Efficiency vs. Equity: A Structured Interdisciplinary Review of AI in Criminal Justice Risk Assessments

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

The included sources identified reducing the bias of human judgment, doing a better job of predicting recidivism, and reducing the number of individuals held in jails and prisons as potential benefits of AI.

“Other benefits include reducing the bias of human judgment [3], doing a better job of predicting recidivism [4], and reducing the number of individuals held in jails and prisons [5].”

• cites [3] Rowland, 2023, AI in Corrections: The Basics and a Way to Experiment, Fed. Probat.Not checked yet
• cites [4] Stevenson, 2018, Assessing risk assessment in action, Minn. Law Rev. sourceSupported
• cites [5] Kleinberg, 2018, Human decisions and machine predictions, Q. J. Econ. sourceSupported

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

This problem raises serious concerns about the legality and usability of these programs, in addition to concerns regarding bias.

“This problem raises serious concerns about the legality and usability of these programs [4], in addition to the concerns regarding bias.”

• cites [4] Stevenson, 2018, Assessing risk assessment in action, Minn. Law Rev. sourcePartly supported

  The Introduction describes practical problems with using risk-assessment tools, including judges ignoring them or misunderstanding risk scores, supporting the usability concern. The Criminal Justice Risk Assessment section cites legal issues related to transparency, but our copy does not establish the broader claim that the programs’ legality is in serious doubt. The citation precedes the separate phrase about bias, so that phrase is not assessed here. (LLM, conf=0.78)

  Experimental — accuracy not yet measured

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Most sources recognize that the broad promises of AI-enabled risk assessment have not been fully realized yet.

“Although most sources recognize that the broad promises of AI-enabled risk assessment have not been fully realized yet [6]”

• cites [6] Ludwig, 2021, Fragile algorithms and fallible decision-makers: Lessons from the justice system, J. Econ. Perspect. sourceNot checked — no full text of the source

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The final sample for review was 56 articles.

“Of the 105 retrieved, 49 articles were excluded, giving a final sample size for review of 56 articles.”

It may state two separate claims as one.

The included sources identified a wide range of potential benefits of AI systems, with the most common being a reduction in the costs associated with the criminal justice system or better distribution of resources.

“As seen in Figure 1, a wide range of potential benefits have been identified by our included sources, with the most common being a reduction in the costs associated with the criminal justice system or better distribution of resources.”

It may state two separate claims as one.

The review included 56 studies that were collected from Google Scholar and the ACM Digital Library, published since 2015, discuss AI, and focus on the adult justice system in the US.

“Included papers were collected from Google Scholar and the ACM Digital Library and have been published since 2015, discuss AI, and focus on the adult justice system in the US, yielding 56 studies.”

Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it.

There is an extremely large number of factors that might cause biased performance by an actuarial or AI tool.

“There is an extremely large number of factors that might cause biased performance by an actuarial or AI tool.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

PCRA was discussed in 4 of the sources.

“The sixth most popular tool for risk assessment in the courtroom is the Post-Conviction Risk Assessment (PCRA) [43], another federally developed tool, discussed in 4 of our sources.”

• cites [43] Johnson, 2011, The construction and validation of the federal Post Conviction Risk Assessment (PCRA), Fed. Probat.Not checked yet

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

These factors range from factors intrinsic to the tools themselves, problems or difficulties with collected real-world data, and the place that these tools occupy in the world.

“These range from factors intrinsic to the tools themselves, problems or difficulties with collected real-world data, and the place that these tools occupy in the world.”

It may state two separate claims as one.

Two examples of tools that were updated after several years of use are the LSI, which later became the LSI-R, and the Static Risk Assessment (SRA), which was later replaced with the SRA2.

“Two examples of tools that were updated after several years of use are the LSI, which later became the LSI-R, and the Static Risk Assessment (SRA), which was later replaced with the SRA2.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Virginia is often held up as an example of biased performance by ARA tools.

“Like Kentucky, Virginia is often held up as an example of biased performance by ARA tools.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.

Frequently, states choose to develop their own actuarial tools, using data that are not publicly available.

“Frequently, states choose to develop their own actuarial tools, using data that are not publicly available.”

It may state two separate claims as one.

Examples of proxies identified by the included sources include a count of stops by police, income, employment history, education, ZIP code, or an individual's first name.

“Examples of proxies identified by our included sources would be a count of stops by police [60], income [61], employment history [17], education [62], ZIP code [61], or an individual's first name [62].”

• cites [60] Westermann, 2020, Evidence-Based Sentencing: Risks and Opportunities, Lex Electron.Not checked yet
• cites [61] Gravett, 2021, Sentenced by an algorithm—Bias and lack of accuracy in risk-assessment software in the United States criminal justice system, South Afr. J. Crim. Justice sourceNot checked — no full text of the source
• cites [17] Elyounes, 2019, Bail or jail? Judicial versus algorithmic decision-making in the pretrial system, Colum. Sci. Technol. Law Rev.Not checked yet

It may state two separate claims as one.

12 (21%) sources identified social or criminal justice reform as a potential benefit of ARA, and 7 (13%) specified that ARA can be used as a replacement for cash bail.

