Getting students to engage in class is tough. But criminal psychology has a secret weapon: it’s an interesting real-world phenomenon! Case studies can bring that to the fore. But who has the time to generate reams of case documentation, right? Actually, AI does…
This semester, I’ve been using AI a lot in my community corrections class. A lot of the topics we cover lend themselves perfectly to case studies: presentence investigation reports, sentencing guidelines, risk assessment, case management plans, parole board decisions. Having students work through those concepts for themselves as a practical exercise is a fantastic way to achieve learning outcomes and make classes that little bit more interactive.
But, as noted at the top, all of those case study exercises require cases. And those cases need documentation.
Take a parole board exercise, where the students work together in small groups to review relevant case documentation to make a parole decision for a fictitious inmate being released back into the community from prison. What documentation does a parole board typically use? Or, more precisely, what documentation do I need to produce to give my students a minimally viable ersatz parole board member experience?
- A Presentence Investigation Report (PSI), including details about the individual’s criminal history, personal background, offense details, and information about their behavior while incarcerated.
- Some form of institutional behavior records, detailing the behavior of the individual while serving their sentence, including disciplinary actions, participation in rehabilitation programs, and any other relevant conduct. This might also include testimonials from prison staff or counselors about the individual’s behavior, participation in programs, and their prospects for successful reintegration into society.
- Some letters of support or opposition, from various sources, such as family members, friends, victims, or concerned parties, presenting arguments either in favor of or against parole. This might also include a formal Victim Impact Statement from the court case.
- A parole plan for the individual’s release, including where they’ll live, employment opportunities, access to support services, and plans for continued rehabilitation or treatment.
- One or more risk assessment reports, evaluating the individual’s risk of reoffending and may include assessments of their mental health, likelihood of rehabilitation, and potential threat to the community.
Those are probably your minimum. At the very least that’s probably a dozen pages of fictional material that you need to create. And you need them all to be interesting, realistic, non-contradictory, ambiguous enough to require critical thought, and have a narrative leading the student to some form of appropriate decision.
That’s days (maybe weeks) of intellectual graft.
This is where your favorite artificial intelligence (AI) tool steps in. When it’s not plotting the downfall of humankind or doing the homework you’ve assigned to your students, AI can produce any type of fictitious case documentation you need it to (within its terms of use). And it can do it in mere seconds, if you give it the right information. In fact, the bulleted list of required documentation you just read above was generated for me, in seconds, by AI. (Ha, fooled ya!)
All you need to do is provide a good prompt. A prompt is the text description you provide to a language-based AI tool for what you want it to do. Developing high-quality prompts that maximize the chances of you getting a good outcome is called prompt engineering. There is myriad “cheat sheets” online for how to do prompt engineering, but they typically recommend structuring your initial prompt with three elements:
- The role you want the AI to play.
- The task you want the AI to complete in this role.
- The format in which you want the result to be produced.
I also found the useful CREATE formula:
- Character: define the AI’s role.
- Request: be specific about what you need.
- Examples: (optional) but providing examples of what you want can help.
- Adjustments: refine the prompt if it doesn’t produce the desired output.
- Type: describe the type and format out the output.
- Extras: incorporate any unique requirements.
For example, you could start with:
You are a criminal justice professor at an elite college in the United States. You are preparing to teach a class on re-entry from custody. As a way to illustrate the concept of a parole board, you would like your students to act as a parole board members, who will review a range of relevant documentation for a fictional case of an adult male inmate returning to the community after a prison sentence. Give him a name. He was convicted of federal armed robbery. Produce a Pre-Sentence Report for this fictional prisoner that includes details about his criminal history, personal background, offense details, and information about their behavior while incarcerated. Generate criminal history and personal information commensurate with someone who was convicted of armed robbery. He should have a reasonably extensive criminal history. The PSI report should be separated into relevant headed sections and ensure that no piece of information contradicts any information in another section. The report should be approximately three pages in length when transferred to word processing software, using single-spaced text and based on US letter-sized paper. Each section should contain at least four sentences of relevant information.
This should allow your AI tool to produce some initial fictional case information. If you’re using a chat-based AI tool (like ChatGPT), you can ask it to tweak elements of your case too, like:
Provide more demographic information for the inmate: a date of birth, sex, ethnicity, and any other variable that would typically be included as demographic information on an official form.
Or:
Without changing the format or the sections of the Pre-sentence Report you have just generated, generate one for a new fictional prisoner that is female, where the conviction was for assault with a deadly or dangerous weapon, and where the defendant has a much shorter criminal history. Then provide the same additional demographic information for this new prisoner.
Boom. Now you’ve got two!
In a chat-based AI environment, you can simply edit the process for any other document you want to replicate. For example, you’ll need some risk assessment documentation for your first case. I used two further prompts, which I’ll tastefully censor, given that I might have asked about a proprietary commercial product:
List the predictors, and any related scores or weightings, for the [insert risk assessment tool]
Based on those predictors, scores, and weightings you have just listed, generate a [insert risk assessment tool] for my original fictitious male case. Ensure that any scores or information that you generate for this assessment are consistent with any other case information you have already generated. Also provide a text description of evidence supporting any score that you generate. At the end of the assessment provide an accurately calculated total numerical score for the whole assessment. Use that score to estimate the category of risk that the score would be associated with: low risk, medium risk, high risk, or very high risk.
If necessary, again you can use further prompts to tweak the results, like:
Include more supporting evidence for each variable.
Other examples would be the testimonies of others:
Please generate a letter of support from the family of the fictitious male prisoner, addressed to the parole board.
Please generate a letter of opposition to the release of the fictitious male prisoner. The author of this letter should be the Chief Operating Officer of a fictitious company that was the victim of the robbery for which the prisoner was convicted. It should emphasize the traumatic effect that the robbery had on those members of staff who were present at the particular location when the robbery took place. Give the company an appropriate name.
None of these prompts are perfect. Someone who considers themselves a prompt engineer would probably find faults throughout. Nevertheless, in less than one hour I had some document files filled with relevant, interesting, and, for the most part, consistent documents for my class exercise.
