Recently, it came to my attention that there are still human beings behind my favourite e-commerce sites. Imagine my surprise. It stands to reason that AI could handle every aspect of e-commerce. When will it? That depends on how quickly AI advances, and it depends on one other very important factor: Will AI be able to do emotionally rich customer service jobs?
Human customer service doesn’t seem like it’s really a part of e-commerce, but it is. When it comes to online shopping hacks, e-commerce coupon site RetailMeNot recommends calling a customer service person and asking them for discounts. In other words, haggle with them.
Additionally, Although some shopping websites automatically find the lowest prices for you, it never hurts to ask a customer service rep to lower an online price by a couple of bucks. RetailMeNot points out that human intervention in the online shopping process is good business because it cultivates return customers for the retailer. It’s also great for the customer who wants to save money.
How long before chatbots replace those human customer service agents? The top e-commerce firms use chatbots to respond to customer inquiries, but the idea of haggling with a chatbot doesn’t make a lot of sense. A chatbot can tell you about discounts. A chatbot can tell you about upcoming sales. But can a chatbot override the system and manually change the price of an item?
To plenty of e-commerce firms, it doesn’t matter. Online retailers will contribute to an increase in virtual agent chatbots of about 35 percent by 2024. With a good knowledge-base, optimization for mobile, and site-wide ubiquity, chatbots can help customers at any time, from anywhere, no human strings attached. If the customer wants to haggle for the type of discounts a bot can’t fathom, they’re out of luck unless they can get a hold of a human customer service rep during regular business hours.
Dr Michael Johnston of Interactions points out that human interactions have always played a critical role in machine learning. Human-assisted AI can handle those tricky customer interactions involving high-level cognition in real-time. The human assistant must simply listen in, or in the case of chatbots, watch the text on the screen, and provide semantic interpretations. If the customer is trying to haggle for a lower price, the human assistant could evaluate whether such a negotiation is reasonable and instruct the chatbot on how to respond. Through machine learning, chatbots could learn how to haggle with customers ” in theory.
Another option is for AI to assist the human rep with a list of possible responses to a query. The rep can then edit and select the most appropriate response, which provides good feedback to the neural network. In either case, the human assisting the AI or vice versa, the human is providing information that can help a deep learning network develop something close to human intelligence, which includes emotional intelligence.
We need to start working on large-scale interaction systems that enable machines to communicate and collaborate with humans rapidly. Machines need to start learning how we conceptualise the world, says Bill Su, CEO of Humanlytics. E-commerce interactions, particularly those involving something akin to haggling, could provide that conceptual apparatus.
If a human assisting an AI chatbot to haggle with a customer eventually gets replaced by the chatbot, it would be true irony. But that is indeed the type of scenario that could play out and should play out if we want AI to actually have the cognitive capabilities of the human brain.
Su points out that we don’t provide AI networks with the thinking process and rationale that leads us to make specific decisions. Instead, we tell AI to categorise this like this, that like that, and then we say, if this is this, then do this. We train AI to react to pre-established, rote signals, based on the probability of outcomes, but we don’t necessarily train it how to ride a bike with no training wheels.
To do so, we need to teach AI how to interact with humans by steering it through conversations where there is a clear reference point ” such as the inventory of an e-commerce store ” but not a predictable outcome. A deep neural network could feasibly learn to think like a human if it’s able to receive repeated input on negotiating unpredictable conversations in real time.
We need to design AIs such that they can interact with humans to understand not only what the human wants, but why they want it. Just like an apprentice to a master craftsman, AIs need to learn on the job, Su says.
When a customer is haggling for a better price, all sorts of emotional content may come into play. My child wants this pair of roller blades for Christmas, she just loves rollerblading, and it would make me so happy to get them for her, but I can only afford to spend $10, and there is a pair of similar rollerblades on this other site for $10. If, out of this fairly complex sentence, AI can learn to understand that the customer is asking for price matching, that familial devotion is driving the request, and that granting the request will generate a heightened level of customer loyalty due to emotional attachment, well then, that’s the type of AI we want.
The question is, will we ever get it?
