Expert: Fine-tuned model AI may become a 'mental patient' for doctors to study

Visual China Picture

ChatGPT When "wisdom emerges", will emotions emerge?

A study from Germany in April 2023 showed that GPT3.5 obtained higher scores in a commonly used anxiety disorder test questionnaire in humans, and after being prompted to enter anxiety situations, as the researchers expected , resulting in specific decision-making judgments and biased behaviors. Does this mean that AI may be trained to develop anxiety disorders?

On May 25, at the first "AI Asks the Brain" dialogue held by Tianqiao Brain Science Institute (TCCI for short), Qiu Jianying, director of the Psychological Counseling and Treatment Center of Shanghai Mental Health Center, and Artificial Intelligence of Beijing University of Posts and Telecommunications Chen Guang, associate professor of the college, communicated and interacted online on this topic, attracting nearly 70,000 people to watch the live broadcast.

As a senior psychiatrist and psychotherapist, Qiu Jianyin is very excited to see that AI’s performance in emotional cognition has reached a new level this time. She said frankly that she is looking forward to AI being able to suffer from anxiety disorders after being professionally trained, thereby helping doctors better understand the manifestations of anxiety disorders, analyze the relevant factors behind them, and open up new ideas for finding more and better prevention and cure methods. .

Chen Guang believes that this type of model learns and generates text based on a large amount of training data. Currently, models trained through fill-in-the-blank and solitaire methods lack true subjective awareness and have no emotions or will. Based on artificial feedback reinforcement learning (RLHF), AI can indeed continuously adjust based on human judgment and guidance on its output results, gradually approaching human tendencies. Although today's imitation can be used for text learning and generation, he believes that it cannot form a stable and consistent "personality."

Qiu Jianying introduced that pathological anxiety refers to a persistent feeling of nervousness and uneasiness for no specific reason, or a premonition of disaster, threat or impending disaster without any realistic basis, accompanied by obvious autonomic nervous system dysfunction and mobility impairment. Restlessness is often accompanied by subjective pain or impairment of social functioning. According to statistics from the National Health Commission in 2020, the incidence rate of anxiety disorders in China is about 5%.

According to the dialogue, doctors have been actively applying artificial intelligence technology to the diagnosis and treatment of mental disorders in recent years, including anxiety disorders. At present, AI has played an active supporting role in suicide prevention and other fields. As the scope of human-computer interaction expands from text to facial expressions, body postures, speech, heart rate, body temperature, and even future brain-computer interfaces, AI will further help reveal the physiological mechanisms behind human emotions and become psychotherapy robots in the community. Or support mental health diagnosis and treatment in remote areas with limited medical resources.

Chen Guang observed from the perspective of computer science that the ChatGPT craze has promoted the widespread application of AI in all walks of life. If the large model is "fine-tuned" combined with professional data in the field of mental health, on the one hand it will help AI better simulate humans. Emotional reactions can simulate a group of "mental patients" for doctors to study. On the other hand, it is also expected to identify human emotional disorders more accurately and efficiently, and play an auxiliary diagnosis and treatment role. The scale of relevant data is key to whether AI technology can penetrate deeply into the field of mental health.

Qiu Jianying and Chen Guang agreed that the identification of mental disorders in traditional medicine is based on subjective observation. The rise of computational psychiatry has promoted the application of tools such as data analysis, machine learning, and artificial intelligence in the classification diagnosis, risk warning, and prognosis prediction of mental disorders, and assisted in the formulation of clinical diagnosis, treatment, and prevention strategies. Large-model AI represented by ChatGPT will bring greater breakthroughs to computational psychiatry and help humans further open the two "black boxes" of mental illness and the brain.

"AI Asks the Brain" is a newly launched light academic exchange conference series, focusing on cross-talk between AI scientists and brain scientists, exploring the two-way innovation of AI technology and brain science, and focusing on AI autonomous consciousness and human consciousness. Conduct open discussions on cloud and other hot public topics.

The above is the detailed content of Expert: Fine-tuned model AI may become a 'mental patient' for doctors to study. For more information, please follow other related articles on the PHP Chinese website!