Artificial Intelligence and Coercive Control: A Complex Relationship

In an increasingly digital world, both sides in contentious family court cases are turning to artificial intelligence (AI) to analyze diaries and communications with former partners for evidence of abuse. However, is it truly ‘expert evidence’ when it comes to coercive control?

The Dark Side of AI in Domestic Violence Cases

As coercive control scholars and expert witnesses, we’re concerned about the growing tendency to treat AI as an authoritative source for determining ‘what’s really happening’ in domestic violence cases. The problem lies in the potential pitfalls of relying on AI-generated analyses in court proceedings.

Coercive control is a complex and multifaceted issue that involves tactics such as strategic love-like acts, intimidation, isolation, monitoring, and physical, verbal, sexual, financial, and litigation abuse. AI analysis lacks the context necessary to accurately detect these patterns, making it prone to false positives and false negatives.

When digital data such as text messages, emails, or parenting app exchanges are uploaded to a large language model like Claude or ChatGPT, the AI is only able to analyze the data with the context provided. This means that AI cannot account for gestures, tone of voice, facial expressions, historical context, or impact, which are all crucial aspects of coercive control.

For example, a text message that says ‘I’m picking up the kids at your house on Friday at 5 pm. Have them ready, please’ may seem harmless, but it could be part of a larger pattern of abuse. AI cannot see whether the message is helpful, neutral, or part of an abusive pattern.

Moreover, AI is prone to pattern hallucination or algorithmic confirmation bias, where it finds patterns even when the data is meaningless. This can lead to false positives, where normal relationship disagreements or assertive communication are labeled as ‘coercive.’

The person setting the AI search criteria may also unknowingly bias the results toward a desired conclusion, making the AI-generated analysis unreliable. This is particularly concerning in domestic violence cases, where the stakes are high and the consequences of misinterpretation can be severe.

The six pitfalls of relying on AI in coercive control cases are:

• Low Context Analysis: AI lacks the context necessary to accurately detect coercive control patterns.
• Pattern Hallucination: AI finds patterns even when the data is meaningless, leading to false positives.
• Mistaking Inputter Bias as Truth: The person setting the AI search criteria may unknowingly bias the results toward a desired conclusion.
• Missing Unusual Presentations: AI cannot accurately analyze unfamiliar material, leading to missed patterns of abuse.
• The Streetlight Effect: AI searches for patterns using only the data provided, missing entire categories of abuse.
• Data Exposure: Victim-survivors and their attorneys are uploading sensitive data into AI apps, potentially putting their safety at risk.

In conclusion, while AI may seem like a convenient and impressive tool in domestic violence cases, it is not a reliable substitute for human expertise. Coercive control is a complex and multifaceted issue that requires a nuanced understanding of the relationship dynamics and power imbalances involved.

By relying on AI-generated analyses, we risk distorting the truth and perpetuating harm. Instead, we must prioritize human-centered approaches that take into account the complexities of coercive control and the unique needs of each individual case.

The Importance of Human Expertise in Coercive Control Cases

Human experts in the field of coercive control have spent years studying the complexities of abuse and developing a deep understanding of the tactics and strategies used by abusers. They can provide a nuanced and contextualized analysis of the data, taking into account the broader relationship dynamics and power imbalances involved.

Moreover, human experts can identify patterns of abuse that AI may miss, such as unusual presentations of control or manipulation. By prioritizing human expertise, we can ensure that the truth is told and justice is served in coercive control cases.