“Twelve (21%) sources identified social or criminal justice reform as a potential benefit of ARA, and 7 (13%) specified that ARA can be used as a replacement for cash bail, allowing defendants identified as low-risk to await their trials outside of jail without the need for them to have access to large cash reserves.”

It may state two separate claims as one.

VPRAI and CPAT were each discussed in 8 of the 56 sources.

“The next two most popular tools, the Virginia Pretrial Risk Assessment Instrument (VPRAI) [41] and the Colorado Pretrial Risk Assessment Tool (CPAT) [42], were both developed at the state level, and both were discussed in 8 of our 56 sources.”

• cites [41] VanNostrand, M., and Rose, K.J. (2026, June 03). Pretrial Risk Assessment in Virginia: The Virginia Pretrial Risk Assessment Instrument (2009). BiblioCEJA, CEJA. Available online: https://biblioteca.cejamericas.org/bitstream/handle/2015/3099/Virginia_Risk_Assessment_2009.pdf?sequence=1.Not checked yet
• cites [42] Terranova, V.A., and Ward, K.C. (2020). Colorado Pretrial Assessment Tool Validation Study Final Report, National Association of Criminal Defense Lawyers.Not checked yet

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The use of AI-enabled risk assessment tools promises many potential benefits to criminal justice system actors and to society as a whole.

“The use of AI-enabled risk assessment tools promises many potential benefits to criminal justice system actors (i.e., defendants, judges, prosecutors, etc.), and to society as a whole.”

It may state two separate claims as one.

COMPAS is the most discussed risk assessment and the most frequently used as evidence of bias in ARA.

“It is the most discussed risk assessment and the most frequently used as evidence of bias in ARA.”

It may state two separate claims as one.

Other concerns include intentional or unintentional actions taken by the tools' developers, hiding or rationalizing human bias in the justice system, inconsistent interpretation of scores or answers to questions on risk surveys, and the need for tools to be regionally calibrated.

“Other concerns include intentional or unintentional actions taken by the tools' developers, which introduce bias in their products, hiding or rationalizing human bias in the justice system, inconsistent interpretation of scores or answers to questions on risk surveys, and the need for tools to be regionally calibrated.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Other risk assessment tools were discussed in fewer than four of the total sources.

“However, each of those tools were discussed in fewer than four of our total sources.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

COMPAS is generally presented as a caution against unintentional bias in risk assessment tools.

“COMPAS is generally presented as a caution against unintentional bias in risk assessment tools [44], while the other tools are presented in both positive and negative light.”

• cites [44] Dressel, 2018, The accuracy, fairness, and limits of predicting recidivism, Sci. Adv. sourceNot checked yet

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.

Many of the perceived sources of bias are more strongly connected to the human presence within the justice system.

“Many of the perceived sources of bias are more strongly connected to the human presence within the justice system.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

A great number of sources have expressed that AI tools, correctly implemented and trained, could make better predictions with less bias than human judges.

“Although many sources expressed concern about AI being used to remove human judgment from the justice system, a great number of sources have also expressed that AI tools, correctly implemented and trained, could make better predictions with less bias than human judges.”

It may state two separate claims as one.

Of the 44 sources that specifically stated that ARA can or does contribute to bias, 32 (73%) cited the COMPAS study as a source, while only 12 (27%) made the claim that ARA shows bias without using COMPAS as an example.

“Of the 44 sources that specifically stated that ARA can or does contribute to bias, 32 (73%) of them cited the COMPAS study as a source, while only 12 (27%) of them made the claim that ARA shows bias without using COMPAS as an example.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.

Interpretability and black-box models are frequently identified as concerns by authors discussing ARA, even when the ML models being used in risk assessments, like decision trees or logistic regressions, are inherently more interpretable than neural networks, which are not being used in courtroom risk assessments.

“It is important to note that interpretability and black-box models are frequently identified as concerns by authors discussing ARA, even when the ML models being used in risk assessments, like decision trees or logistic regressions, are inherently more interpretable than neural networks, which are not being used in courtroom risk assessments.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

While many sources express the view that AI, as an inanimate program, can only acquire bias from human-sourced data, the concern remains that ARA limit advances in fairness in our justice system.

“While many sources express the view that AI, as an inanimate program, can only acquire bias from human-sourced data [3], the concern remains that ARA limit advances in fairness in our justice system.”

• cites [3] Rowland, 2023, AI in Corrections: The Basics and a Way to Experiment, Fed. Probat.Not checked yet

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

This study is widely cited to show AI's synergistic potential for identifying high-risk offenders for incarceration, allowing offenders with lower risk scores to be given alternative sanctions.

“This study is widely cited to show AI's synergistic potential for identifying high-risk offenders for incarceration, allowing offenders with lower risk scores to be given alternative sanctions.”

It may state two separate claims as one.

Most ARAs have been built from relatively simple models, with logistic regressions being the most commonly referenced in the literature.

“The burden of training models on datasets of limited size [6] and on using models that are relatively interpretable [17] mean that most ARAs have been built from relatively simple models, with logistic regressions being the most commonly referenced in the literature.”

