AI for psychology: why mental health needs its own model
In a talk hosted by Felicis (a Slingshot seed investor), co-founder and CEO Daniel Cahn lays out why mental health can't be scaled with humans alone, how a model purpose-built for psychology differs from general-purpose AI, and how Ash is trained, kept private, and brought to market.
• The supply–demand gap: demand is escalating (about 1 in 5 adults reported severe depression last year; roughly 2 in 3 teens a depressive episode) while therapist supply stays flat — ratios reaching \~10,000:1 — so it can't be solved with humans alone.
• Why prior mental-health tech fell short: most embedded one psychologist's single approach (usually CBT) top-down, but fit matters both culturally and individually — so Ash instead learns at scale from a wide range of therapists and modalities, an "AI-first," not "one-opinion," approach.
• General AI isn't enough: assistants like ChatGPT optimize to validate and solve problems fast, yet short-term helpful can be long-term harmful; a good helper optimizes for the long-term trajectory and will push back — which is why the model's persona, not just its data, matters.
• Data via "give-to-get" partnerships: Ash partners with behavioral-health organizations (and has looked closely at OpenAI and Anthropic, who don't want this data) to build what it says is the largest dataset of its kind — privacy-first and opt-in.
• Business model: roughly half of mental-health spend is cash-pay (BetterHelp is a \~$1B business at \~$90 a session), and Ash is going direct-to-consumer at Netflix/Spotify-like prices (\~$10–20 a month).
