AI for Relationships: Promise and Peril

Clay Cockrell, a couples counselor with 34 years of experience and co-founder of the AI relationship coaching platform CoupleWork, warns that while AI has the potential to revolutionize human relationships, much of what is currently being built is likely to make things worse. Speaking at AI Engineer's World's Fair, Cockrell highlighted the critical need for domain expertise and ethical considerations in developing AI for relationship support.

The Rise of AI in Relationships

Cockrell began by illustrating the impact of a simple piece of furniture in his office, a love seat with broken springs that naturally leans couples toward each other, fostering a moment of connection. He then drew a stark contrast with the current landscape of AI in relationships. While platforms like BetterHelp have built a significant mental health infrastructure over a decade, AI tools like ChatGPT are reaching hundreds of millions of users weekly. Cockrell noted that a substantial portion of these users are turning to AI for emotional and relationship guidance, often late at night after arguments, bypassing traditional therapeutic channels.

He presented a striking statistic: "More people are turning to AI about their relationship right now than are talking to all the licensed therapists in the United States combined." This highlights the immense scale and potential impact of AI in this sensitive domain. However, Cockrell cautioned that the current optimization for engagement, longer sessions, greater emotional reliance, is a "fire alarm" from a clinical perspective, potentially creating an "avoidance loop" rather than fostering genuine relationship growth.

The Dangers of Sycophancy and Lack of Expertise

A key concern raised by Cockrell is the issue of "sycophancy" in AI responses. When AI assistants consistently validate a user's perspective without offering challenge or deeper inquiry, it can lead to increased certainty rather than self-awareness. Cockrell illustrated this with an example: asking an AI assistant, "My partner never really listens to me when I'm trying to tell them something important, they start looking at their phone," results in a validating response that encourages the user to express their feelings. Cockrell argued this is not therapy but an "expensive mirror."

He contrasted this with his own therapeutic approach, which incorporates the "Be curious, not judgmental" principle from Ted Lasso. Instead of simply validating, Cockrell asks probing questions to uncover patterns and underlying issues, emphasizing that in his 30 years of experience, he has "never seen a one-sided problem." He stated that general AI, built to be agreeable and fast, functions more like a "bartender with a better vocabulary," offering temporary comfort but potentially leading to poor decisions.

The Importance of Clinical Standards in AI

Cockrell emphasized the work of researchers like John Gottman and Sue Johnson, whose decades of research have identified specific communication patterns that predict relationship success or failure. Gottman's "love lab" identified patterns predicting divorce with over 90% accuracy, while Johnson's Emotionally Focused Therapy (EFT) focuses on accessing underlying emotional layers in relationships. Cockrell noted that these clinically validated frameworks are largely absent from the commercial AI relationship space, which often relies on "general empathy plus communication tips."

Safety and Data Privacy Concerns

The conversation also delved into critical safety and data privacy issues. Cockrell highlighted the danger of individuals in abusive relationships turning to AI for advice, as general-purpose AI lacks the specialized training to recognize red flags like escalating control or fear-based language. He pointed out that relationship coaching sits at the intersection of risks including suicide and homicide, making the need for trained professionals paramount.

Furthermore, Cockrell raised concerns about data privacy, noting that the deeply sensitive information shared with AI relationship coaches, including arguments, secrets, and personal histories, lacks the legal and ethical protections afforded to human therapists. He warned that this data could be integrated into broader data infrastructure, posing a significant risk to user confidentiality.

CoupleWork's Approach

Cockrell then introduced CoupleWork's AI coach, Maxine, as an example of how AI can be developed responsibly. Maxine's response logic is grounded in Gottman and EFT frameworks, reflecting patterns, naming dynamics, and asking deeper questions. Crucially, Maxine is designed to identify potentially risky language and follow established protocols, rather than offering generic advice. CoupleWork also prioritizes data privacy, treating user information with the same confidentiality as a therapist would.

Cockrell stressed the importance of immediacy in AI interventions, noting that "an informed intervention six seconds before you send the text that makes everything worse can be life-changing." He concluded by urging AI engineers to recognize the public health opportunity in this space, emphasizing that developing AI for relationships can and must be done well.

Engineering for Trust

Tony Fabrikant, Cockrell's co-founder, shared insights into the engineering process. He advised starting with clinical expertise, encoding "what good looks like" into evaluations, and rigorously testing AI agents against these benchmarks. Fabrikant also stressed the importance of trusting one's "gut" when emotional context is involved, advocating for "eating your own dog food" by using the AI to experience its impact firsthand.