Implications for Artificial Intelligence in Forensic Psychological Practice and Board Certification

Current Uses of AI in Forensic Practice

Artificial intelligence (AI) is a term with multiple definitions, although is generally identified as programs, algorithms, and/or technology designed to perform activities that require human intelligence (e.g., content generation, decision-making, analysis, etc.). Unlike human intelligence, machine-based learning and decision-making reduces error, including within the practice of medicine. Even though AI is used for life-saving medical operations, its widespread use is inhibited due to difficulties in understanding the underlying algorithms, the liability of using AI to diagnose, and poor security of patient data (Matsuzaki, 2018).

AI use has been documented in the mental health field since the 1960s, with the advent of ELIZA, a computer program designed to replicate Rogerian communication styles and interact with others. Currently, AI programs often perform administrative tasks, such as writing emails or articles for private practitioners, onboarding employees, and filling in calendars. From the standpoint of a forensic practitioner, the number of applications and uses for AI is expansive. While few forensic practitioners may quibble over such assistance, some programs, like AutoNotes (2024), can assist with forensic session notes (e.g., interviews, testing) and treatment planning (e.g., risk management plans, mitigation evaluation recommendations), whereas others review the contents of the examinee’s personal history and clinical file to suggest relevant diagnoses (Minerva & Giubilini, 2023). Some practitioners use AI to interpret test data or jail records, similar to how some medical professionals use AI to summarize medical charts or hospital records. The rapid advent of AI into psychological practice invites numerous questions about the future potential of this technology in forensic psychology.

Future Directions of AI in Forensic Practice, Research, and Policy

Potential Advantages

Advancements in AI platforms often occur in fits and starts, resulting in large shifts in the quality of output and quantity of data analytic capacities in short periods of time. In addition to the services described earlier, current AI programs could be modified to assist with other forensic tasks, such as transcribing forensic interviews, synthesizing multiple records, summarizing media files pertaining to discovery (e.g., body-worn camera footage), or identifying empirical articles to bolster report conclusions. AI platforms are also in development to track natural language to identify risk factors and prevention strategies for diseases (Baclic et al., 2020). These algorithms have the potential for comparable forensic research like threat assessment (e.g., language content analysis) and risk assessment (e.g., identifying unique risk factors and evaluating risk management effectiveness). AI could also be used to generate prevalence data in forensically-relevant populations or to analyze other types of data that would facilitate policy change (e.g., an algorithm distilling evaluator reports and/or medical charting of forensic hospital patients). Across research, practice, and policy development domains, repetitive tasks like record reviews could become automated, resulting in increased efficiency and providing evaluators with time and energy to focus on innovation and critical thinking.

Areas for Concern

Despite the advantages of AI, forensic users need to be aware of the limitations of the current tools available. For example, AI technology has infamously been highlighted as racially biased due to the databases from which they draw (Lohr, 2018), resulting in reluctance by larger public health entities to adopt AI platforms to reduce healthcare disparities (Nazer et al., 2023). Similarly, when ChatGPT was used as a “researcher,” it failed in its task without the guidance of a human assistant (Jungwirth & Haluza, 2023). Such limitations suggest AI is currently ill-equipped to address complex forensic issues, such as assessing violence risk or answering the psycholegal question. Comparable criticisms exist regarding embodied AI and advanced therapeutic questions (Fiske et al, 2019). While AI may be suitable for generating images or text alone (e.g., summarizing records), it is likely to generate errors if asked to synthesize or interpret mental health records or court documents. Furthermore, AI is unable to synthesize observational data during forensic interviews or video-recorded interrogations.

ChatGPT exhibits limited capacities when tasked with abstract or critical thinking. For example, we asked ChatGPT to interpret a Total Score on the Inventory of Legal Knowledge (ILK). The ILK is a 61-item binomial forced-choice test of response style in evaluations of adjudicative competence. Specifically, it assesses feigned legal knowledge deficits. Scores below 47 are suggestive of a feigned response style and scores below 24 suggest deliberate selection of the wrong answer since those score values fall below what would be expected by chance when considering a 95% confidence interval. When asked to interpret an ILK Total Score of 10, ChatGPT provided a blatantly wrong response. It stated, “A score of 10 on the Inventory of Legal Knowledge indicates a severely limited understanding of legal concepts and processes” and “a significant deficit in the comprehension necessary for participating in legal proceedings.” Thus, for all potential advantages, AI platforms inconsistently recognize their own limitations or when human assistance may be required, leaving inexperienced users at risk of taking the program’s output as fact and not applying critical reasoning to ensure accuracy.

