Page 5 - AI Vol 2: Risks of AI

INCORPORATING A "HUMAN IN
THE LOOP", REQUIRING HUMAN
OVERSIGHT OF AI-GENERATED
DECISIONS, IS A CRITICAL
MITIGATION MEASURE FOR
ADDRESSING POTENTIAL BIASES
IN AI SYSTEMS, ESPECIALLY
WITHIN PUBLIC AGENCIES.

When bias is identified after deployment, public
agencies should first and foremost ensure that essential, encouraging employees to question and
the bias does not impact agency decisions. challenge AI recommendations when necessary.
Additionally, agencies can work with developers to Agencies should also have protocols for reporting
adjust the model or seek alternative solutions that potential biases or inaccuracies in AI decisions.
demonstrate a stronger commitment to unbiased This feedback can inform the agency's decision
outcomes. While the technical details of these on whether to use a specific AI system and may

corrections are managed by specialists, agency also be communicated to the developer to enhance
officials should understand the broad strategies the model, aiming to mitigate bias.
used for bias correction, such as diversifying
training data or adjusting the algorithm's decision- Bias may also be minimized by engaging in
making parameters. prompting strategies which promote transparency

and are less prone to bias. Instead of asking an AI
Incorporating a “human in the loop,” requiring system to make broad assessments or decisions,

human oversight of AI-generated decisions, a more targeted approach in questioning can
is a critical mitigation measure for addressing
potential biases in AI systems, especially within
public agencies. This approach ensures that AI-
generated decisions are not made in isolation but
are instead reviewed and, if necessary, adjusted by “THE THING THAT’S GOING
human oversight. For this approach to be effective, TO MAKE ARTIFICIAL
agency personnel involved in overseeing AI INTELLIGENCE SO POWERFUL
decisions need appropriate training. They should IS ITS ABILITY TO LEARN, AND
THE WAY AI LEARNS IS TO
be aware of the potential biases AI systems might LOOK AT HUMAN CULTURE.”
hold and be equipped to identify signs of such
biases in decision outputs. Additionally, fostering - DAN BROWN
a culture of critical engagement with AI tools is

RISKS OF AI | LOZANOSMITH.COM VOLUME 2 | 5