Yun Liu, PhD, is a senior staff research scientist at Google Health/Research, focused on developing and validating machine learning models for medical imaging, including pathology, dermatology, and ophthalmology.

Yun Liu, PhD

Patient Awareness of Dermatologic Conditions Using AI, With Rory Sayres, PhD, and Yun Liu, PhD

, Rory Sayres, MD, and Yun Liu, PhD, of Google Research, went into greater depth about the implications of their recent 

 study, focusing on how artificial intelligence (AI)-generated information is interpreted by consumers and what improvements may be needed to better support clinical decision-making.

A key theme in this video segment was the gap noted by the 2 speakers between improved condition recognition and limited gains in understanding appropriate next steps. While AI tools can meaningfully increase a user’s ability to identify a likely skin condition, Sayres expressed, translating this recognition into actionable decisions, such as whether to seek care, remains more complex. Such a disconnect, he noted, is consistent with prior data observed on use of online health information, suggesting influencing behavior is inherently more difficult than improving knowledge alone.

“There are definitely a lot of diseases [in ophthalmology], but when an ophthalmologist is looking at like an image of a retina, that number of possible diseases is somewhat smaller,” Sayres explained. “Whereas when you look at dermatology, one of the first things we saw is the set of conditions you might want to assess for numbers at least in the hundreds.”

The investigators emphasized that accuracy of the underlying AI predictions plays a critical role in shaping outcomes. The study’s “Wizard of Oz” arm, which presented dermatologist-generated differentials, demonstrated stronger improvements in next-step decision-making compared with the AI arm. This finding highlights that even modest gaps in prediction quality can influence how users interpret and act on information. Liu noted that improving the precision and ranking of suggested conditions may therefore be an important step toward enhancing the clinical usefulness of these tools.

Outside of prediction accuracy, the 2 speakers pointed to the value of how information is presented to users. In their study design, they included structured condition “cards” with images, descriptions, and treatment information, but their results indicate the providing of more data is not sufficient. Instead, Sayres indicated future iterations of AI tools may require guiding users through decision-making processes, potentially by providing them with more apparent recommendations or contextual cues regarding the level of severity and urgency.

In their discussion, Liu and Sayres also touched on user behavior and variability in how individuals act on their health information. Participants brought differing levels of prior knowledge, interpretation strategies, and confidence, all of which can impact patient outcomes regardless of the use of AI tools. As a result, the 2 investigators describe optimizing user experience and interface design as just as necessary as algorithmic accuracy improvements.

Looking ahead, Sayres and Liu noted the promise held by AI tools as adjuncts to, not replacements for, clinical care. Although current data support their role in improving consumer understanding of skin conditions, further refinement was described necessary to ensure these technologies can reliably guide patients toward appropriate next steps.

The quotes contained in this summary were edited for the purposes of clarity.

Disclosures: Sayres and Liu reported stock ownership from Google outside the submitted work.

GBD 2017 Disease and Injury Incidence and Prevalence Collaborators. Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet. 2018 Nov 10;392(10159):1789-1858. doi: 10.1016/S0140-6736(18)32279-7. Epub 2018 Nov 8. Erratum in: Lancet. 2019 Jun 22;393(10190):e44. 

Sayres R, Jain A, Liu Y, et al. Consumer Understanding of Skin Concerns With an AI-Powered Informational Tool. JAMA Dermatol. 2026 Apr 15:e260597. 

. Epub ahead of print. PMID: 41984449; PMCID: PMC13084546.

Videos

Liu and Sayres emphasize that while AI tools improve consumer recognition of skin conditions, gaps remain in prediction accuracy and information.

AI Tool Use Linked with Understanding of Skincare Issues, With Rory Sayres, PhD, and Yun Liu, PhD

 access have been identified in the field of dermatology, with only 28% of skin conditions reported by patients being assessed through a dermatologist.

, a pair of investigators named Rory Sayres, MD, and Yun Liu, PhD, discussed the key findings authored in their recent 

 study regarding ways in which artificial intelligence (AI) tools may be impacting consumers’ understanding of skin health. While the study found certain gaps remain, the AI applications assessed were linked with increased accuracy and confidence in terms of consumer understanding.

“I think one of the key goals we're trying to understand with this study is how use of these at home tools affects decision making and affects the quality of information people get,” Sayres explained.

In this segment of their interview, Sayres spoke on how his team’s primary aim was to evaluate how at-home, AI-powered applications can impact both decision-making and the quality of health information attained by consumers. With many already turning to online platforms for questions on their own dermatologic concerns, Sayres, Liu, and colleagues had set out to better understand how these tools compare with traditional information-seeking methods, as well as how users interpret and act on the data given to them.

The team specifically looked at the ways in which assessments by lay individuals served them, with retrospective dermatology cases and images with limited clinical context being used. These images would thus mirror real-world scenarios in which individuals might implement smartphone-based tools to assess their various skin health concerns.

The investigators focused on 2 key outcomes: participants’ ability to correctly identify a skin condition and their understanding of appropriate steps for the conditions’ management. Compared with a control arm using standard resources such as web searches, those in the study who utilized AI-supported tools were significantly more likely to both attempt and correctly name the condition depicted.

Investigators also included a “Wizard of Oz” group, the participants of which were shown dermatologist-generated differential diagnoses presented in the same format as the AI tool. This allowed for isolation of the impact of prediction accuracy on user performance. Notably, improvements in recognition of different skin conditions were seen in both the AI and dermatologist-guided arms, with the highest accuracy seen when participants were provided with expert-level predictions.

However, the results revealed a more nuanced picture when it came to clinical decision-making. While participants improved in their identification of conditions, this did not consistently translate into better judgment regarding next steps, such as whether to seek medical care or manage the issue independently.

Only the dermatologist-guided arm demonstrated a significant improvement in next-step accuracy. This finding, Sayres suggested, indicates the necessity of higher-quality or more precise predictions to impact patient decision-making in a meaningful way. Together, their data highlight both the promise and current limitations of AI in dermatology.’

For any additional information on these findings, view the video segment posted above.

This interview highlights new data assessing whether artificial intelligence (AI)–powered tools can help consumers better understand dermatology cases.

Yun Liu, PhD

Articles