IEEE Technology and Society Magazine - September 2023 - 83

code labels can be found in Supplemental Materials
(Appendix) Table C.
Despite code labels under the same principles
demonstrating strong proximity as expected, subprinciples
belonging to different principles also occur in
the same or close sections of documents frequently.
For instance, " Preventing Stigmatization " displays
strong proximity to " Disparities, " as long as " Race,
Sex, and Ethnicity. " While the latter two are categorized
under the general principle " Inclusiveness and
Equity, " it is interesting yet not surprising to see the
core codes connect and echo with the principle of
" Human Safety and Public Interest, " the parent principle
of " Stigmatization. " Inclusiveness is the key
requirement for AI healthcare applications to ensure
equal, appropriate, and nondiscriminatory access to
whichever services they provide. This further suggests
that ensuring inclusivity and equity in AI healthcare
applications is not only important in its own right,
but is also closely linked to ensuring the safety and
well-being of users. By addressing disparities and preventing
stigmatization based on factors such as race,
sex, and ethnicity, AI healthcare applications can promote
greater access and equity while also promoting
the safety and well-being of all users. Overall, this analysis
highlights the importance of considering these
two principles together and ensuring that AI healthcare
applications are designed and implemented in a
way that prioritizes both inclusivity and safety.
Another pair demonstrating a great amount of
intersection is " Regular Evaluation " and " Public
Consultation. " Despite being categorized as different
subprinciples, the codes " Regular Evaluation "
and " Public Consultation " demonstrate a significant
overlap in the policy documents. While regular
evaluation is essential for monitoring and assessing
AI systems' performance and impact, public
consultation emphasizes engaging with a diverse
range of stakeholders, including patients and caregivers,
in the development and deployment of AI
in healthcare. The intersection between these two
codes underscores the importance of transparency,
accountability, and stakeholder participation in
ensuring the responsible and ethical development
of AI in healthcare. This highlights the need for a
collaborative and participatory approach to AI ethics,
where stakeholders' input is valued and taken
into account throughout the development and
deployment process. By engaging stakeholders and
ensuring transparency and accountability, we can
September 2023
Figure 5. Code model of intersection
analysis.
develop AI systems in healthcare that are responsive
to the needs and concerns of the communities
they serve. Prioritizing these principles can allow us
to work toward building a more just, equitable, and
effective healthcare system that leverages the power
of AI to improve health outcomes for all.
Modal verb analysis
As described in the coding section, a set of modal
verbs is utilized to differentiate coded segments that
are proposing legislations, frameworks, and future
practices from the rest. Figure 6 shows the percentages
of coded segments labeled by the modal verb search
under each principle. The principle of " Human Safety
and Public Interest " has the highest proportion of proposed
practices (66.67%), while the percentage of
" Inclusiveness and Equity " is the lowest of all six principles
(48.54%). Overall, there is no significant difference
between any two principles with all the percentages
staying within the range between 40% and 60%.
In the comparison of modal word labeling and
ethical principles, inclusiveness and equity possess
the lowest percentage of content related to proposed
practices-48.56%. In this study, inclusiveness and
equity include biases in the data sampling process, disparities,
and diversity in languages, race, and gender.
Ideally, equitable access to the benefits of AI applications
in healthcare should be ensured regardless of
race, gender, and other socio-economic factors. The
subprinciples Sampling Bias and Race, Sex, and Ethnicity
display a noticeable level of frequency in the
code cloud analysis, meaning that such challenges
83
IEEE Technology and Society Magazine - September 2023

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - September 2023
