Saturday, 22 February 2025
28.5 C
Singapore
33.4 C
Thailand
21.1 C
Indonesia
26.4 C
Philippines

An AI-powered health chatbot struggles with accuracy

Explore the challenges and developments of WHO's new AI health assistant, SARAH, which provides health information but may not always be accurate.

The World Health Organisation (WHO) has embarked on an innovative project using artificial intelligence to disseminate basic health information via a human-like avatar. This venture introduces SARAH, a virtual health assistant designed to offer 24/7 support in eight languages on topics such as mental health and nutrition. However, the WHO cautions on its website that this early prototype, launched on April 2, might not always deliver precise answers.

Challenges with current AI technology

SARAH, or Smart AI Resource Assistant for Health, cannot diagnose, similar to platforms like WebMD or Google. Instead, it directs users to WHO resources or advises consulting with healthcare providers. Despite its empathetic design, SARAH’s responses can sometimes be misleading due to its outdated training data and the inherent limitations of its AI model, which include generating incorrect or “hallucinated” answers.

Ramin Javan, a radiologist and researcher at George Washington University, remarked, “It lacks depth, but I think it’s because they just don’t want to overstep their boundaries, and this is just the first step.” WHO asserts that SARAH is meant to complement the work of human health professionals rather than replace them.

AI’s potential and pitfalls

Image credit: AI Business

Trained on OpenAI’s ChatGPT 3.5 with data up to September 2021, SARAH does not have information on more recent medical developments or advisories. For example, it incorrectly stated that the Alzheimer’s drug Lecanemab was still in clinical trials, although it was approved in January 2023 for early disease treatment. Even WHO’s latest statistics can confound SARAH, requiring multiple prompts to fetch the correct information.

In its efforts to simulate empathy, SARAH analyses users’ facial expressions for 30 seconds before deleting this data, ensuring anonymity. WHO’s communications director, Jaimie Guerra, highlights the privacy measures in place to secure these interactions, stating that all sessions are unlinked from any identifiable data to prevent cybersecurity risks.

Future directions and ethical considerations

The development of SARAH builds on WHO’s previous virtual health worker, Florence, and involves the same New Zealand-based company, Soul Machines, for the creation of the avatars. While the current avatar appears white, there is potential for users to select from a variety of avatars in future iterations to enhance personalisation.

WHO recently issued ethical guidelines for the use of AI in health care, emphasising data transparency and user safety. As AI continues to evolve, the integration of these technologies in public health remains a delicate balance between innovation and accuracy.

Hot this week

Hitachi Vantara: Building AI success without falling into financial traps

Discover how Hitachi Vantara guides Southeast Asia firms to maximise AI's ROI through strategic planning, scalable infrastructure, and targeted use cases.

Google updates Family Link to help kids stay focused

Google updates Family Link with improved screen time controls, school time features, and contact supervision, helping parents manage digital habits.

Researchers train AI to understand animal emotions

Researchers are training AI to recognise animal emotions, helping farmers, vets, and pet owners detect pain and distress in animals more accurately.

Nvidia acknowledges RTX 5090 and 5070 Ti manufacturing defect

Nvidia confirms a rare manufacturing defect in the RTX 5090 and 5070 Ti, affecting less than 0.5% of GPUs. Affected users can request a replacement.

ASUS launches ZenScreen Duo OLED MQ149CD, a portable monitor with dual OLED displays

ASUS unveils the ZenScreen Duo OLED MQ149CD, a portable dual-screen monitor with OLED technology, delivering stunning visuals and flexible work setups.

Nvidia acknowledges RTX 5090 and 5070 Ti manufacturing defect

Nvidia confirms a rare manufacturing defect in the RTX 5090 and 5070 Ti, affecting less than 0.5% of GPUs. Affected users can request a replacement.

DJI’s RS 4 Mini stabiliser now features advanced subject tracking

DJI’s RS 4 Mini stabiliser introduces subject tracking, improved battery life, and better handling, making it an excellent tool for content creators.

American Airlines introduces AirTag location sharing for lost luggage

American Airlines now supports Apple’s AirTag location sharing, making it easier for passengers to track and recover lost luggage.

Google may launch YouTube Premium Lite in more countries

Google may launch YouTube Premium Lite in the US, Australia, Germany, and Thailand, offering a cheaper plan with fewer ads. Pricing is yet to be confirmed.

Related Articles