Monitoring in Prisons Using Artificial Intelligence

Legislators should be cautious when expanding the use of artificial intelligence, given its shortcomings.

FREMONT, CA: As artificial intelligence (AI) has become more powerful and accessible, governments have become increasingly interested in its potential benefits. A fiercely contested application of AI is monitoring talks between inmates and outside callers within jails and correctional facilities to identify specific words or phrases that may indicate danger for inmates.

Stay ahead of the industry with exclusive feature stories on the top companies, expert insights and the latest news delivered straight to your inbox. Subscribe today.

Reuters reported that a group of congressional lawmakers made a request to the Department of Justice requesting a report on the potential use of AI in federal prisons, indicating that lawmakers may be receptive to the concept of implementing this technology on a wide scale. Reuters' David Sherfinski and Avi Asher-Schapiro wrote:

A crucial House of Representatives panel has requested a report on using AI to analyze prisoners' phone calls, paving the way for prisons in the United States to receive more technological assistance in monitoring inmate speech.

Families and advocates for prisoners argue that depending on AI to interpret messages leaves the system vulnerable to errors, misunderstandings, and racial bias.

The request for the Department of Justice (DOJ) to further investigate the technology to help prevent violent crime and suicide is included in an $81 billion-plus budget package for the DOJ and other federal agencies in 2022 that was approved by the Appropriations Committee last month.

The device can automatically transcribe inmates' phone calls by studying their communication patterns and detecting specific words or phrases, including slang, that are pre-programmed into the system by officials.

In an emailed statement, a Democratic staffer to the House of Representatives urged the Department of Justice to "consult with stakeholders while considering the viability of employing such a system."

Several state and local facilities around the nation, including Alabama, Georgia, and New York, have already begun implementing the technology.

Current problems with AI in prisons indicate early efforts to adopt the software may offer more risks than benefits. Comparing talks is hampered by the limited data available to the software in the current state of AI call monitoring. Developers focused on popular languages and dialects in the early days of analyzing language with AI. As a result, contemporary AI that investigates conversations struggles to comprehend some communication formats more than others.

This aspect of contemporary AI becomes troublesome when its usage in the criminal justice system is considered. Although most Americans speak English, there are over 30 primary varieties of American English. Currently, a significant proportion of inmates in American prisons do not talk about the kind of English that many developers train AI systems to recognize. According to research, AI constantly misinterprets African American English (AAE) compared to other dialects. According to a new study by Stanford Engineering, the technology that drives the nation's premier automatic speech recognition systems makes twice as many mistakes when interpreting words uttered by African Americans than when interpreting the exact words spoken by whites.

As a result, the use of AI in its current form may unwittingly discriminate against certain persons by flagging more of their discussions for human review than others. Consequently, utilizing AI in prisons before the technology can accurately catalog all offenders' languages would undoubtedly cause issues for populations already subject to discrimination.

The second obstacle to growing the use of AI in correctional facilities is not the technology's limitations but rather the extent to which management should rely on AI for effective oversight. AI can assist personnel in doing jobs more efficiently; nonetheless, correctional institution administrators should avoid responding to present challenges by depending excessively on AI in jail management. When the AI flags a communication, there must be a legitimate review and appeal process; it cannot be assumed that the AI system is always correct.

The labor-saving potential of AI has already attracted the attention of correctional administrators across the nation. As in other industries, technology has brought significant advancements to the field of corrections, but an over-dependence on new surveillance methods for convicts might have harmful effects. Instead of providing enough accommodations for convicts with COVID-19, New Orleans invested $70 million in technologically advanced camera upgrades.

Even if callers know that AI software is on the line, failure to physically supervise external calls may pose safety risks for inmates. Even if AI were capable of comprehending all inmates' calls, it is possible that some convicts would attempt to mislead the program, just as some inmates attempt to smuggle contraband into institutions or continue outside criminal operations while confined. Suppose officials opt to rely solely on AI to monitor phone calls. In that case, convicts might easily use codewords or other methods to circumvent AI software, making it easier for them to plot risky activities that could endanger inmates and cops. On the other hand, AI that incorrectly identifies innocent terms as problematic may result in unjust punishment of inmates.

However, policymakers should not overreact by concluding that the technology should be outlawed due to the genuine flaws of AI as it exists today. Researchers are already addressing some practical challenges associated with deploying AI to monitor prisoner discussions. If AI reaches the level of sophistication required to monitor prisoner discussions successfully and corrections staff accept it as a tool rather than a replacement, the technology might be revolutionary.

Additionally, a complete ban on the use of AI in jails would prevent prisoners from benefiting in the future from this technology. In a population where at least half of individuals are mentally ill and where present prisons only raise the likelihood of acquiring mental illness and further behavioral difficulties, we should pursue technologies that allow us to enhance the health of prisoners across the nation.

Before permitting expansions of AI's use, legislators should be cautious of the technology's existing flaws, but they must also avoid prematurely limiting this technology's future life-saving potential.