APMA News - September/October 2023 - 36

coding
can also introduce bias. For example, the features
selected for training an AI system or the
weight assigned to certain factors can unintentionally
introduce or amplify biases. The algorithms
themselves can also exhibit biases in how
they process and interpret data.
Addressing inherent bias in AI will require a combination
of approaches, including diverse and representative
training data, careful algorithmic design,
and thorough testing and evaluation to identify and
mitigate biases. It will be essential to have transparency,
accountability, and ongoing scrutiny of AI
systems to ensure they are fair, unbiased, and align
with ethical and societal values.
Overcoming inherent bias in AI requires concerted
efforts and the implementation of several approaches.
Here are some possible ways to address
and mitigate inherent bias:
* Diverse and Representative Training Data:
Ensuring that the training data used to train AI
models is diverse and representative of the entire
population is crucial. It involves collecting data
from a wide range of sources and demographics,
taking into account factors such as age,
gender, race, and socioeconomic background.
By having a more inclusive dataset, the AI system
can learn from a broader range of examples,
reducing the risk of bias.
* Bias Detection and Evaluation: Implementing
thorough testing and evaluation procedures to
identify and measure bias in AI systems is essential.
This process involves examining the outputs
of the AI system and assessing whether they
align with fairness and non-discrimination principles.
Various metrics and evaluation frameworks,
such as demographic parity and equalized odds,
can be used to assess bias. If bias is detected,
it can help pinpoint areas for improvement and
inform corrective measures.
* Algorithmic Design and Regular Auditing: Careful
consideration of algorithmic design choices is
necessary to mitigate bias. This process includes
examining the features used for training, ensuring
they are relevant and fair, and paying attention to
the weighting of different factors. Regular audits
and reviews of the algorithms can help identify
and rectify biases that may emerge over time. It
is important to involve multidisciplinary teams,
including ethicists, social scientists, and domain
experts in the design and evaluation process.
* Ethical Guidelines and Standards: Establishing
clear ethical guidelines and standards for AI
development and deployment is crucial. These
36
apmanews
SEPTEMBER/OCTOBER 2023
guidelines should explicitly address issues of bias
and discrimination, emphasizing fairness, transparency,
and accountability. Organizations and
developers should adhere to these guidelines
throughout the AI life cycle, from data collection
and model development to deployment and
ongoing monitoring.
* Collaboration and Stakeholder Engagement:
Addressing inherent bias in AI requires collaboration
and engagement from various stakeholders,
including researchers, developers,
policymakers, and impacted communities.
Engaging diverse perspectives and involving
those who are affected by AI systems can help
identify biases and ensure that the development
and deployment of AI technologies are more
inclusive and equitable.
* Continuous Monitoring and Iterative
Improvement: Bias mitigation should be
an ongoing process. AI systems should be
continually monitored and evaluated for bias,
even after deployment. Feedback loops, user
input, and real-world performance data should
be leveraged to identify and rectify biases.
Regular updates and iterations of AI models
can help improve their fairness and mitigate
any emerging biases.
The consequences of inherent bias in AI can
be significant. It can lead to unfair or discriminatory
outcomes, exacerbate existing social
inequalities, and reinforce stereotypes. Moreover,
it can disproportionately affect marginalized
groups, perpetuating systemic biases and creating
further disadvantages.
It is important to note that completely eliminating
all forms of bias in AI systems may be challenging.
However, by implementing these strategies
and fostering a commitment to ethical AI practices,
it is possible to significantly reduce inherent bias
and promote fair and unbiased AI systems.
References
1
https://www.ama-assn.org/practicemanagement/cpt/cpt-appendix-s-ai-taxonomymedical-services-procedures
2
Hoffman
KM, Trawalter S, Axt JR, Oliver MN.
Racial bias in pain assessment and treatment
recommendations, and false beliefs about
biological differences between Blacks and whites.
Proc Natl Acad Sci U S A. 2016 Apr 19;113(16):4296301.
doi: 10.1073/pnas.1516047113. Epub 2016 Apr
4. PMID: 27044069; PMCID: PMC4843483.
Id.

https://www.ama-assn.org/practice-management/cpt/cpt-appendix-s-ai-taxonomy-medical-services-procedures https://www.ama-assn.org/practice-management/cpt/cpt-appendix-s-ai-taxonomy-medical-services-procedures https://www.ama-assn.org/practice-management/cpt/cpt-appendix-s-ai-taxonomy-medical-services-procedures

APMA News - September/October 2023

Table of Contents for the Digital Edition of APMA News - September/October 2023