• cites [6] Ludwig, 2021, Fragile algorithms and fallible decision-makers: Lessons from the justice system, J. Econ. Perspect. sourceNot checked — no full text of the source
• cites [17] Elyounes, 2019, Bail or jail? Judicial versus algorithmic decision-making in the pretrial system, Colum. Sci. Technol. Law Rev.Not checked yet

It may state two separate claims as one.

17 (30%) of the sources identified ARA as a tool for increasing the objectivity of the justice system, while 14 (25%) specifically identified ARA as a tool for increasing the transparency of the justice system.

“Seventeen (30%) of our sources identified ARA as a tool for increasing the objectivity of the justice system, while 14 (25%) specifically identified ARA as a tool for increasing the transparency of the justice system.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.

Some tools have emerged in the literature as attracting more substantial attention and more frequently serving as the baseline for tools developed by state commissions.

“However, some tools have emerged in the literature as attracting more substantial attention and more frequently serving as the baseline for tools developed by state commissions.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

There is a common and widespread claim that ARA can be used to make the process of criminal justice less biased.

“However, this demonstrates that there is a common and widespread claim that ARA can be used to make the process of criminal justice less biased.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The claimed benefits of AI and ARA extend beyond the walls of the courtroom.

“The claimed benefits of AI and ARA extend beyond the walls of the courtroom.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The other tools are presented in both positive and negative light.

“COMPAS is generally presented as a caution against unintentional bias in risk assessment tools [44], while the other tools are presented in both positive and negative light.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.

There is very little consensus amongst states as to which risk assessments should be employed in courtroom settings.

“There is very little consensus amongst states as to which risk assessments should be employed in courtroom settings.”

Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.

AI is an extremely influential field of study both inside and outside of the courtroom.

“AI is an extremely influential field of study both inside and outside of the courtroom.”

It may state two separate claims as one.

Efficiency and effective use of limited resources are the most widely cited benefits of AI's use in the courtroom, with 38 sources (68%) indicating that AI can contribute to better resource allocation.

“Efficiency and effective use of limited resources are the most widely cited benefits of AI's use in the courtroom, with 38 sources (68%) indicating that AI can contribute to better resource allocation.”

Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.

The COMPAS dataset is more than 12 years old as of the writing of this paper, and many reforms have been made to risk assessment, some inspired by the shortcomings identified in COMPAS.

“The dataset is more than 12 years old as of the writing of this paper, and many reforms have been made to risk assessment, some inspired by the shortcomings identified in COMPAS.”

Probably not a substantial claim.

Limiting the investigation to the adult justice system gives the strongest and most consistent conclusions.

“The analysis presented in this work focuses on the uses of AI in risk assessment rather than on the actual applications of justice, so limiting the investigation to the adult justice system gives us the strongest and most consistent conclusions.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The list of synonyms for AI or algorithmic tools was compiled by conducting an initial review of the literature before collecting the list of included papers.

“This list was compiled by conducting an initial review of the literature before collecting our list of included papers.”

It may state two separate claims as one.

While a belief exists in the literature that deep learning will be applied to risk assessment in the future, the risks, lack of interpretability, and need for large training data sets currently outweigh the potential benefits.

“While a belief does exist in the literature that deep learning will be applied to risk assessment in the future [22], the risks, lack of interpretability [23], and need for large training data sets [24] currently outweigh the potential benefits.”

• cites [22] Cyphert, 2020, Reprogramming recidivism: The first step act and algorithmic prediction of risk, Seton Hall Law Rev. sourceNot checked — no full text of the source
• cites [23] McKay, 2020, Predicting risk in criminal procedure: Actuarial tools, algorithms, AI and judicial decision-making, Curr. Issues Crim. Justice sourceNot checked — no full text of the source
• cites [24] Berk, R., Berk, D., and Drougas, D. (2019). Machine Learning Risk Assessments in Criminal Justice Settings, Springer. sourceNot checked — no full text of the source

Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.

It is difficult to reach any solid conclusions about ARA tools themselves because the nature of the tools varies widely from state to state and is often not open to public review.

“However, it is difficult to reach any solid conclusions about ARA tools themselves because the nature of the tools varies widely from state to state and is often not open to public review.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.

The review's protocol is not registered, but care has been taken to develop a source collection protocol that is thorough, complete, and resistant to bias.

“This protocol for this review is not registered but care has been taken to develop a source collection protocol that is thorough, complete, and resistant to bias.”

Probably not a substantial claim.

Transparency and interpretability are highly valued in risk assessment tools.

“Transparency and interpretability are highly valued in risk assessment tools, which may be why classical ML techniques are preferred in this context.”

Probably not a substantial claim.

Publications that received peer review in an academic publication were treated by the authors as being of equal validity for the purposes of collection.

“Publications that received peer review in an academic publication were treated by the authors as being of equal validity for the purposes of collection our results.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Research in the early 2000s indicated that anywhere from 75% to 97% of probation, parole, and community corrections agencies were using a risk assessment tool.

“There has been rapid and expansive growth in RNA tools over the last hundred years, with research in the early 2000s indicating that anywhere from 75% to 97% of probation, parole, and community corrections agencies were using a risk assessment tool [13].”

