human reader to interpret digital chest X-rays in people aged 15 and older, and it is not currently recommended for children under 15. The recommendation applies to products whose external evaluation shows performance not inferior to the products WHO reviewed in 2020. A score is not a percentage and does not directly give a person's risk of disease.
What the handbook asks of programs
The handbook treats the software as a tool that needs local setup. Each program is expected to calibrate the software to pick a threshold score, because the thresholds supplied by manufacturers perform inconsistently across settings. Performance should be monitored after calibration. Human reader capacity should be kept, for example for children and for abnormal images where a disease other than tuberculosis is suspected. Costs, licensing, and infrastructure differ by setting, and the handbook says so.
This is the kind of benefit that is easy to overstate. The software does not diagnose tuberculosis on its own authority. It helps decide who goes on to a confirmatory test, which matters most where trained readers are few.
Indigenous language tools built with communities
The second example is a Canadian public research effort. The NRC's Indigenous Languages Technology project fact sheet, last modified April 2, 2026, says the NRC works with Indigenous language experts, instructors, and communities to build speech- and text-based technologies for language reclamation, revitalization, and stabilization.
The listed work is concrete. Verb conjugation software exists for two Mohawk dialects, Algonquin, and Michif, with Mi'kmaq being prepared. ReadAlong Studio highlights words in Indigenous-language books as they are read aloud and has been applied to 22 languages. Predictive text covers Plains Cree and Makah, and SENĆOŦEN keyboards. Four community teams created online courses for Kwak'wala, Mi'kmaq, Naskapi, and Southern Michif. Named partners include the Onkwawenna Kentyohkwa Mohawk-language immersion school, the First Peoples' Cultural Council, and the Council of Yukon First Nations.
Why these tools stay unfinished by design
The fact sheet is plain about limits. Most Indigenous languages in Canada are polysynthetic, which makes verb conjugation among the hardest things for learners. Older recording formats are degrading, and the page states that fewer than 40 Elders remain fluent in N̓syilxčn̓. Several tools are still in user testing or research: an Inuktitut reading and writing suite is in testing, speech recognition labeling tools start with Inuktitut and Cree, and machine translation between English and Inuktitut is described as research. The software supports teachers and learners; the languages are carried by the people who speak them.
A checklist for judging a benefit claim
| Question | Tuberculosis screening example | Language project example |
|---|---|---|
| What is the exact task? | Reading digital chest X-rays for screening and triage | Conjugating verbs, read-along books, predictive text, transcription support |
| Who is the intended population? | People aged 15 and older | Communities and learners of named languages |
| Who checked the claim? | WHO guideline process and external product evaluation | Community partners and the NRC's own reporting |
| What is written down as a limit? | Not for children under 15; local calibration needed | Tools in testing; translation still research |
| Where do people stay involved? | Human readers kept for children and unclear images | Language experts, instructors, and Elders lead the content |
Claims that fail these five questions, such as a promise that AI will fix health care or save languages in general, give readers nothing to verify.
FAQ
Does AI diagnose tuberculosis by itself?
No. In WHO's framework the software supports screening and triage by reading chest X-rays, and a positive result leads to further testing. Diagnosis rests on confirmatory tests and clinical judgment.
Is the WHO recommendation valid for children?
The operational handbook says the software is not currently recommended for children under 15. The recommendation covers people aged 15 and older.
Are the NRC language tools finished?
Some are in use, such as verb conjugation software for several languages, and others are in testing or research. The fact sheet lists Inuktitut machine translation as research.
Does this page say AI is good for society overall?
No. The two cases show that documented benefits exist in narrow tasks. Wider effects, including harms, are covered in the related articles below. This article is general information and not a recommendation of any product or medical advice.
Related reading
- The impact of AI on modern society
- AI, sustainability, and society
- Documented harms of AI
- AI and the future of work
More in AI Basics.
Last verified: October 2026
Sources
- WHO consolidated guidelines on tuberculosis: module 2: screening: systematic screening for tuberculosis disease, World Health Organization, March 22, 2021
- Operational handbook on tuberculosis, Module 2: screening, section 3.1.3, CAD technologies for screening and triage, World Health Organization, 2021
- Fact sheet: Indigenous Languages Technology project, National Research Council Canada, page modified April 2, 2026