Many Challenges in AI, But Possibilities are Endless
You're reading Entrepreneur India, an international franchise of Entrepreneur Media.
Have you tried talking to anyone about Artificial Intelligence? Those words instantly conjure up images of killer robots, Jarvis and superhero movies for most people.
Artificial Intelligence has gone through several decades of ups and downs. Investors and founders have gained and lost billions through these cycles, while tech leaders like Andrew Ng, Yann LeCun, Geoffrey Hinton and a long history of neuroscientists from Hubel and Weisel, to Sejnowski and Olshausen continued to push the boundaries on brain research, Neuromorphic Engineering, Robotics and Machine Learning, breaking new grounds, cycle after cycle.
Much of the technology industry waited for a big bang event to announce the big arrival of AI. But that day never came. Instead, AI has found a way to slowly and steadily creep into our lives without us even knowing it.
Whether it is Google’s autocomplete and predictive search, Apple’s voice recognition Siri or Facebook’s automatic friend tagging, we have come a long way without realising how dependent we have gotten on AI.
For the past 5–10 years, AI has gone a bit further, slowly and steadily finding its way into eCommerce and everyday shopping. Just like most other applications mentioned above, it arrived quietly on scale, almost ten or more years ago with Amazon in the US. It has slowly grown its web across the world. But so many different kinds of online stores have grown since then and the companies have all been competing for attention from consumers around the globe.
Unlike Amazon in the US, almost all of them have been struggling to find the Holy Grail, the key to differentiating themselves from the sea of options in the market. Big Data Analytics and Image based learning are emerging as two areas in AI that are slowly and steadily finding their way into helping these companies differentiate from each other.
Product discovery and personalisation have become clear goals in this march towards differentiation.
Solving product discovery
Online shoppers not only have to decide which site they go to, they also have to decide which brands to pick, what price to buy at, the quality of the product and much more. But what one sees when they visit a store online is typically a sea of products sourced from a variety of places. Every site sorts and tags each item they list. So the user can search to some degree to find what they want if they use the right search keywords or filters. But more often than not, most of us are surfing deals and when not just looking at deals, we want choice.
How does a site open up meaningful discovery? This is a question Amazon’s Collaborative Filtering based Recommendation system began to answer. Users were delighted to see a carousel that read ‘Users that viewed this item, also viewed this’.
The idea was simple. People rely on recommendations of other people when shopping for products. The recommendation system built, basically digitised this concept for the online shopping world.
It opened up discovery in a much more meaningful way than just looking at rows of products, each one of which had to be opened in order to see all product details, just to see if it was even useful. And it made sense, especially in categories like books.
Over time, several different kinds of recommender systems emerged. Many collect data on the user’s journey on the shopping portal and use the click stream data to understand which categories of products people like and lean towards.
These systems crunch large amounts of data about all the consumers that use
their portals to target users with their brands and products of choice on one hand, and on the other, make discovery more meaningful. They also use natural language processing (NLP) to correlate text on the portals with patterns in user behaviour. AI here, is a classic case of crunching large amounts of consumer data, to find patterns and help users find their way around a site while increasing their likelihood of seeing, clicking and buying.
Fashion is a visceral experience; a visual experience. Sometimes shopping for fashion online, can feel more like shopping at a crowded wholesale market, overwhelmed by choices and so much noise, it’s hard to tell if any of them fit your look, your style, your preferences.
Fashion is a whole different beast online. While its conversions rates hover around one to two per cent, they are also the largest margins for most companies. AI has found applications extensively in fashion over the last ten years. Despite several failures, companies around the world are actively experimenting, growing and changing their appetite for AI to fix fashion.
What visual intelligence looks like
Computer Vision or Image intelligence based AI has been around for ten years now. The technology expertise required in this space is immense, but talent and skills are scarce.
Like.com was one of the first visual intelligence based portals. It solved the problem of visual similarity. The premise: See something you like at a store window? Take a picture of that zebra printed top and instantly find it online with Visual Search. Since Like.com, many other B2C visual search applications have gone live around the globe. Cortexica, CamFind and Slyce have been experimenting with this use case for years now.
At Mad Street Den, we launched our first fashion product line Vue.ai, a range of visually powered AI products for fashion with Voonik. Visual Recommendations are the first of these to hit the market. The premise was simple — fashion is visual. Recommendation systems need to show the user dresses that are visually similar to the one they are seeing, and not just show what 10 million other people also bought.
Recommendations appear next to or at the bottom of each product in a carousel, showing outfits that have similar colours, patterns, fit, size, style and so on. As a user, you click on an outfit when you suspect you like it. AI can look at that image, go and collect all the other clothes that are similar and show them to you, right next to the product you clicked.
Artificial Intelligence, especially the kind based on visual recognition systems, can actually begin to draw your fashion persona(s). They can understand over time, the clothes you like, the colours and fits you buy, the prints you actively avoid, the sleeve length and collar style you prefer and much more.
The road ahead
While many of these applications and the potential of AI are exciting no doubt, there are many challenges ahead on this road. Visual search has not scaled and stayed a sticky use case for most consumers around the globe.
Pictures that users take of products they want to find online, often include backgrounds, cats, chairs, people nearby and more in the image. This requires consumers to crop the image to focus on the product. The steps required to find precise results can be laborious if not executed well in the app. User Experience and Design is of utmost importance and has failed in many of the examples mentioned above.
In short, we’re just scratching the surface of this absolutely exciting space of AI and eCommerce. The challenges ahead are many, but the possibilities are absolutely endless.