• cites [13] Taxman, F.S. (2017). Risk and needs assessment in probation and parole: The persistent gap between promise and practice. Handbook on Risk and Need Assessment: Theory and Practice, Routledge.Not checked yet

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The review included only works that focus on the adult, rather than juvenile, legal system.

“Focuses on the adult, rather than juvenile, legal system”

Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it. It may state two separate claims as one.

ARA tools may be perceived by researchers and the general public as inherently biased or introducing greater bias than other RNA tools, when the reality might simply be the reinforcement of existing implicit biases.

“As a result, ARA tools may be perceived by researchers and the general public as inherently biased or introducing greater bias than other RNA tools, when the reality might simply be the reinforcement of existing implicit biases.”

Probably not a substantial claim.

The review included only works that specifically focus on the application of ARA in the United States legal system.

“Specifically focuses on the application of ARA in the United States legal system”

Probably not a substantial claim. The paper may give no reason why it holds.

Conclusions that may be clear using data collected from one country do not necessarily hold in another.

“Conclusions that may be clear using data collected from one country do not necessarily hold in another.”

Probably not a substantial claim.

The scientific standard of peer review was used as the quality assessment for included works.

“The scientific standard of peer-review has been used in this investigation as our quality assessment for included works.”

Probably not a substantial claim.

The review excluded self-published works.

“self-published works have been excluded from this study.”

It may state two separate claims as one.

ARA tools use criminal records as a major factor in the calculation, thereby solidifying the effect of said bias.

“ARA tools use criminal records as a major factor in the calculation, thereby solidifying the effect of said bias.”

It may state two separate claims as one.

RNA tools are used in almost every decision point in the criminal justice system with the goal of offering a uniform, evidence-based approach to properly target resources and increase public safety.

“Risk need assessment (RNA) tools are used in almost every decision point in the criminal justice system with the goal of offering a uniform, evidence-based approach to properly target resources and increase public safety [11,12].”

• cites [11] Gottfredson, 2006, Statistical risk assessment: Old problems and new applications, Crime Delinq. sourceNot checked — no full text of the source
• cites [12] Taxman, F.S. (2017). The value and importance of risk and need assessment (RNA) in corrections and sentencing: An overview of the handbook. Handbook on Risk and Need Assessment: Theory and Practice, Routledge. sourceNot checked — no full text of the source

Probably not a substantial claim.

The review excluded book chapters.

“Because of this, book chapters have been excluded from this survey.”

Probably not a substantial claim.

The review excluded papers whose intention is primarily persuasive.

“Therefore, papers whose intention is primarily persuasive have been excluded from this investigation.”

It may state two separate claims as one.

The fifth generation of risk assessment tools introduced ML to streamline decisions and attempt to increase predictive accuracy.

“Introduced in the late 1990s and early 2000s, the fifth generation of risk assessment tools introduced ML to streamline decisions and attempt to increase predictive accuracy [12,13].”

• cites [12] Taxman, F.S. (2017). The value and importance of risk and need assessment (RNA) in corrections and sentencing: An overview of the handbook. Handbook on Risk and Need Assessment: Theory and Practice, Routledge. sourceNot checked — no full text of the source
• cites [13] Taxman, F.S. (2017). Risk and needs assessment in probation and parole: The persistent gap between promise and practice. Handbook on Risk and Need Assessment: Theory and Practice, Routledge.Not checked yet

It may state two separate claims as one.

A more accurate explanation of racial disparity in the criminal justice system is implicit bias in the administration of the criminal justice system and the compounding effect of past discriminatory decisions and policies.

“A more accurate explanation of such racial disparity is implicit bias in the administration of the criminal justice system and the compounding effect of past discriminatory decisions and policies [27,29].”

• cites [27] Clemons, 2014, Blind injustice: The Supreme Court, implicit racial bias, and the racial disparity in the criminal justice system, Am. Law Rev.Not checked yet
• cites [29] Butler, 2013, Poor people lose: Gideon and the critique of rights, Yale Law J.Not checked yet

Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.

ML finds patterns in data without direction or insight from human analysts, which puts it at a greater risk for capturing bias recorded in historical data, misaligning our tools with their intentions.

“ML finds patterns in data without direction or insight from human analysts, which puts it at a greater risk for capturing bias recorded in historical data, misaligning our tools with their intentions.”

Probably not a substantial claim. The authors may not put it forward as true. It may state two separate claims as one.

ARA tools can reinforce human bias and inequality, as the historic data used to train models often reflect biased policing and judicial practices.

“Of course, others have argued that ARA tools can reinforce human bias and inequality, as the historic data used to train models often reflect biased policing and judicial practices [14].”

• cites [14] Kaur, 2024, AI in criminal justice: Implications for justice, fairness, and potential biases, Nyaayshastra Law Rev.Not checked yet

Probably not a substantial claim. The paper may give no reason why it holds.

The extent to which ARA tools' claims to incorporate AI are true is not entirely clear, owing largely to a lack of a unified definition for AI across the many disparate disciplines engaged in reviewing and discussing ARA tools.

