Page 9 - AI Vol 2: Risks of AI
P. 9
01 C. ACCOUNTABILITY
While AI can significantly enhance efficiency,
decision-making, and service delivery, it also
brings forth critical questions of accountability.
The common theme of this section is that a
public agency is ultimately responsible for their
decisions, even if AI systems are utilized to inform
or advise on decision making. Accordingly, it
is critical for humans to be informed about the
limitations and risks of AI, oversee all uses of
AI and independently evaluate outputs, and
ultimately come to independent decisions a single data-point in a comprehensive decision-
informed by, but not reliant on, information and making process. Unquestioning reliance on AI
recommendations produced by AI. This is not outputs may present a variety of legal risks for
only critical to ensure agency decisions are well public agencies.
reasoned and legally permissible, but it provides
the foundation of transparency that is imperative Accountability is crucial from the outset,
for ensuring public trust in AI deployments. encompassing decisions on AI implementation,
its intended use, and the selection of AI systems.
The cornerstone of integrating AI into public These initial steps set the foundation for how
agency operations is the recognition that, despite effectively a public agency can harness the
the advanced capabilities of these technologies, power of AI while maintaining the trust and
human decision-makers hold the ultimate confidence of the public it serves. Organizational
responsibility for outcomes. This principle is accountability is essential and includes decisions
pivotal in ensuring that AI serves as a tool for on the deployment, purpose, and choice of AI
enhancement, not as a replacement for human systems utilized by a public agency. These early
judgment and accountability. As discussed in decisions lay the groundwork for effective use of
this section and elsewhere in this series, while AI and it is critical for agencies to develop clear
AI systems are powerful and likely to provide a policies on AI adoption and usage within their
myriad of benefits to public agencies, they are still operations. Without these guidelines, there is
prone to errors, bias, and lack human wisdom, a risk of employees independently using AI for
understanding, and ethical reasoning. Agency staff various tasks, potentially exposing the agency to
utilizing AI systems must maintain a “human in the risks such as privacy breaches, bias in decision-
loop” actively evaluating AI outputs to identify and making, and regulatory non-compliance. When
prevent errors. Staff and decision makers alike need deciding to implement AI systems, it is crucial to
to be informed about the benefits and limitations of establish specific protocols that define who can
AI systems and should only ever use AI outputs as use AI systems and for what purposes. Moreover,
RISKS OF AI | LOZANOSMITH.COM VOLUME 2 | 9