The Role of Artificial Intelligence in Shenzhen’s Public Health Future

AI Medical Startups in Shenzhen are seemingly ubiquitous, fittingly as the market is expanding steadily, with a projected size exceeding 1.2 trillion just in Mainland China (The State Council of the PRC). 

It is not inaccurate to suggest that the integration of AI in the revamping of medical systems has not contributed to Shenzhen’s economic wellbeing. However, in the harnessing of AI, such risks — poor user adherence, misdiagnosis, and underrepresentation of certain population subgroups in training data — must not be overlooked. 

Several educational institutions, including CUHK Shenzhen and the University of Shenzhen openly offer courses on AI’s application in medicine. In deploying AI, many such use a hybrid approach, which combines AI with human expertise. However, this presents the risk of dependence, leading to inability for doctors to be unable to make an independent diagnosis without consulting AI. The notion of AI being an existential threat is not a novel one; big-tech founders such as Sam Altman have attested to such, stating ‘AI will most likely lead to the end of the world…’. 

A recent study showed that Open AI’s GPT-4 answered open-ended medical questions incorrectly nearly 2/3rds of the time (University of Waterloo). Provided that AI makes an incorrect treatment plan, then the burden of risk will fall entirely upon the patient. Possibly dependence may also lead to the erosion of diagnosis skills. A study has found that after the use of AI-aided detection, performance in detecting polyps has declined from 28.4% to 22.1% (Spadaccini et al.) 

Unencumbered by resource constraints, such as a shortage of medical personnels, public health funding in Shenzhen has been shown to be more than sufficient in meeting the demands of our institutions without AI. Despite this, Shenzhen, being at the lead of the AI frontier, has continued to push funding towards an increasingly congested market. This creates a disjuncture between demand for the sake of commercial interests and the actual practical needs of the local population. 

AI tools often perpetuate algorithmic biases that cause misdiagnosis as well as amplify pseudo-science misinformation. Noting the fallibility of the models as well as the importance of early identification, it is not just enough to push against using AI for self diagnosis, when the tools themselves are becoming institutionalized (Norori et al.).

In the deployment of AI in the healthcare sector, caution must be exercised, and innovation and corporate priorities must not take precedence over the safety of the public. In the testing of models, medical questions cannot serve as a proxy for how AI may communicate with humans. Given its opacity, if AI were to be used, it must be  checked in several dimensions, such as individual medical history, in which it may then solidify its role in assisting in clinical decision-making.

By Jillian Ye

Bibliography

“AI’s Medical Diagnostic Skills Still Need a Check-up | Waterloo News.” Waterloo News, 22 May 2025, uwaterloo.ca/news/media/ais-medical-diagnostic-skills-still-need-check.

Norori, Natalia, et al. “Addressing Bias in Big Data and AI for Health Care: A Call for Open Science.” Patterns, vol. 2, no. 10, 8 Oct. 2021, p. 100347, pmc.ncbi.nlm.nih.gov/articles/PMC8515002/, https://doi.org/10.1016/j.patter.2021.100347.

“China’s Core AI Industry Scale Tops 1.2 Trln Yuan in 2025: Official.” Www.gov.cn, 2025, english.www.gov.cn/archive/statistics/202603/05/content_WS69a96a11c6d00ca5f9a09932.html.

Spadaccini, Marco, et al. “AI and Polyp Detection during Colonoscopy.” Cancers, vol. 17, no. 5, 26 Feb. 2025, p. 797, pmc.ncbi.nlm.nih.gov/articles/PMC11898786/, https://doi.org/10.3390/cancers17050797.