“The extent to which these claims are true is not entirely clear, owing largely to a lack of a unified definition for AI across the many disparate disciplines that are currently engaged in reviewing and discussing ARA tools.”

Probably not a substantial claim.

The review included only works published in a peer-reviewed, highly regarded journal or conference with a focus on law, criminal justice, or computer science.

“Published in a peer-reviewed, highly regarded journal or conference with a focus on law, criminal justice, or computer science”

It may state two separate claims as one.

Deep learning models infrequently appear in ARAs, but simple tools share many of the same dangers as larger models.

“Deep learning models infrequently appear in ARAs [19], but the simple tools share many of the same dangers as larger models [20].”

• cites [19] Partnership on AI (2019). Report on Algorithmic Risk Assessment Tools in the US Criminal Justice System, Partnership on AI.Not checked yet
• cites [20] 2022, Algorithmic governance from the bottom up, BYU Law Rev.Not checked yet

Probably not a substantial claim.

Risk assessment tools are used in criminal justice to evaluate an individual's likelihood of reoffending by calculating a risk score based on personal history, criminal record, and other causal factors.

“Risk assessment tools are used in criminal justice to evaluate an individual's likelihood of reoffending by calculating a risk score based on personal history, criminal record, and other causal factors [1,2].”

• cites [1] Matz, A.K. (2022). The Debate and Concerns of Risk Assessment with Historically Marginalized Populations. Handbook on Inequalities in Sentencing and Corrections Among Marginalized Populations, Routledge. sourceNot checked — no full text of the source
• cites [2] Kang, 2023, False positives vs. false negatives: Public opinion on the cost ratio in criminal justice risk assessment, J. Exp. Criminol. sourceNot checked — no full text of the source

Probably not a substantial claim.

The review excluded works that were themselves summaries, such as survey papers or reviews of existing literature.

“Therefore, works which were themselves summaries, e.g., survey papers or reviews of existing literature, have been excluded.”

Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.

The specifics of ARA tools are rapidly changing, meaning that solid conclusions from just a few years ago may need to be re-reviewed for consistency in the current judicial landscape.

“The specifics of ARA tools are also rapidly changing, meaning that solid conclusions from just a few years ago may need to be re-reviewed for consistency in the current judicial landscape.”

Probably not a substantial claim.

The review included only works that applied transparent investigative techniques.

“Applied transparent investigative techniques”

It may state two separate claims as one.

Actuarial and statistical assessment models became the standard, as they are considered more accurate, consistent, and efficient than human judgment, and less likely to introduce bias.

“Actuarial and statistical assessment models became the standard, as they are considered more accurate, consistent, and efficient than human judgment, and less likely to introduce bias [13].”

• cites [13] Taxman, F.S. (2017). Risk and needs assessment in probation and parole: The persistent gap between promise and practice. Handbook on Risk and Need Assessment: Theory and Practice, Routledge.Not checked yet

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

The search was conducted jointly by both authors, with either author being able to exclude works that failed to meet the inclusion criteria or did meet the exclusion criteria.

“The search was conducted jointly by both authors, with either author being able to exclude works that either failed to meet the inclusion criteria, or did meet the exclusion criteria.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

There is widespread interest in adapting ML tools to address the concerns of this problem space.

“This demonstrates that there is widespread interest in adapting ML tools to address the concerns of this problem space.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

LLMs and other NLP techniques have applications in the courtroom for transcribing and summarizing conversations.

“Even though LLMs and other techniques from natural language processing are not currently being used for risk assessments, they do have applications in the courtroom for transcribing and summarizing conversations [3] and for obfuscating race-related information in prosecutorial documents [37].”

• cites [3] Rowland, 2023, AI in Corrections: The Basics and a Way to Experiment, Fed. Probat.Not checked yet

Probably not a substantial claim. The paper’s conclusions may not depend on it.

This benefit is a possibility that is largely dependent on the implementation of the ARA tool being used.

“It is important to remember that this benefit is a possibility that is largely dependent on the implementation of the ARA tool being used.”

It may state two separate claims as one.

If risk assessments are believed to be a purely mechanical or computational process, and therefore immune to expressing bias along lines of race, gender, age, and wealth, then those biased assessments are more likely to be accepted at face value, rather than investigated for any potential expressions of unfairness.

“If risk assessments are believed to be a purely mechanical or computational process, and therefore immune to expressing bias along lines of race, gender, age, and wealth, then those biased assessments are more likely to be accepted at face value, rather than investigated for any potential expressions of unfairness.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.

Continued improvements in AI-enabled risk assessments require the regular collection of fresh data.

“Continued improvements in AI-enabled risk assessments require the regular collection of fresh data.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.

Proxy variables can be difficult to identify without independent exploratory factor analysis.

“Proxy variables can be difficult to identify without independent exploratory factor analysis, again highlighting the importance of human judgment and review in the process of developing ARA tools.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

LLMs and other NLP techniques have applications in the courtroom for obfuscating race-related information in prosecutorial documents.

“Even though LLMs and other techniques from natural language processing are not currently being used for risk assessments, they do have applications in the courtroom for transcribing and summarizing conversations [3] and for obfuscating race-related information in prosecutorial documents [37].”

