Can machines save the human soul? Here comes an AI chatbot that can "talk" to you!

The significance of artificial intelligence (AI) chatbots goes far beyond the superficial "you ask me and I answer", there are some more important things, such as "allowing everyone to enjoy the same medical resources."

To increase access to mental health services, a research team from University College London and Eberhard Karlsruhe University of Stübingen has developed a personalised self-service chatbot - Limbic Access .

An observational study involving 129,400 people in England showed that the use of the system led to more people, particularly from minority groups, using National Health Service (NHS) talking therapy services , and an increase in the number and diversity of groups served.

Figure | Limbic Access chat interface on the NHS Talking Therapies website. a) Initial message interface, which is customized for a specific service; b) Interface for collecting demographic information. (Source: The paper)

The related research paper, titled "Closing the accessibility gap to mental health treatment with a personalized self-referral chatbot", has just been published in the scientific journal Nature Medicine.

This research confirms the potential of digital tools to improve access to mental health services , providing strong support for the promotion of more comprehensive and equitable services.

What has changed?

Mental health is a widespread problem worldwide, but not everyone can use mental health services when they need them, and access to mental health services is restricted by many barriers .

Limbic Access aims to address this problem by making services more accessible.

In this observational study, use of the system led to a larger and more diverse group of people entering talking therapy services on the National Health Service (NHS), particularly some historically neglected groups.

According to the paper, the uniqueness of Limbic Access lies in its personalized design, which reduces the social stigma of mental health issues by actively guiding users through the self-referral process while allowing users to explore their mental health issues in depth .

The chatbot has been put into use on the online portals of 14 NHS talking therapy services in England. During the three-month study period, these centres saw a 15% increase in self-referrals, compared with just a 6% increase in the 14 centres that did not use the tool.

Figure | Total number of recommendations before and after the implementation of Limbic Access. Use of chatbots is shown in pink, and non-use is shown in gray. (Source: The paper)

The increase in self-referrals was even more significant among non-binary, Asian/Asian British and African/African British groups, increasing by 179%, 39% and 40% respectively.

Figure | Percentage changes in socio-demographic groups using chatbots (pink) and control groups (grey). (Source: the paper)

Notably, the increase in the number of referrals did not result in an increase in waiting times or a decrease in the number of clinical assessments.

How is it done?

So the question is, how does Limbic Access do it?

It is reported that the working principle of Limbic Access includes collecting personal information and actively guiding users to complete the self-referral process and obtain mental health services.

The chatbot predicts the patient’s most likely diagnosis by using multiple machine learning models and administers customized questions based on this prediction .

The model uses free text input, standardized questionnaire scores (such as PHQ-9), demographic information, and behavioral indicators (such as reaction time and typing speed) as inputs for its predictions. The prediction model combines the Transformer model for natural language processing for free text input and the gradient boosting model for other input modalities.

Furthermore, three self-referral formats were used in the study to ensure equal length, approximately 120 questions in total, but with differences in question selection and presentation.

Figure | Research design and treatment pathways for NHS talk therapy services. a) Research design of a multi-site retrospective observational study, showing the 3-month pre-implementation and 3-month post-implementation phases of different services; b) Treatment pathway, showing the process from patients visiting the NHS talk therapy service website and self-referring to entering treatment. (Source: The paper)

Limbic Access is personalized based on the information provided by the participant, whereas generic chatbots and web forms do not change based on the participant's answers.

One of the key findings from the study was that Limbic Access significantly increased self-referral rates, particularly among minority groups.

The introduction of Limbic Access not only addresses problems with existing mental health services, but also has a positive impact on the entire field by improving the accessibility and inclusiveness of services.

By encouraging users to think deeply about mental health issues, the system is expected to reduce the social stigma of mental health issues and encourage more people to seek help proactively .

However, despite the positive findings in promoting mental health services, the study has some limitations .

First , the observational design of the study limits the accurate judgment of causal relationships because the intervention variables cannot be manipulated; second , the study only covers observational data of 129,400 people in England, so caution is needed when generalizing the results to people from other cultural and geographical backgrounds; in addition , the study failed to fully consider the individual's digital literacy and technological adaptability, which may have an impact on the experience of using self-service referral tools.

Furthermore , the study focused on observing the positive impact of self-help referral tools on minority groups, but did not delve into the social inequalities and potential biases that may exist therein.

The research team said future research could investigate more deeply the differences in usage and benefits between different groups to ensure that these innovative tools do not exacerbate social inequalities.

Finally , while the study looked at the increase in the number of self-help referrals, it did not delve into the quality and effectiveness of the services provided. Further follow-up of participants’ mental health and treatment outcomes would help to more fully assess the long-term impact of this innovative approach.

Therefore, people need to carefully weigh these limitations when applying these results to actual clinical practice.

In the future, as technology continues to develop, AI chatbots are expected to become more intelligent and provide more personalized and targeted mental health services. We can expect further expansion in the field of personalized AI chatbots.

Reference Links:

https://www.nature.com/articles/s41591-023-02766-x