The System for Reporting Sexual Harassment Is Broken—Here’s How to Fix It

Many people experience harassment at work, but studies show that most never report it. The Equal Employment Opportunity Commission reported in 2016 that between 6 and 13 percent of people who experienced harassment at work filed a formal complaint, meaning 87 to 94 percent of people do not report it. And for those who do report, research shows that investigative processes are often highly contentious: complainants often face retaliation or victimization from the alleged perpetrator or other coworkers. Even the reports themselves can be flawed, with shoddy evidence and inconsistent stories that weaken both the quality of the case and the response to it. In other words, the system for reporting workplace harassment is broken.

So how can we fix it? A 2018 report from the British Equality and Human Rights Commission points to the need for organizations to have an anonymous online reporting option. The commission found that this—combined with strong anti-harassment policies and the construction of a civil workplace culture—was the best way to help improve the reporting process and reporting outcomes, particularly for workplace sexual harassment. When victims can report anonymously, they’ll feel safer and more comfortable coming forward, the authors of the report suggest.

How can we ask people to remember and report workplace harassment without compromising their identities while also building strong cases?

But anonymity and good policies are insufficient if the way that victims share their stories is through a poorly designed web portal or questionnaire. The reporting process itself also needs to help create a strong case, extracting information of high evidentiary quality that will be taken seriously and acted upon. This leads to a bigger question: how can we ask people to remember and report workplace harassment without compromising their identities while also building strong cases? The solution might lie in a combination of memory science and artificial intelligence (A.I.).

A better way to interview victims

As memory scientists who study how people remember important emotional events, we spend a significant amount of time understanding how to get quality information out of people—information that will stand up in a court case.

The manner and skills employed by interviewers invariably influences the experience of interviewees and the outcome of the experience. Research shows that there is a direct correlation between interviewer behaviors and both the quantity and quality of relevant information gained. The goal when interviewing victims or witnesses should be sourcing information that is accurate, timely, and case-relevant in a manner that is effective, efficient, and legally compliant.

But, in a complex and demanding workplace setting, there are inherent dangers associated with poor interview practices. For instance, poorly conducted interviews can not only lead to conviction of the innocent, and potentially help set the guilty free, but may also result in false denials, court acquittals, and, in rare cases, false confessions resulting in miscarriages of justice. Even though it’s important to let people talk freely about what happened without being interrupted, humans find this challenging. Studies have found that during one of the most important parts of the cognitive interview, the free-recall phase, interviewers interrupt interviewees every few seconds. Human interviewers seem to be able to last only between two and ten seconds (on average) before interrupting.

There are two main interview styles that encourage people to open up. The first is called ethical interviewing, or cognitive interviewing. It uses open-ended questions that encourage people to talk freely about what they remember. The second is called accusatorial interviewing, where the interviewer presumes that something has happened and uses questions to confirm their assumptions. Accusatorial interviewing is problematic and often involves closed-ended questions that can make people give unreliable details or confessions.

Research shows that in order to collect complete and accurate information, it is imperative to use cognitive interviewing rather than accusatorial techniques. Most people who work in H.R. and compliance settings are taught how to do investigations, but they are rarely taught the fundamentals of evidence-based memory interviewing. And even those who are taught to use the cognitive interview (or similar procedures) often find themselves straying from what they were trained to do, resulting in being unable to generate reliable accounts.

Could A.I. help?

Online A.I. tools rooted in best interview practices can offer a reliable and scalable way to help people recall and report what happened—while bypassing the biases of their human counterparts. Such tools could provide a judgement-free service that doesn’t rush or fluster users and that is available at any time to employees who wish to report inappropriate behavior.

This accessibility is critical; research shows that if businesses acknowledge reports within 24 hours, they help to create a positive organizational climate. When complainants are satisfied with the organization’s handling of cases, they are less likely to pursue legal redress against an organization. Retaliation is also least likely when complainants make a formal rather than an informal report, something which can be done more easily (and anonymously) using A.I. tools.

A.I. tools could perform better than human interviewers by asking appropriate questions every time, without going off script or asking leading questions.

Reports created by A.I. tools that are capable of gathering information of high evidentiary quality could be taken seriously and acted upon. A.I. tools could perform better than human interviewers by asking appropriate questions every time, without going off script or asking leading questions. One of the critical aspects of the cognitive interview technique is that the free-recall stage should be followed by appropriate “probes” that prompt the user to go into further details. A.I. tools could ask appropriate open-ended probes by picking the key words from the interviewee’s free-recall account. For example, if the user reports, “I was harassed by my boss,” A.I. could pick up the word “boss” and ask an appropriate probe: “Please tell me more about boss.” These probes enhance memory recollection, because the quality of recorded memory is evidence and is critical to being believed.

We created such an A.I. tool—Spot—to help employees record and report workplace harassment and discrimination.

A.I. tools, like Spot, can even help organizations better structure their response procedures, to make the whole reporting process better—more respectful and more effective. For instance, after a user participates in an interview on Spot, the tool creates a certified, private, and time-stamped PDF report that can be used as high-quality evidence in internal grievance procedures. It also allows the organization to respond to complaints directly through Spot even if employees have chosen to stay anonymous, by allowing employers to send a follow-up interview with questions that are also in accordance with the cognitive interview.

Of course, A.I. tools are only one part of what must be a society-wide effort to address sexual harassment. But we can’t address sexual harassment effectively if a vast number of victims still feel unsafe reporting it, or if their stories are mangled by other humans.