Artificial intelligence (AI) has a lot of promise for learning. However, most applications so far are using AI to replace human interaction while there are less examples of AI increasing the productivity of interaction. This article presents two retail cases that approach interaction from a different angle.
Written by Teemu Ylikoski
In general, the ways in which AI impacts work fall into two categories (Acemoglu & Restrepo 2018). These are the displacement and productivity effects, which can coexist. AI and robotization replace human work in certain tasks. This creates a displacement effect, reducing the need for human labor. On the other hand, AI also reduces the cost of automated, simpler tasks, increasing the demand for labor in more complex tasks. False hopes, measurement issues and the general approach in which AI is currently being fostered may be the reason behind the productivity paradox with AI: while transformative new technologies are present, their effect on productivity remains to be seen (Brynjolfsson et al. 2018).
The traditional way to think of AI, robots and machine learning, is through a human metaphor. It is natural to think of the learning machine as parallel to humans, to the extent of dystopia (Dick 1968). However, this type of thinking may have limited the way in which we think of AI in a systemic perspective. It may also be the reason why so many AI applications seek productivity; while less are approaching the problem truly outside the box.
Similarly, in learning the focus so far has largely been on productivity and efficiency, e.g. with MOOCs, teacher-less self-study programs, and current education apps. We are only beginning to see more applications of AI in supporting human to human learning (Lamson 2018).
Case: A robotized grocery store with a focus on human service
Honestbee is an Asian ecommerce grocery retailer that also operates an AI-based physical grocery store in Singapore. The store, Habitat, is a completely new take on buying groceries. It has no traditional aisles or checkout counters. The customer makes checks in to the store with a mobile app, makes purchases with the app, scanning items as they are added to the cart, and checks out of the store with the app. A robot packs the customer’s purchases and delivers them in bags next to the exit.
Automation is nothing new in retail. Where Habitat differs is that it is far from automated self-service. Dozens of human assistants called “Bees” roam the store floor, ready to help the customer in finding items, operating interfaces, and making purchase decisions. Should the customer have technical issues or perhaps a need for shopping ideas, the next Bee is always right there to help.
Although a great deal of automation is present in the store, it is used to augment human interaction, not replace it. Habitat uses automation and artificial intelligence to replace phases in the shopping path that add little value to the customer, such as queuing at the check-out or packing items. Instead, these phases are automated, freeing the customer to spend time meaningfully and use other services in the grocery store, including multiple cafeterias and a library.
The purpose is twofold. First, making the store a nice place to spend time increases the likelihood that the customer’s average purchase increases. Secondly, Habitat is also a stepping stone to online shopping, which is where Honestbee makes its turnover. Shopping inside the store is done with the same mobile app that customers use to order online. Having practiced mobile shopping inside the store with the help of a friendly Bee lowers the adoption threshold.
Case: Using AI to optimize the time for service
Singtel is one of the largest mobile operators in the world, with over 600 million customers in Asia. Its market share has been dwindling and competition increases. Customer retention is important in a mature market, which makes every customer contact point important. Singtel’s flagship store in Singapore has been designed from the ground up, using AI to increase customer experience and to improve customer retention.
Customers come to a telecom operator for service issues and for new services or products. Typically, a customer needs to queue for 30-45 minutes before a customer representative is available. Waiting is a customer pain point with little value. But this does not have to be so. The Singtel store uses AI to make an experience that is more like a cafeteria than an telco store. Anonymized facial recognition systems track the moods of customers entering vs. leaving the premises, customer movements within the shop are used to optimize the store and make recommendations and make the need for service desks unnecessary.
The key to the experience is a free wifi that replaces the queue. Instead of taking a ticket when entering the shop, customers sign in to the wifi which assigns them a queue number automatically. In a country where cellular data is expensive, the wifi provides value in itself. In signing in, customers tell what services they are interested in. The system uses customer data and tracks where the customer spends time in the store. It then makes suggestions on additional services and items the customer may find interesting.
When the customer’s turn comes, the system feeds all of this data to the customer representative, along with the location of the customer in the store. The queue number is projected on the floor close to the customer. As a result, customer experience and satisfaction are on the rise. Even more importantly, the typical customer ends up buying more and making more expensive purchases.
In the two cases, AI is not treated as a cost saver or a replacement of human contact. In fact, in both cases, it is driving costs up. However, in the end, what matters is that it results in improved overall results. I believe this represents a focus on the productivity of AI.
In some respects, AI-based learning systems appear to have focused on the displacement effects, such as automating current learning processes. There areless examples of focusing AI primarily on the outcomes of learning.
Probably the most typical way AI is currently used in learning is to replace the human component (think apps and MOOCs). Students study materials online at their own pace, watch videos, take quizzes and take part in peer to peer assessments through the app. However, human interaction is often minimal. Although education exists for the purpose of increasing knowledge, digital learning tends to emphasize the cost efficiency of learning. The marginal cost of another MOOC participant approaches zero which makes it tempting to think of AI as a way to create the same results with lower cost.
What if we approached this from another direction? What would a learning environment look like where the focus would be on the value for the learner, asopposed to efficiency? What would an AI-assisted, but human interaction-centered learning environment look like? The process of studying has multiple phases, not all of which create value for the learner. What if we created a learning environment where automation would assist and empower a more meaningful interaction with the teacher, rather than replace it?
Acemoglu, D., & Restrepo, P. 2018. Artificial Intelligence, Automation and Work (No. w24196). National Bureau of Economic Research.
Brynjolfsson, E., Rock, D., & Syverson, C. 2018. Artificial intelligence and the modern productivity paradox: A clash of expectations and statistics. In The Economics of Artificial Intelligence: An Agenda. University of Chicago Press.
Dick, Philip K. 1968. Do Androids Dream of Electric Sheep? New York: Ballantine Books.
Lamson, Melissa 2018, The Role of AI in Learning and Development. Inc Magazine.