Lori Gottlieb sees a role for AI between therapy visits and a limit at the relationship itself
Lori Gottlieb and Sword Health’s Virgílio Bento discussed when AI could help people seek support and what clinicians bring to therapy.
LAS VEGAS — Psychotherapist Lori Gottlieb said Monday at HLTH that patients can use artificial intelligence to rehearse difficult conversations or reflect between appointments. She said the more important question is what people can do afterward in their relationships, without turning to a chatbot for each decision.
Gottlieb, the author of Maybe You Should Talk to Someone, shared the Revv Stage with Sword Health founder and CEO Virgílio Bento in a discussion about AI and the future of psychotherapy. The two took turns questioning one another. Bento focused on whether technology could make support available sooner, while Gottlieb described what she looks for when patients already use it.
From rehearsal to a conversation
Gottlieb told the audience about a client whose husband consulted an AI tool after disagreements to try to understand his wife’s perspective. The apparent result was that she felt less understood when he brought the machine’s interpretation back to her. Gottlieb suggested that they try using it together so each could speak for themselves and the exercise might improve the way they talked the next time.
She described a therapist’s role as helping the person see parts of a situation they might have missed, an approach she called “wise compassion.” A chatbot designed to agree readily with a user can reinforce that user’s account of a disagreement. In her example, the value of the tool depended on how the couple used it to learn something about each other.
Gottlieb said good tools should help people “learn these skills” and apply them in real conversations. She made the comparison to psychotherapy, where the aim can be for a patient to use skills independently. In the same way, a chatbot interaction may be more useful if it changes what a person can do after the exchange ends.
Reaching people sooner
Bento said a childhood experience watching his brother struggle to get intensive rehabilitation led him to work on access. He framed Sword’s approach as “AI in combination with clinicians,” describing technology as a way to reach patients sooner while retaining a clinical role. The company describes Sword Mind as a service that combines AI with licensed psychologists.
On mental health specifically, Bento said the system often “waited for a person to crash” before offering help. He suggested that some users might disclose problems to a chatbot earlier because they feel less judged. The speakers did not present data comparing how much sooner they would seek a clinician or whether the chatbot would reliably direct someone to appropriate care.
Gottlieb said some people might find it easier to tell a machine something they have never told anyone. She compared it with the immediacy of letters people write to her as an advice columnist, where someone may describe in a first paragraph what could take weeks to bring up in therapy. She suggested that an early disclosure could serve as a bridge to speaking with another person.
Gottlieb described one place the two uses can intersect. A person already in treatment may face a difficult situation between sessions and use technology to think through what to raise at the next visit. That person also has a clinician who can hear the context, ask questions the tool did not ask and notice patterns across visits. A similar exchange for someone with no clinician in the loop may end with the chatbot’s response.
The distinction matters for where the technology is offered. A tool that people use on their own may be available as soon as they type a question. In a program tied to a clinical service, the organization can also decide how to bring a clinician into an exchange, how much information that clinician can see and which uses are outside the service’s remit. Neither of the speakers presented results comparing those pathways onstage.
Gottlieb also addressed why an easy first disclosure and a therapeutic relationship can look different. A person speaking to another person has to contend with how that disclosure might be received. That can delay what a patient says, she noted, and it can also make the response more meaningful when a clinician hears and responds to it. Bento’s question about perceived judgment helped identify what an always-available tool might change at the beginning of care.
Her advice-column example illustrated another distinction. She described hearing from a woman who discovered that a partner used AI to write intimate messages. Gottlieb said the partner could be trying to express genuine feelings and still need the skill of expressing them directly. She used that situation to connect the panel’s questions about dependency to a familiar one from clinical practice: how to give someone a way to deal with an interaction when a therapist or tool is absent.
The psychological association’s guidance treats patients’ AI use as something clinicians can ask about. That discussion can reveal whether someone is using a tool to organize thoughts, drafting questions to discuss in therapy or relying on outputs for decisions beyond a chatbot’s intended purpose. It is also one way to learn what happens after a patient closes the app, which the design of an individual conversation may not capture.
Drawing the line between support and treatment
The American Psychological Association reported in June that more than three-quarters of surveyed psychologists said patients were discussing AI in therapy. The survey reflects what clinicians reported encountering, rather than a population estimate of chatbot use. It gives context to why the situations Gottlieb described are already coming into treatment rooms.
The American Psychiatric Association has warned about potential harms from use of unsupervised chatbots for therapy or medication advice. Its concerns highlight the difference between a person asking a tool to help find words for a difficult conversation and relying on the same tool for decisions ordinarily made with a clinician.
Gottlieb said developers should work with clinicians, and she described the relationship a patient forms with a therapist as central to the work. For health systems considering the kinds of services Bento discussed, the design question is who is responsible when a conversation goes past routine support. A tool’s ability to answer at any hour says little about who sees a clinically important disclosure and how that person follows up.
The panel offered ways technology could widen the points at which someone can seek help. Its clearest examples placed the benefit outside the chat itself, whether that meant bringing a concern to a clinician sooner or handling a difficult conversation with another person more effectively.