The Impact of AI on Prison Safety and Security

FREMONT, CA: Governments are increasingly exploring the potential benefits of artificial intelligence (AI) in monitoring conversations between inmates and outside callers in jails and correctional facilities. They aim to identify words or phrases that may indicate danger for inmates. 

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. Even though most Americans speak English, there are over 30 prior varieties of American English. Presently, a considerable proportion of inmates in American prisons do not talk about the kind of English many developers train AI systems to determine. Based on the research, AI continually misunderstands African American English (AAE) as against other dialects. Based on a new study by Stanford Engineering, the technology that operates the nation's premier automatic speech identification systems makes twice as numerous mistakes when interpreting words spoken by African Americans than when analyzing the exact words spoken by whites.

Therefore, the application of AI in its present form may inadvertently distinguish specific persons by fading more of their arguments for human review than others. Therefore, using AI in prisons before the technology can precisely catalog all offenders' languages would cause problems for residents already subject to discrimination.

The second barrier to raising the application of AI in corrective facilities is not the technology's boundaries but instead the degree to which management should depend on AI for effective management. AI can help personnel do jobs more effectively, yet, correctional institution administrators should evade responding to offer challenges by relying excessively on AI in jail management. When the AI flags a communication, there must be a fair review and request process; it cannot be assumed that the AI system is often correct.

AI's labor-saving capacity has already lured correctional administrators' attention nationwide. As in other industries, technology has brought significant advancements to the field of disciplines, but an over-reliance on new surveillance strategies for convicts might have detrimental effects. 

Even if callers understand that AI software is on the line, failure to physically handle external calls may pose safety risks for prisoners. Even if AI could learn all inmates' calls, some convicts would likely attempt to mislead the program, just as some inmates attempt to smuggle contraband into institutions or proceed outside criminal operations while restricted. Suppose officials choose to rely only on AI to watch phone calls. If so, convicts might readily employ codewords or other ways to circumvent AI software, making it more comfortable to connive risky actions that could threaten inmates and cops. Conversely, AI that wrongly identifies innocent terms as inappropriate may follow in the unjust punishment of detainees.

Still, policymakers should not exaggerate 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.

Also, a total ban on the application of AI in jails would stop prisoners from availing in the future from this technology. In a population where leastways half of the individuals are psychologically ill and where present prisons just raise the possibility of getting mental illness and further behavioral difficulties, we should pursue technologies that permit us to improve the health of prisoners across the nation.

Before enabling expansions of AI's application, legislators should be careful of the technology's present flaws, but they must also dodge prematurely restricting this technology's future life-saving possibility.