Researchers at the University of Utah posted a framework, in advance of publication, to assess how conversational AI and large language models could automate therapy work. The project was led by Zac Imel with collaborators including Vivek Srikumar and Brent Kious. The team sets out four categories of automation from low to high.
Category A covers scripted systems that follow prewritten content and decision trees. Category B describes AI that reviews sessions and gives feedback or ratings. Category C covers AI that assists therapists with suggestions during care, and Category D refers to autonomous agents that provide therapy directly, possibly with supervision.
The researchers compare potential usefulness and risks and warn that simple tools differ greatly in risk from fully autonomous systems. They are partnering with a statewide crisis text line to develop tools that evaluate counselors' sessions and give timely feedback. The paper recommends beginning with lower-risk tools while researchers study benefits and harms.
Difficult words
- framework — A structure for organizing ideas or plans.
- automate — To make a process work without human help.
- evaluate — To judge or measure the quality of something.
- feedback — Information about performance to improve work.
- decision tree — A step-by-step choice process for decisions.decision trees
- autonomous — Able to operate without human control.
- supervision — The act of watching or guiding work.
Tip: hover, focus or tap highlighted words in the article to see quick definitions while you read or listen.
Discussion questions
- Do you think tools that give counselors feedback would be helpful? Why or why not?
- What risks worry you most about autonomous agents providing therapy? Explain briefly.
- Would you prefer a human therapist or an AI assistant for some support? Give one reason for your choice.
Related articles
When Basic Needs Fail: How Local Shocks Become Wider Emergencies
From close experience in Turkey, the author shows how shortages in food, health and services turn into medical and social crises. Simple support like school meals, cash and cross‑city cooperation can reduce risks and costs.
