LLMs and mental health chatbots
- Specific benefits and concerns for each AI technology
- Risk and eligibility assessments and facial recognition
- Robotics
- LLMs and mental health chatbots
We found similar perceptions as those described for other AI uses when we asked about LLMs and mental health chatbots.
Improvements in efficiency and faster access to support are the most popularly selected benefits for general-purpose LLMs and mental health chatbots, respectively.
Specific to LLMs, the public are worried about the generation of biased content (50%) or harmful content (47%)
Table 5: Most commonly selected benefits for LLMs and mental health chatbots
Which of the following, if any, are ways you think the use of this technology will be beneficial?
| Technology | Not asked in 2022/23 | Percentage |
|---|---|---|
| Large language models (e.g., ChatGPT) | 1 Not asked in 2022/23 | |
| 2 Not asked in 2022/23 | ||
| 3 Not asked in 2022/23 | ||
| Mental health chatbots | 1 Not asked in 2022/23 | |
| 2 Not asked in 2022/23 | ||
| 3 Not asked in 2022/23 | ||
| Technology | Top three chosen benefits 2024/25 | Percentage |
| Large language models (e.g., ChatGPT) | 1 Efficient by automating tasks | 56% |
| 2 Tool for learning and skills development | 50% | |
| 3 Enhance creativity | 38% | |
| Mental health chatbots | 1 Faster support | 50% |
| 2 Improve accessibility | 46% | |
| 3 Feels human, preventing isolation | 33% |
As with robotic care assistants, losing human interaction was a popular concern for mental health chatbots. 68% of the public are concerned about mental health chatbots leading to isolation by replacing human interactions. Specific to LLMs, the public are worried about the generation of biased content (50%) or harmful content (47%), as well as attrition in individual problem-solving skills (66%).
Table 6: Most commonly selected concerns for LLMs and mental health chatbots
Which of the following, if any, are concerns that you have about the use of this technology?
| Technology | Not asked in 2022/23 | Percentage |
|---|---|---|
| Large language models (e.g., ChatGPT) | 1 Not asked in 2022/23 | |
| 2 Not asked in 2022/23 | ||
| 3 Not asked in 2022/23 | ||
| Mental health chatbots | 1 Not asked in 2022/23 | |
| 2 Not asked in 2022/23 | ||
| 3 Not asked in 2022/23 | ||
| Technology | Top three chosen concerns 2024/25 | Percentage |
| Large language models (e.g., ChatGPT) | 1 Reduced problem solving and critical thinking skills | 66% |
| 2 Biased content | 50% | |
| 3 Generating harmful content | 47% | |
| Mental health chatbots | 1 Loss of human interaction | 68% |
| 2 Unclear that you are not interacting with a human | 63% | |
| 3 Providing misleading advice | 62% |
Image credit: Wes Cockx & Google DeepMind / https://betterimagesofai.org / https://creativecommons.org/li…