• cites [37] Chohlas-Wood, A., Nudell, J., Yao, K., Lin, Z., Nyarko, J., and Goel, S. (2021, January 19–21). Blind justice: Algorithmically masking race in charging decisions. Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society, Virtual. sourceNot checked — no full text of the source

It may state two separate claims as one.

Sufficiently large datasets for training ML span a large data collection period and risks capturing unfair or unconstitutional practices.

“Sufficiently large datasets for training ML span a large data collection period and risks capturing unfair or unconstitutional practices [17].”

• cites [17] Elyounes, 2019, Bail or jail? Judicial versus algorithmic decision-making in the pretrial system, Colum. Sci. Technol. Law Rev.Not checked yet

It may state two separate claims as one.

AI and ML have been influential on the development of risk assessments, even if the newest tools of AI have yet to make an appearance.

“AI and ML have been influential on the development of risk assessments, even if the newest tools of AI have yet to make an appearance.”

Probably not a substantial claim. It may be too vague to check against evidence. It may state two separate claims as one.

This is not true of all ML models, but it does indicate that specifically interpretable models should be applied if ML is going to be used in ARA.

“This is not true of all ML models, but it does indicate that specifically interpretable models should be applied if ML is going to be used in ARA.”

It may state two separate claims as one.

The high stakes of criminal justice make courts slow to adopt new technologies, and the state of the art in AI has little impact on criminal risk assessment.

“The high stakes of criminal justice make courts slow to adopt new technologies [52], and the state of the art in AI has little impact on criminal risk assessment.”

• cites [52] Novokmet, 2022, Pretrial risk assessment instruments in the US criminal justice system—What lessons can be learned for the European Union, Int. J. Law Inf. Technol. sourceNot checked — no full text of the source

It may state two separate claims as one.

The COMPAS dataset is now more than a decade old, and it is not clear that its performance is exemplary of the ARA tools currently in use by American courts.

“The COMPAS dataset is now more than a decade old, and it is not clear that its performance is exemplary of the ARA tools currently in use by American courts.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.

Judges with access to risk assessment tools are not always beholden to them and are usually free to utilize an individual's risk score in whatever way they feel is appropriate at the time.

“Judges with access to risk assessment tools are not always beholden to them and are usually free to utilize an individual's risk score in whatever way they feel is appropriate at the time.”

It may state two separate claims as one.

ARA tools that are specifically identified as ML are frequently developed by statisticians based on an analysis of historical records, and so can still be impacted if or when those data have captured the biased performance of human risk assessors.

“Even ARA tools that are specifically identified as ML are frequently developed by statisticians based on an analysis of historical records, and so can still be impacted if or when those data have captured the biased performance of human risk assessors.”

It may state two separate claims as one.

The risk assessments themselves undergo a lengthy period of development and validation, and then must operate in the wild for several years before large datasets can be collected.

“The risk assessments themselves undergo a lengthy period of development and validation [47], and then must operate in the wild for several years before large datasets can be collected.”

• cites [47] Hamilton, 2016, Designed to fit: The development and validation of the STRONG-R recidivism risk assessment, Crim. Justice Behav. sourceNot checked — no full text of the source

It may state two separate claims as one.

The SRA had to be updated because, as a purely data-driven tool, it had captured some confusing and illogical patterns in its training data.

“While the LSI was updated largely to improve its predictive performance [47], the SRA had to be updated because, as a purely data-driven tool, it had captured some confusing and illogical patterns in its training data.”

• cites [47] Hamilton, 2016, Designed to fit: The development and validation of the STRONG-R recidivism risk assessment, Crim. Justice Behav. sourceNot checked — no full text of the source

Probably not a substantial claim. It may be too vague to check against evidence.

Tools that learn patterns in their training data without human guidance have a particular need for human review before they can be used with faith in real courts.

“Tools that learn patterns in their training data without human guidance have a particular need for human review before they can be used with faith in real courts [17].”

• cites [17] Elyounes, 2019, Bail or jail? Judicial versus algorithmic decision-making in the pretrial system, Colum. Sci. Technol. Law Rev.Not checked yet

It may state two separate claims as one.

Kentucky moved towards the PSA risk-assessment tool in 2013, although this move did not appreciably improve outcomes.

“This disappointing outcome led Kentucky to move towards the PSA risk-assessment tool in 2013, although this move did not appreciably improve outcomes [4].”

• cites [4] Stevenson, 2018, Assessing risk assessment in action, Minn. Law Rev. sourceNot checked yet

It may state two separate claims as one.

Deep learning has shown no additional predictive power in this particular problem space while suffering from additional needs for training data.

“Deep learning has shown no additional predictive power in this particular problem space while suffering from additional needs for training data [24], while exacerbating existing concerns regarding interpretability and transparency [33].”

• cites [24] Berk, R., Berk, D., and Drougas, D. (2019). Machine Learning Risk Assessments in Criminal Justice Settings, Springer. sourceNot checked — no full text of the source
• cites [33] Jorgensen, 2022, Algorithms and the individual in criminal law, Can. J. Philos. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

It may state two separate claims as one.

