Page 14 - AI Vol 2: Risks of AI
P. 14
01 D. IMPACTS OF WIDESPREAD AI ADOPTION
It is important to recognize that the impacts of full image or video deepfakes of public agency
AI on public agencies will not just be limited to officials. As we head into the election season, AI
how agencies themselves chose to deploy (or not) disinformation is one of the primary concerns
these systems. As AI begins to proliferate through of experts studying the impact of AI on society.
society, public agencies will have to learn to adapt Given AI’s potential to make these disinformation
to new challenges these systems bring. campaigns easier, we can expect these strategies
to be employed at the local level as well as state-
As constituents and parents increasingly turn to wide and federal elections. This is particularly
AI systems for information regarding agencies, likely as political activists and groups have
policies, and laws, as well as to shape their increasingly targeted their activism at the local
understanding of the agency's decisions, reliance level, a strategy that first gained prominence
on these systems is expected to rise. Although during the COVID-19 pandemic.
the information people receive from LLMs
may be incorrect for any number of reasons It is also possible that as the general public
(bad prompting, hallucinations, poor quality gains an understanding and acceptance of AI
information retrieval), people are primed to technology, their expectations about public
believe that these systems are providing accurate agency use of these systems may shift. Currently,
information, particularly when it comes from given the attention on hallucinations, bias, and
companies they have high confidence in (e.g., other problems associated with AI systems, many
Google, Microsoft). For example, many people communities may be skeptical of any public
have come to trust the accuracy of information agency deployment of AI technology. However,
obtained from a Google Search result and may it’s possible that over time, as AI systems
incorrectly assume the information obtained
from a Google Gemini interaction is of equal
quality. It is likely that agencies will begin to
see misinformation spread in their communities
based on shared interactions with LLMs, a fact
that will be further exacerbated by people’s biases
for trusting AI sources.
AI can also be used to spread disinformation
intentionally. This includes things like mass
producing targeted social media posts, voice
cloning systems being used to make it appear
that an official said something they did not, and
14 | VOLUME 2 RISKS OF AI | LOZANOSMITH.COM