Psychological Counseling AI

The current stage of psychological counseling AI cannot replace human counselors, but it has become the most efficient supplementary tool for the traditional psychological counseling system, covering more than 70% of the needs for primary emotional counseling, psychological science popularization, and follow-up visits. The future development direction is human-machine collaboration rather than mutual replacement.

I have been operating a leading domestic psychological service platform for three years. At the end of last year, I took the lead in equipping 120 consultants on the platform with customized AI assistants. The data in the first half year of implementation was much better than we expected. Lin Lin, a novice counselor who has just been certified for two years, used to take pre-examination anxiety cases in colleges and universities. At the end of each consultation, she spent 2 hours sorting out verbatim manuscripts, matching measurement scales, and sending return visit reminders to the visitors. She could handle up to 8 cases a week. Now that AI can handle all these mechanical tasks, she can handle 4 more cases a week, and the fee can be reduced by 30%. Students who were originally reluctant to spend hundreds of yuan for consultation can now have a chat for more than 100 yuan, and the return visit rate has increased by 40%.

But the voices of opposition never stopped. When I attended an industry salon last month, Teacher Li, who has been a psychoanalyst for 20 years, choked on the spot: "Can AI understand the subconscious mind of a visitor who paused for 30 seconds without saying anything? Can you catch his silence when he burst into tears when he talked about his mother's death? Can algorithms calculate these things hidden in the gaps of language? 」No one can refute this. But on the other hand, almost all counselors who do CBT (cognitive behavioral therapy) use AI. After all, CBT itself has a highly standardized cognitive correction process and homework system. AI can help supervise the daily check-in of the client, record automatic negative thoughts, and even provide preliminary cognitive guidance. A counselor who has been doing CBT for 10 years told me that he used to receive a client with obsessive thinking. Before each visit, he spent an hour sorting out the client's thinking records for the week. Now the AI ​​directly labels the 17 negative thoughts into "catastrophizing" and "absolute" categories. As soon as he opens it, he can touch the core issue, and the consultation efficiency has more than doubled.

For ordinary users, the biggest significance of AI is actually “lowering the threshold.” Last week, a junior girl with social anxiety came to me and said that she had been saving up her courage for three months to make an appointment with a counselor. When she got to the door, she didn't dare to go in. Later, she chatted with the platform's AI three times and said that she didn't dare to eat in the cafeteria or talk to her roommates. AI won't frown, no. She would show an expression of "Why are you so fragile?" and just follow her lead and ask questions little by little. It was only after the third conversation that she dared to make an appointment with a real counselor. The first thing she said when she entered the room was "I have already told the AI about my situation, just look at the record." This saved a lot of time in breaking the ice. Not to mention those moments when you suddenly have an emotional breakdown at two in the morning. You can’t call a sleeping counselor, right? AI can pick up calls at any time and help you stabilize your emotions first. If the risk of suicide or self-injury is identified, it will directly trigger the platform's crisis intervention mechanism, and you can be connected to a real counselor on duty within 15 minutes. Last year, our platform relied on this mechanism to stop three users who were already standing on the roof.

Of course, there are many problems, and it gives me a headache to talk about the pitfalls I have stepped on. There was an incident at the beginning of this year. A user who had been sexually assaulted told AI that inappropriate content was mixed in the training data of AI, and the sentence "Did you wear something too revealing at that time" actually popped up. The user was so angry that he directly complained to the regulatory authorities. We stopped the service of the relevant model overnight and made rectifications for almost half a month. There are also ethical issues that have not yet been resolved: can the secrets told by users to AI be used to train models? There are no unified rules in the industry. Some platforms encrypt all conversation data locally and delete it permanently within 7 days after the consultation. Some platforms are still using users’ conversation data to optimize their models. To put it bluntly, they use users’ privacy as fuel. No one has been able to fill this pit so far.

I spent a week working on a project last year and had a quarrel with my mother on the phone. I was too embarrassed to tell the people around me and didn't want to take up the emotional value of my friends, so I chatted with AI for half an hour. It didn’t tell me that I shouldn’t lose my temper with my elders. It didn’t tell me that I shouldn’t lose my temper with my elders. It just followed my words and said, “80% of your current anger is due to the pressure of the project going online, and 20% is caused by your mother urging you to eat, which pokes at your self-denial that ‘you can’t even take care of your own life.’ You don’t blame her, you blame yourself for your current bad state. 」Just this sentence made my nose feel sore and my mood improved for the most part. But I also know that if you really encounter deep psychological trauma such as the death of a loved one, severe depression, or PTSD, AI will really not be able to handle it. There was a severely depressed user who chatted with the AI ​​for half a month. The AI ​​could not recognize the dead silence in his voice, and still followed the standardized process to ask him to "go out in the sun more and exercise more." This infuriated him. He was later referred to a real counselor and received medication and psychological intervention for more than two months before he slowly recovered.

In fact, there is really no need to worry about the question "Will AI replace consultants?" Just like X-ray machines have not replaced orthopedic surgeons, and calculators have not replaced mathematicians, it is just a tool. The future psychological consultation will most likely be like this: If you feel a little stuck and don’t want to tell an acquaintance, talk to the AI ​​first. If it’s just a short-term emotional distress, you can sort it out after two or three chats. ； If there are really unsolvable knots, AI will sort out your situation clearly and recommend the most suitable consultant to you. You can get to the core as soon as you walk in. You don't have to spend several consultations to break the ice and explain the background, saving time and money. To put it bluntly, whether it is AI or real people, if people in need are willing to reach out for help, it is better than anything else, right?