While frameworks exist to improve the interpretability of complex ML, these frameworks have done little to quell the concerns of authors in the legal space.

“While frameworks exist to improve the interpretability of complex ML [63], these frameworks have done little to quell the concerns of authors in the legal space.”

• cites [63] Nguyen, H.T.T., Cao, H.Q., Nguyen, K.V.T., and Pham, N.D.K. (2021, January 17–21). Evaluation of explainable artificial intelligence: Shap, lime, and cam. Proceedings of the FPT AI Conference, Virtual.Not checked yet

Probably not a substantial claim. The paper’s conclusions may not depend on it.

One source noted that with correct and unbiased training data, even COMPAS performs without bias.

“One source noted that with correct and unbiased training data, even COMPAS performs without bias [36].”

• cites [36] Wang, 2023, In pursuit of interpretable, fair and accurate machine learning for criminal recidivism prediction, J. Quant. Criminol. sourceNot checked — no full text of the source

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

One study showed that the application of simple AI tools to the task of predicting recidivism can reduce crime by 24.8% while also reducing the incarcerated population by 42.0%.

“One study showed that the application of simple AI tools to the task of predicting recidivism can reduce crime by 24.8% while also reducing the incarcerated population by 42.0% [5].”

• cites [5] Kleinberg, 2018, Human decisions and machine predictions, Q. J. Econ. sourceNot checked yet

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.

It is commonly understood that some attributes describing defendants should be omitted from consideration in judging the risk presented by a defendant, either because they are constitutionally protected in the United States or because they are immaterial to the assessment.

“It is commonly understood that some attributes describing defendants should be omitted from consideration in judging the risk presented by a defendant, either because they are constitutionally protected in the United States or because they are immaterial to the assessment.”

Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.

Court records must be collected by independent researchers, which itself is a slow process.

“Court records must be collected by independent researchers, which itself is a slow process.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Northpoint, the company that developed COMPAS, has disputed claims of bias in COMPAS.

“These analyses uncovered that COMPAS recommended 24% longer sentences for young defendants [35] and that COMPAS was more than twice as likely to incorrectly identify a black defendant as a future criminal as compared to a white defendant [58] although Northpoint, the company that developed COMPAS, has disputed claims of bias in COMPAS [58].”

• cites [58] 2019, Algorithmic risk governance: Big data analytics, race and information activism in criminal justice debates, Theor. Criminol. sourceNot checked — no full text of the source

Probably not a substantial claim.

The review focused entirely on the justice system within the United States.

“First, we have focused entirely on the justice system within the United States.”

Probably not a substantial claim.

COMPAS now stands as an example of the harms and biases that can arise from ARA and other AI tools when trained on biased data, rather than as an example of the harms and biases that do currently exist within the justice system.

“If anything, COMPAS now stands as an example of the harms and biases that can arise from ARA and other AI tools when trained on biased data, rather than as an example of the harms and biases that do currently exist within the justice system.”

It may state two separate claims as one.

The abundance of data related to COMPAS makes it one of the easiest tools to analyze, but it does not necessarily make it the most representative ARA tool.

“While the abundance of data related to COMPAS makes it one of the easiest tools to analyze, it does not necessarily make it the most representative ARA tool.”

Probably not a substantial claim. It may state two separate claims as one.

The review focused exclusively on the adult justice system, omitting several papers whose focus was on the juvenile justice system.

“Second, we have chosen to focus exclusively on the adult justice system, omitting several papers whose focus was on the juvenile justice system.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Civil rights groups have expressed concerns regarding ARA usage.

“Civil rights groups have expressed concerns regarding their usage [67].”

• cites [67] Green, 2021, Algorithmic risk assessments can alter human decision-making processes in high-stakes government contexts, Proc. ACM Hum. Comput. Interact. sourceNot checked yet

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

ARA tools have existed at least since the 1970s, and many of the early tools were simple actuarial tools rather than AI or ML.

“While ARA tools have existed at least since the 1970s [22], many of the early tools were simple actuarial tools rather than AI or ML.”

• cites [22] Cyphert, 2020, Reprogramming recidivism: The first step act and algorithmic prediction of risk, Seton Hall Law Rev. sourceNot checked — no full text of the source

It may state two separate claims as one.

Because the predictive policing software currently on the market is closed-source and proprietary, it is impossible to know whether interpretable models are being used, or even what models are being used.

“Because the predictive policing software currently on the market is closed-source and proprietary [70], it is impossible to know whether interpretable models are being used, or even what models are being used.”

• cites [70] Wexler, 2018, Life, liberty, and trade secrets: Intellectual property in the criminal justice system, Stan. Law Rev. sourceNot checked — no full text of the source

It may state two separate claims as one.

Judges express reluctance to adopt new risk assessments in their courtrooms, expressing concerns that risk assessments limit judicial options and undermine elected officials.

“Likewise, judges express reluctance to adopt new risk assessments in their courtrooms [55], expressing concerns that risk assessments limit judicial options [55] and undermine elected officials.”

