Contextual AI: what it is and why it matters in fashion retail
Katharina Vandamme-Eybesfeld
Scalable solutions for luxury fashion and VIP comms.
I recently came across an interesting article talking about the building blocks of a successful relationship between humans and AI. We have come a long way with AI and whilst it is making many aspects of our lives easier and more efficient, we - as consumers - don't fully trust it yet. Oliver Brdiczka, the author, notes that Artificial intelligence (AI) is powering more and more services and devices that we use on a daily basis, such as personal voice assistants like Alexa, film recommendation services like Netflix and driving assistance systems like ACC. And while AI has become a lot more sophisticated, we all have those moments where we wonder: “Why did I get this weird recommendation?” or “Why did the assistant do this?”
One of the reasons for this distrust is that most current AI systems operate as a ‘black box’ with limited interaction capabilities, human context understanding and explanations.
So, what steps can we take to increase humans' trust in machine-driven applications? One suggested answer can be found in an emerging approach called "contextual AI".
Contextual AI is technology that is embedded, understands human context and is capable of interacting with humans.
At this stage it is important to understand that this should be viewed as a concept, rather than a new algorithm or machine learning method. This concept is based on a human-centric view and approach to AI. Its foundation is defined by 4 pillars that enable a more collusive relationship between humans and AI.
I-Chun Han (Lead Data Scientist) and I (Founder) feel it's time to shine some light on these pillars and examine why they are particularly relevant in the context of fashion retail and our internal AI approaches at BECOCO.
- Intelligibility: Trust & transparency.
No black box. I want to know what you're doing and why. For example: if you recommend and outfit to me, tell me why it is not just a straight copy of an influencer and why it will look particularly flattering on me [me!] versus someone else. I-Chun: "Compared to (in fashion) frequently used black box approaches, we are using machine learning models like SVMs and decision trees that are more transparent for humans. This allows us to have a proper understanding of what is going inside and through it, trust it more."
2. Adaptation: Work in several contexts.
I expect that your technology works in different environment. For example: computer vision based clothing recommendations have to be adapted to numerous clients, who use different fashion photography and technical environments. I-Chun: "Deconstructing the various style elements (such as 3D colour spaces, contextual styles or garment features) within fashion allows us to get to the ultimate backbone of it. This backbone can be transferred to different clients and then be adopted by training it with their unique features."
3. Customisation: Control.
I'm ruling you - not the other way around. For example: you recommend that my best shade of red is strawberry-red, but I'm also really into brick-red right now, so take that into consideration and build it into my suggestions. I-Chun: "We analyse each component of fashion and transfer it to our machine learning models as a specific input. That way, our models can be trained transparently and can be personalised to specific user profiles giving users full flexibility and control."
4. Context awareness: All encompassing.
I want you to see everything that I see. For example: I have various functional and also emotional needs when it comes to a digital styling solution and I want you to address them all. Chun: "Contextual AI can be seen as a more advanced version of transfer learning, which is a new and popular deep learning method. However contextual AI is more focused on the human behaviour which requires a lot of domain knowledge within fashion retail and all the related human interactions within. This is why our approaches focus strongly on collecting as many elements as possible within this space, which will enable us to prepare a fruitful environment for contextual AI.
Conclusion: At BECOCO we believe that AI should serve the purpose of homogenising data, link the right products to consumers through personalised recommendations that can scale across channels and geographies, and with it deliver an overall better shopping experience. Therefore, it is important that we leverage the power of Contextual AI to help move the industry forward and harness its potential to continually innovate.
Katharina & I-Chun