When a friend says they just bought a new TV, what questions occur to you? Other than the question of how they got the nerve to buy a TV when they still owe you money, which attributes of the television come to mind most readily for you?
Maybe the size, which everyone knows is measured diagonally in inches.
Maybe the resolution, which fewer people know used to denote the height in pixels, as in 1920 x 1080 for 1080p, but is now a rounding-up of the width in pixels, as in 3840 x 2160 for 4K.
Maybe you’re a geek, maybe you’re not. Today, everyone knows you want the biggest 4K TV your budget and living room allow for.
This is an example of background knowledge: the broad, intuitive and context-aware intelligence about familiar topics that people develop from everyday experience. There’s a rough consensus that, if you had to pick two attributes, size and resolution are the way people describe a TV.
In contrast, expert knowledge is gained through deep mastery of a narrow domain. You would not expect a person you meet on the street to be able to converse with you on the details of the subject you know the most about. In your domain you feel uniquely competent, at least in your immediate environment, to be the one to make the judgement calls, determine relevancy, and behave consistently. Your expertise is built on endless cycles of success and failure, which gradually builds a knowledge base, and rules emerge out of that.
Automating and Scaling Human Expertise
If you wanted to open an ecommerce store selling a single, well-known consumer product, you would probably have a pretty good idea of how to structure the product page and which content to include.
However, if you wanted to sell a broad range of products, from the best sellers to the long-tail, in a variety of categories, your background knowledge would quickly be exhausted. Your expert knowledge, too, would only ensure perfect product pages for a very limited catalog.
The goal, therefore, is to automate background knowledge and scale expert knowledge. Computers have traditionally excelled at scale and automation when operating on a rule-based system. The challenge here, however, is that background and expert knowledge are not strictly rule-based. Automating and scaling this type of intelligence requires a flexible system, one that can adapt to a changing marketplace.
A rule-based system will never be flexible enough to elegantly handle even trivial examples of shifting markets. For instance, in the product titles for mobile phones, the word “Touch Screen” was virtually absent before the iPhone, then became ubiquitous as manufactures and consumers quickly adopted the feature, and then disappeared again as it ceased to be a differentiating feature.
Whether a feature doesn’t yet exist, or it is not promoted because its inclusion is simply assumed, the situation looks identical to a rule-based system trying to determine whether “Touch Screen” should be part of a product title. “Touch Screen” will follow a similar trajectory in the marketing of laptop computers. We’re currently in the phase where the feature is marketed prominently in product titles. People will intuitively feel when it’s time to drop the feature from the titles, and they’re flexible enough to handle these situations on a case-by-case basis. This flexibility is one of the hallmarks of human intelligence, but no person could hope to gain the necessary background and expert knowledge to follow these shifts for thousands or millions of products in disparate categories.
Applying Artificial Intelligence to Product Pages
The solution is a system that is trained today and learns as it goes, getting better over time while keeping up with the ever-changing landscape. This type of system is an A.I. or artificial intelligence, a combination of Big Data, Natural Language Processing and Generation (NLP+G), and Machine Learning. Deep Learning and Neural Networks are related concepts, referring to the multiple layers, backpropagation, and feedback loops which are loosely modeled on the neocortex of the human brain.
These systems combine the best attributes of classical computers and human experts. It’s the only way to effectively and efficiently pick up the long-tail, continuously adapt to changing markets and consumer preferences, and stay ahead of copy-cat competitors. The blend of consistency, which is critical for things like search and category filters, and flexibility, important for dealing with novel product features and evolving marketplaces, is what sets machine learning systems apart from legacy databases and earlier rule-based systems.
Nevertheless, as good as these systems are, people are still required for their creativity, empathy, and intuition. The shift going on right now in every industry is a hand-over of the tasks that computers and AIs excel at, while people get to spend more of their time adding meaning, humor, and connection to their work. Whether AIs will ever grasp a concept like love is an open question in philosophy and computer science. What is certain is that, today, the very best work is being done by people and AIs working off of each other’s strengths. We’re simply better together.
The late Steve Jobs made this point very well:
“I think one of the things that really separates us from the high primates is that we’re tool builders. I read a study that measured the efficiency of locomotion for various species on the planet. . . . Humans came in with a rather unimpressive showing, about a third of the way down the list, but somebody at Scientific American had the insight to test the efficiency of locomotion for a man on a bicycle. . . . A human on a bicycle blew the condor away, completely off the top of the charts. And that’s what a computer is to me . . . it’s the most remarkable tool that we’ve ever come up with; it’s the equivalent of a bicycle for our minds.”
The chart below visually makes the point we’ve been building to. Combining the strengths of AI and people enables you to reach points that were previously inaccessible. People’s efforts roughly follow the law of diminishing marginal returns – putting in twice the effort does not result in twice the quality after the initial steep learning curve. AI systems initially fare poorly against their human counterparts, but the quality generally scales well with additional effort. The best possible place to be at the moment is at the sum of AI and people-powered efforts, the green point on the chart.
In this series, we’ll focus on examples of real-world use of machine learning for product information. We’ll explore how automating and scaling product catalog management positively impacts ecommerce businesses. Stay tuned for deep dives into categorization, attribute prioritization, and generation of unique SEO-optimized titles.
Artificial Intelligence and Machine Learning are the keys to enabling effortless commerce. When you see it working, it feels like magic.
See for yourself with our AI-powered Batch Manager for product information. You can play around with your own trial account today for free. Let us know how much time, money, and effort your business can save. And ask yourself, if we had perfect product pages, effortlessly, what else would we sell?