Artificial intelligence has attracted the interest of both the general public and legislators, including members of the New Mexico Legislature.
During their meeting on Monday, the Interim Legislative Science, Technology, and Telecommunication Committee delved into the topic of embedding transparency into artificial intelligence programming concerning public resources and services.
The session commenced with a presentation by Professor Cristopher Moore from the Santa Fe Institute, emphasizing a focus on algorithmic issues rather than generative AI.
Moore highlighted the utilization of automated decision systems, known as automated methods, in both public and private sectors for decision-making processes that impact individuals. These systems are employed in various domains such as pretrial detention, predictive policing, healthcare, fraud detection, case worker recommendations, and management of public housing waiting lists in the public sector. Conversely, in the private sector, these algorithms play a role in automated recruitment, credit assessment, lending practices, and tenant screenings.
One significant term discussed in the realm of AI is “black box AI,” referring to systems that do not reveal their inputs and operations to users or stakeholders, thus lacking transparency.
Expressing his disapproval of black box AI technology, Moore highlighted the unconventional nature of these algorithms, emphasizing that they are not explicitly programmed by humans. Instead, they are provided with data and tasked with deriving rules that optimize performance based on that data, a process known as the training process. Moore stressed the importance of various levels of transparency beyond this training process.
According to Moore, transparency should extend to individuals’ rights to comprehend the basis of decisions made by governmental agencies that affect them, citing it as a matter of due process. Additionally, considerations should be given to how the AI or algorithm was trained, with many systems being trained on data from diverse geographical locations and time periods, potentially leading to inaccuracies in the context of New Mexico’s unique demographic composition.
The presentation raised more questions than answers, prompting discussions on the approaches adopted by other states to enhance AI transparency. Vermont, for instance, enacted a law in 2022 requiring vendors to disclose adequate information about algorithms to facilitate independent audits before any state agency engages in a contract for the procurement, lease, or utilization of such tools.
