Good Practices for Making Your App Trustworthy, Effective and Transparent for End-Users
This page describes the recommended steps to make your app feel and work transparent and effective for end-users
In the previous step, you successfully finished the app flow. Your application can now collect high-quality images, process them with AI image analysis, and even recommend relevant products.
Before launching live, we recommend reviewing this optional guide towards improving the perceived quality of your app.
Indicate That Your App Uses AI
In the modern world, mixing sources and leveraging AI to do multiple jobs, previously possible only to a limited circle of trained specialists, it is important to explicitly communicate about AI usage to manage end-user expectations correctly.
Advice #1: Disclaimer about "not a medical-grade device"
If your app does not involve humans in end-user communication about analysis results or recommendations, it is highly recommended that you add a disclaimer stating that the app is "not a medical-grade device." By stating this, you help your end-users better understand the degree to which your app can be trusted in making any decisions affecting their health.
We recommend stating in a disclaimer that:
The application is not intended to be a substitute for professional medical advice, diagnosis, or treatment.
None of the provided skin health parameters estimations can be treated as health advice.
Advise users to consult with a physician or other qualified healthcare provider about any medical condition or treatment before undertaking any healthcare regimen.
Advice #2: Disclose the AI limitations
Any algorithm, AI, or data processing method is prone to error to some degree.
We recommend adding a visible note to the screens with the AI image analysis results presentation, saying, "Our AI estimates skin health, but it might make mistakes."
Advice #3: Indicate if collected data will be used for AI improvement
According to the general Data Processing Agreement with Haut.AI, Haut.AI acts as a data processor, and our clients remain the data controllers. Data collected in the SaaS platform is not used for AI training without written approval from clients (you).
If you plan to use the collected data for any other AI model training or are using some other service for personal data processing with AI, we highly recommend disclosing the amount and type of end-user personal data used for AI model training.
If you do not plan to use collected images, you can configure your Dataset to delete images as soon as they are processed. See API for Datasets
Increase End-User Trust in AI Results by Highlighting Detected Concerns with Masks on Their Images
Masks are additional results of image analysis, representing the spatial information about detected skin/hair health returned by the selected set of Algorithms. Masks can help end-users to better understand the scale of the problem, and help you build trust into your application.
Check the description of the Algorithm Results for more details.
There are two options for how to present images with overlaid Masks to your end users:
Download Images with Overlaid Masks from SaaS API
You can download Images with the Overlaid Masks via the API. These Images are raster objects and can not be modified. Follow these steps:
Use the rasterization service: Access the rasterization service using the provided Swagger link. This service allows you to obtain images with applied masks.
Invoke the correct method: Call the appropriate method in the rasterization service to request the image with the masks applied. The API will return an image that includes the specified masks.
Wait for processing: Before retrieving the images, ensure that the image has been processed by the algorithms. This step is crucial to guarantee that the masks are correctly applied to the image.
Retrieve results: The rasterization method can also return results directly alongside the masked image. Be sure to check the response for any additional data you may need for analysis or display.
Draw (Interactive) Masks in Your App Frontend
You can draw Masks on top of Images directly on the front of your app. Follow one of these guides to start with:
Follow this guide to see how to draw Masks for web applications: Visualize Masks on Frontend [JavaScript]
Follow this guide to see how to draw Masks with server-side image rendering: Visualize Masks on Server [Python]
Follow this guide to see how to draw Masks in mobile apps (Incoming Soon)
Your Masks visualization may leverage interactions with masks, such as zooming by clicking to exact points/polygons of the mask, etc.
Communicate AI Confidence and Quality Issues to End-Users
Haut.AI provides a Quality Algorithm, included in the Face Metrics 2.0 Application, and designed to estimate uploaded image quality.
Your application can leverage the responses of this algorithm to manage end-users' expectations and correct end-users' behavior.
Ideas:
Example: Pores algorithm requires high intensity of illumination to correctly detect and clasidy all visible pores on a face. This algorithm results can be shown Smoothed for the recent 3-4 photos in order to account for variability.
When the Quality Algorithm marks an image as "low quality" in its result, your app can show a warning or prompt for a better image.
Your app can use the Metric of Quality Algorithm (Quality Score) as an approximate "AI confidence" in predicted AI analysis results.
Offer Feedback Mechanism in Your App
Your end-users may feel more engaged if they can indicate what they like or dislike in your app. Collecting this feedback can help you iterate on product improvements faster.
Ideas:
Allow users to rate the recommendations and provide feedback about AI image analysis results and/or product recommendations.
Offer the possibility for users to opt for human expert review in case they are uncertain about the AI’s analysis or recommendations.
Stay Tuned to the Latest Recommendations from Haut.AI
Haut.AI is committed to delivering you the best quality product and service, and we regularly update our guides and recommendations by sharing the best practices among our clients.
Stay tuned!
Last updated