AI and Board Certification in Forensic Psychology

Potential Advantages

AI has the potential to influence each stage of the board certification process in forensic psychology, both positively and negatively. Regarding the Written Examination, AI could be used to assist with the study process by synthesizing chapters from the American Board of Forensic Psychology (ABFP)’s Suggested Reading List for Written and Oral Examinations, summarizing case law holdings, or generating mock test questions after pooling publicly available sample questions. Similarly, AI could be used by ABFP to assist in the creation of newer versions of the Written Exam. There are already AI-powered writing assistants (e.g., Grammarly) that could be used to modify practice samples in sanctioned ways. AI could be used by the ABFP to streamline the screening process for Practice Samples or standardize the criteria for approving submissions. AI has implications for the Oral Exam as well. It could be used to generate broad, content exam questions by the Oral Examination Committee or summarize research on the Candidate’s areas of expertise for review and preparation. AI could be used by the Committee to transcribe examinations or summarize the Candidate’s performance. Given the emphasis of this step on determining the Candidate’s understanding and application of ethics, it may be relevant for Committees to start asking about the implications of AI in forensic practice during the Oral Examination itself.

Areas for Concern

Less savory Candidates could exploit AI to cheat during the administration of the Written Exam. This potential may be higher with remote administrations, which limit opportunities to monitor the covert use of smartphones and other AI-compatible devices. While some of AI-generated modifications to Practice Samples may be benign (e.g., correcting typos and grammatical mistakes), more significant changes could border on plagiarism. The ABFP Practice Sample Guidelines, updated as recently as May 2024, provide no guidance on the use of AI assistance in submitted samples. Arguably, the instructions regarding “Authorship and Review” bear on this issue. For example, Practice Samples must be of the Candidate’s sole authorship and must not have been reviewed or critiqued by “any other person.” However, the extent to which these restrictions apply to AI is unclear.

Candidates could use AI to inappropriately identify shortcomings in their practice samples that may be addressed during the Oral Exam. Typically, the Oral Exam includes a brief break, during which time the Candidate could use AI to research recently asked Committee questions and return to the exam with “renewed” thoughts on the topic. In light of the above considerations, the ABFP (or ABPP, more generally) should seek to incorporate AI into its certification policies and instructions. 

Ethical and Legal Considerations

The field of forensic psychology has yet to offer suggestions for addressing AI in informed consent, report writing, or expert testimony; thus, the considerations detailed here are offered based on what is known about AI and forensic issues of evidence and transparency. One of the most notable concerns is whether the use of AI, or content generated by AI, will be admissible in court. Different jurisdictions use different standards for making this determination. In Daubert v. Merrell Dow Pharmaceuticals, Inc. (1993), the Supreme Court emphasized that expert testimony should be informed by methods that are peer-reviewed, testable, generally accepted in the field, or have known error rates. Currently, the federal system and most state jurisdictions adhere to this standard. The use of AI may conflict with Daubert’s requirements in that many AI algorithms are proprietary and guarded by private companies, preventing review and critical analysis. Indeed, a pivotal question addressed in State v. Pickett (2021) was whether a private company could be forced to divulge source code for AI genotyping software used to connect a defendant to a murder.  The company argued the code was a “trade secret,” although it was eventually ordered to release the code under seal for review.  Few states employ the Frye standard, which requires the methods used to form an opinion to have “general acceptance” in the field. There is currently no research, case law, or official positions from professional organizations to suggest the use of AI in forensic psychology is “generally accepted.” 

The Specialty Guidelines for Forensic Psychology stress that the competent forensic psychologist produces work that is transparent, with data being provided for review and judicial scrutiny when needed. Forensic psychologists, therefore, have an obligation to accurately communicate the limitations of AI in the context of expert testimony, especially since jurors and judges may rely heavily on the opinions of forensic mental health experts and assume them to be unbiased and accurate in their conclusions (Blackwell & Seymour, 2015; Zapf et al., 2004).  Thus, it seems possible for an expert to discuss the use of AI in their evaluation and have those procedures go unchallenged despite the concerns and limitations noted above.

Concluding Perspectives

The use of AI in forensic practice has great promise. While AI programs currently have the capacity to assist with basic tasks, this technology continues to struggle with systemic biases, flawed logic, and poor abstract reasoning. The implications of AI in forensic ethics, board certification, and admissibility of evidence are expansive and unresolved. Forensic practitioners who use this technology should remain aware of these issues, exercise transparency, and proceed with caution until more cohesive and informed guidance is available.