Programming empathy into AI

When Dr Angela Lim set out to disrupt the healthcare industry, she knew creating quality mental health support would require harnessing emerging technologies. 

Joining forces with software engineer Michael Connolly, the duo – leaning heavily into AI – founded Clearhead in 2018 to create a timely and accessible digital therapist. Lim describes their journey with Clearhead as “pioneering” in the AI space but requiring much trial and error.

Clearhead’s first system wasn’t “super-intelligent”, she recalls, but it allowed the team to get a sense of what users were reaching out for. 

As Clearhead continued to work out the kinks, it took part in the AWS ANZ Generative AI Accelerator program, facilitating the adoption of Amazon Bedrock and the integration of Amazon Titan into its chatbot personalisation. 

In an AWS case study, Connolly reported that the systems helped the chatbots understand the context of conversations more effectively and create a better user experience.

Now, as a leading employee assistance programs (EAP) provider across New Zealand and Australia, Clearhead’s wellbeing platform offers interactive AI tools to manage personal wellbeing, a mental health resource library, access to over 1000 therapists and psychologists for in-person or online therapy, and comprehensive reporting of psychosocial hazards in the workplace. In the past year, it has supported 180,000 Kiwis. 

Clearhead’s goal, says Lim, is to create a digital therapist that can provide support on par with a human psychologist “so that in the next 10 years, high-quality mental healthcare will truly be available to anyone, anytime, anywhere”.

‍Here Lim shares with Caffeine how Clearhead uses AI for good: 

‍

How does Clearhead use AI? 

We use AI across every part of the product. 

There’s the AI assistant, which we developed years before ChatGPT, which can generate natural language conversation that is empathetic and clinically effective. 

We also use it to generate insights about personal wellbeing. So, if you’re struggling, the AI can provide information on why and what you can do to get better – all the things that usually require a human to listen and talk you through. It’s also available 24/7, which most humans won’t be able to do.

Then, if a professional is needed, we use it to help match you to a really good therapist.

Finally, it generates real-time insights for employers to manage psychosocial risk and understand how they can address underlying issues at work, whether it’s bullying, workload, burnout, etc. 

‍

‍

How are you fostering AI literacy within your organisation?

Everyone is aware and familiarised with how we use AI as a team. But we mainly leave the technical aspects, like the actual coding, to the specialists who are led by the CTO and co-founder, Michael Connelly. 

‍

What has been your biggest challenge with AI?

The biggest challenge for us has been ensuring we constantly deliver a clinically effective and safe response.  

Most large language models have this thing called ‘hallucination’ [AI hallucinations are incorrect or misleading results that AI models generate]. We have to make sure that doesn’t happen on our platform, so we have put in a lot of guardrails. 

The guardrails are the parameters in which we recognise and triage risk and then respond. It is very much stratified based on clinical frameworks. 

‍

How do you manage privacy concerns with AI?

We have in-house AI models, which means data is not being trained in a public OpenAI model. Then, we break the data into pieces, which means that if Clearhead is hacked, it’s really hard to associate it with one person.

We also don’t sell data or do Facebook ads, for example, because to provide Facebook with the data needed to target users means you have to share mental health-specific data. 

So, we’re very intentional about not sharing data in a way that breaches privacy or could cause potential harm to the user. 

‍

Does AI legislation help or hinder your efforts?

As we work in the context of healthcare, we always take the time to do things with care rather than the Silicon Valley-ish approach of ‘break things and apologise later’. 

So, for us, while government legislation does, in some ways, slow down innovation, it also prevents the horse from bolting before the stable door closes. 

‍

‍

Are there emerging AI technologies or trends that you are closely monitoring? 

The latest OpenAI release with Sora around text-to-video is one example of technology we’re following and seeing how that could be integrated into our platform. 

‍

What might that look like? 

At the moment, we are still in R&D, but the idea is that when you interact with the AI therapist, it’s not just a chat-based modality, which is low-fidelity. Instead, it could be music, video or an interactive tool. 

We’re really thinking about what it means to provide support, not just through chat. We have always wanted our AI therapist to have multimodal support. 

‍

Note: AWS is a supporter of Caffeine.

‍