Responsible AI Deployment (RAID) with NeoSkin

TLDR:

‍

The team at Beauty by Holition (BBH) has been pioneering emerging technology solutions since 2015. From virtual try-on solutions for makeup and AI-powered shade matching (part of Neo Beauty Studio), to accurate AI skin diagnostics by NeoSkin, BBH has consistently delivered state-of-the-art digital experiences.

BBH's patented FaceSDK (our proprietary software AR solution library) harnesses machine learning (ML) and artificial intelligence (AI) to analyse thousands of data points on the human face to deliver superior visualisations and accurate skin analytics. Not only are we the pioneers of AI and AR beauty solutions, our technology has also proven to increase purchase intent x5, increase product interaction time to 2+ minutes on average, and increase conversions by 24%. The success of our virtual beauty solutions can be attributed to the following factors: 

1. ‍Leading with human-centred design‍
2. Responsible AI deployment
3. Industry-leading technology, promising unparalleled AI accuracy & visual reproduction.

‍Find out more about the role human-centred design plays in the development of our solutions here.

‍

Understanding Responsible AI

Responsible AI (RAID), as a practice, targets to increase AI transparency whilst also reducing AI bias, and it is guided by 6 core principles, as set out by the International Organisation of Standardisation (ISO) : Fairness, transparency, non-maleficence, responsibility, privacy & inclusiveness.

Fairness, Non-Maleficence, & Responsibility

At Beauty by Holition, the team follows the following three-step system to align with the RAID principles:

‍

‍

1. Creating Balanced Data Sets: We curate our own datasets to ensure balanced outcomes  when training our AI algorithms. For example, in order for NeoSkin to detect uneven skin tone accurately, the image files used for training had to contain varied degrees of post inflammatory hyperpigmentation, melasma, dark spots & sun spots on a full representation of skin types, tones, genders, & ages.

‍

2. Peer Review: Upon creating balanced data sets of various skin concerns, types, & ages, each category has to undergo several rounds of peer review. The aim of this step is to preempt bias which may have occurred during labelling of each data set.

‍

3. Iterative Bias Testing: The final step of this process is running the AI-enabled face-scan to reveal any underlying bias & to identify how the data sets & AI training can be fine tuned further.

‍

Transparency, Inclusiveness, & Privacy

Furthermore, to address transparency, inclusiveness, & privacy in particular, the team at BBH takes on the following considerations;

‍

4. Explainable AI: Ensuring that the user journey clearly signposts how & when AI is being used. Users are made aware of what exactly the journey entails at the onboarding page, and exactly how our AI informs the product & routine recommendations. During the Questionnaire stage of the journey, the user can self-select their concerns, which are taken into consideration when the AI outputs recommendations from the photo analysis. The AI makes 2 recommendations; first, is what the AI detected to be the highest priority concerns, and second, is helping users prioritise between the 2 self-selected concerns - Find out more here.

‍

5. User Research & Testing: To provide the utmost intuitive experience & to test that the AI is transparent & explainable, NeoSkin underwent multiple rounds of user testing including concept and usability testing. 

‍

6. Privacy: The team at BBH really prides itself In developing an AI solution which respects & prioritises user privacy. Typically, AI processing is cloud hosted or outsourced to third parties which have few obligations to the user, making it incredibly easy to store personal data. NeoSkin operates completely locally, within the user's device, meaning that personal data is never transmitted, shared, or stored with anyone at any time. 

‍

‍

Ready to put NeoSkin's responsible AI to the test? 

Reach out to our team today to get started with a 60-day free trial.

‍

‍

By Edit Kruk

21.05.2024