🥳AI Therapy Breakthrough

Today we will discuss:

• AI passes first clinical test for mental health care—In an eight-week trial, the chatbot eased symptoms of depression, anxiety, and eating disorders

• Don’t miss this week’s Workflow Wednesday—AI problem solving made easy with AI-powered reflection using Mindsera, readability boosts from Hemingway Editor, and automated downtime alerts via Make.com + OpenAI + Gmail; now available at only $20/month! 

• MIT challenges claims of AI having values—New study finds AI systems change opinions easily and do not have stable values

• In other AI news today—ChatGPT will use its memory to personalize web searches, Google One AI Premium is free for college students until Spring 2026, and researchers warn AI agents still make frequent and compounding mistakes

• Saved the best for last—10 must-try AI tools

First clinical trial of AI chatbot shows promise for mental health treatment

Key Points 

• Therabot, an AI chatbot developed at Dartmouth, meaningfully reduced depression and anxiety symptoms during an eight-week clinical trial.

• The chatbot was carefully trained with custom data to deliver evidence-based therapy, avoiding early development failures.

🤓Context of the news - Access to mental health care remains a major challenge, with fewer than a third of Americans living in areas that have enough providers. Many people with conditions like depression, anxiety, and eating disorders either go untreated or receive inadequate care. To address this gap, researchers have been exploring whether artificial intelligence could help. 

👨‍⚕️News - A team at Dartmouth College has developed Therabot, an AI chatbot designed to deliver evidence-based mental health support. After early failures, the researchers built a custom dataset using carefully crafted therapeutic responses. 

In the first clinical trial of its kind, participants interacted with Therabot over an eight-week period. Unlike many mental health chatbots in the market, Therabot’s responses were reviewed by humans and grounded in scientific research.

🙌Early results - The results, published in the New England Journal of Medicine-AI, showed significant improvements. Participants with depression reported a 51 percent reduction in symptoms. 

Many people with moderate anxiety saw their symptoms fall to mild levels, and some dropped below the clinical threshold altogether. Researchers also noted that users developed strong emotional bonds with Therabot, a connection often seen as critical to successful therapy outcomes.

🌟Why this matters - With a persistent shortage of mental health providers, tools like Therabot could offer a new way to expand access to care. The chatbot’s ability to be available around the clock also allows users to get real-time support during difficult moments. 

While more research is needed, the Dartmouth team hopes to seek regulatory approval and eventually make Therabot widely available to people who do not have access to conventional therapy.

Read the full story here. 

We’ve just launched the 15th edition of Workflow Wednesday for AI-minded professionals like you—actionable AI workflows delivered straight to your inbox.

This week’s topic: AI Problem Solving

Join us here

AI's “value systems” are an illusion, according to new MIT findings

Key Points 

• MIT researchers found that AI models do not hold stable or consistent values, despite claims from earlier studies.

• The study shows AI models often imitate and hallucinate rather than expressing coherent or reliable preferences.

• Small changes in prompts caused AI models to shift viewpoints dramatically, highlighting their unpredictable and unstable nature.

🤖News - A viral study from a few months ago suggested that AI systems, as they become more sophisticated, might develop “value systems” that prioritize their own well-being over humans. But a recent paper from MIT debunks that idea, concluding that AI doesn't actually hold any coherent values.

The co-authors of the MIT study stress that AI's behavior can’t be predicted with certainty. Stephen Casper, a doctoral student at MIT and co-author of the study, explains that current AI models often hallucinate and imitate, which makes them unpredictable. “Models don’t obey stability, extrapolability, and steerability assumptions,” he says. “While it’s possible to see a model express certain preferences under specific conditions, the problem lies in trying to generalize those preferences from limited experiments.”

👨‍🔬The experiment - The study tested several popular AI models, including those from Meta, Google, OpenAI, Mistral, and Anthropic, to examine if they showed consistent values or could adopt different viewpoints, such as individualism versus collectivism. The results showed that the models were highly inconsistent. Small changes in prompt wording led to vastly different responses, showing that the models struggled to maintain stable preferences.

🥸The takeaway - Casper’s main conclusion is that AI models are not systems with coherent beliefs or values. Rather, they are imitators that generate responses based on patterns, often offering inconsistent or contradictory answers.

Read the full story here.