• cites [55] Pruss, D. (2023, January 12–15). Ghosting the machine: Judicial resistance to a recidivism risk assessment instrument. Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency, Chicago, IL, USA. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

Although ML has been shown to predict crime and recidivism with greater reliability than humans, these products require additional testing and public scrutiny before they can reliably be used in the real world.

“Although ML has been shown to predict crime and recidivism with greater reliability than humans [49], it is also clear that these products require additional testing and public scrutiny before they can reliably be used in the real world.”

• cites [49] DeMichele, 2021, The intuitive-override model: Nudging judges toward pretrial risk assessment instruments, Fed. Probat.Not checked yet

Probably not a substantial claim.

The other results presented in Section 5 are similarly robust to the presence of un-collected sources.

“The other results presented in Section 5 are similarly robust to the presence of un-collected sources.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

It is possible that some appropriate publications were not collected during the query process.

“Although care was taken to ensure that this review is complete by using broad queries to search well-supplied sources of literature, it is possible that some appropriate publications were not collected during our query process.”

It may state two separate claims as one.

Although deep learning algorithms are not being widely used for ARA, they have begun to appear in the newer field of predictive policing.

“Although deep learning algorithms are not being widely used for ARA [19], they have begun to appear in the newer field of predictive policing [20].”

• cites [19] Partnership on AI (2019). Report on Algorithmic Risk Assessment Tools in the US Criminal Justice System, Partnership on AI.Not checked yet
• cites [20] 2022, Algorithmic governance from the bottom up, BYU Law Rev.Not checked yet

It may state two separate claims as one.

AI-derived tools might be better predictors of recidivism than judges and can be less biased than human actors.

“While AI-derived tools might be better predictors of recidivism than judges [6] and can be less biased than human actors [6], these benefits are meaningless if the recommendations of AI systems are ignored.”

• cites [6] Ludwig, 2021, Fragile algorithms and fallible decision-makers: Lessons from the justice system, J. Econ. Perspect. sourceNot checked — no full text of the source

It may state two separate claims as one.

Because of the disconnect between political and judicial priorities, risk assessment tools often take many years to develop, and their risk scores are often ignored by judges.

“Because of the disconnect between political and judicial priorities, risk assessment tools often take many years to develop [24], and their risk scores are often ignored by judges [55].”

• cites [24] Berk, R., Berk, D., and Drougas, D. (2019). Machine Learning Risk Assessments in Criminal Justice Settings, Springer. sourceNot checked — no full text of the source
• cites [55] Pruss, D. (2023, January 12–15). Ghosting the machine: Judicial resistance to a recidivism risk assessment instrument. Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency, Chicago, IL, USA. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

No formal inter-rater agreement statistic was calculated, though the authors discussed and reviewed each others' assessments informally.

“No formal inter-rater agreement statistic was calculated, though the authors discussed and reviewed each others' assessments informally.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

This review summarizes the conclusions presented by the collected sources, and while some observations may apply to broader populations, the authors make no explicit claim of that.

“This review summarizes the conclusions presented by the collected sources, and while some observations may apply to broader populations, we make no explicit claim of that.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Explainable AI is a field of study that produces AI models whose reasoning is transparent to outside scrutiny.

“Explainable AI is a field of study that produces AI models whose reasoning is transparent to outside scrutiny [69].”

• cites [69] Longo, 2024, Explainable Artificial Intelligence (XAI) 2.0: A manifesto of open challenges and interdisciplinary research directions, Inf. Fusion sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

It may state two separate claims as one.

ARA predictions are correct when viewed through the lens of the training data but still reflect bias inherent in that data.

“This results in ARA predictions that are correct when viewed through the lens of the training data, but still reflect bias inherent in said data.”

Probably not a substantial claim.

The review limited the time over which papers were collected to one where AI has been influential.

“Finally, we have limited the time over which papers were collected to one where AI has been influential.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Queries and paper assessments were performed and reviewed by both authors independently to help ensure the completeness of this review.

“Queries and paper assessments were performed and reviewed by both authors independently to help ensure the completeness of this review.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

Some products currently offered for use by police departments include Gotham, HunchLab, and PredPol, all made by Palantir, which is headquartered in Miami, FL.

“Some of the products currently offered for use by police departments include: Gotham, HunchLab, and PredPol, all of which were made by Palantir [33], which is headquartered in Miami, FL.”

• cites [33] Jorgensen, 2022, Algorithms and the individual in criminal law, Can. J. Philos. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

It may state two separate claims as one.

Pre-trial detention increases the likelihood of conviction and future crime.

“The financial inaccessibility of bail, for example, increases the likelihood of pre-trial detention, which increases the likelihood of conviction and future crime [66].”

• cites [66] Copp, 2022, Pretrial risk assessment instruments in practice: The role of judicial discretion in pretrial reform, Criminol. Public Policy sourceNot checked — no full text of the source

It may state two separate claims as one.

COMPAS was the subject of a 2016 ProPublica article, which gave it a much higher public profile than other ARA tools.

“COMPAS was the subject of a 2016 ProPublica article [65], which gave it a much higher public profile than other ARA tools.”

• cites [65] Washington, 2018, How to argue with an algorithm: Lessons from the COMPAS-ProPublica debate, Color. Technol. Law J.Not checked yet