Robots, Chatbots, and Conversational AI


Recently I visited the Innorobo show in Paris, a gathering of robot companies from around the world, which brought together industrial robots, service robots, toy robots, family robots, and many other robot types, all under one roof. One of the key highlights was SoftBank Robotics launching Pepper Partners Europe, inviting developers and companies in Europe to build applications for SoftBank’s flagship robot, Pepper. As a part of the launch, SoftBank showcased solutions from 23 existing partners, spanning verticals including healthcare, retail, banking, education, eldercare, and many more. This initiative is part of SoftBank’s expansion of Pepper outside Japan, where more than 3,000 Pepper robots have already been deployed at over 1,000 companies. SoftBank Robotics, largely composed of the French robotics pioneer Aldebaran (of which Softbank owns 95%), has more than 500 employees globally, with the majority based in Paris at the Aldebaran facility.


SoftBank has big ambitions for Pepper, and with more than 20,000 Pepper robots deployed worldwide in both consumer and enterprise environments, SoftBank is easily the leading player in this space. Pepper is a powerful robot, with multiple sensors, lasers, cameras, microphones, and a tablet that acts as the main control interface. Pepper also comes with integrated speech recognition and image recognition algorithms, which allow the robot to do facial recognition, take voice commands, and have conversations with humans.

Among the use cases, there are a few common themes:

  1. Pepper’s basic use case is for meet and greet purposes at retail stores, banks, hospitals, train stations, and so forth to provide basic information to customers or visitors. It can also perform check-ins for appointments and guide customers to the appropriate locations. According to SoftBank, on average there is a 3x increase in customer traffic in stores due to the presence of Pepper (although this includes the novelty factor, which will naturally diminish over time).
  2. Pepper can also go further and perform the role of a retail floor assistant, who can answer questions and even provide product recommendations based on the knowledge it has gathered, or data stored in a customer relationship management (CRM) system. Pepper can also take orders and complete purchase transactions, after which it can conduct a customer satisfaction survey.
  3. Pepper is also good at entertaining, which includes dancing, singing, storytelling, or playing games. This is a common use case in elderly care homes in Japan where Pepper has become a popular companion and entertainer.
  4. Pepper also has some smart capabilities around gathering customer information like age, gender, and mood, all of which can help in A/B testing of a promotion within a store, similar to a mobile app or website.

Pepper relies largely on conversational algorithms or artificial intelligence (AI) in the background to perform its tasks. Conversational AI is a hot area these days, with chatbots and messaging platforms beginning to utilize virtual AI assistants for retail or customer service. Most adoption of conversational AI is on mobile, with messaging platforms like Facebook Messenger, WhatsApp, WeChat, or Google’s upcoming Allo being touted as the next battleground for e-commerce – an evolution of website-based commerce, also being called conversational commerce.

Pepper is hoping to latch onto the trend of conversational commerce, but specifically within physical brick and mortar locations, rather than virtually on a mobile device or PC. There is one problem, though – most of the conversational commerce we see gaining traction is that of text-based chatbots, rather than voice-based. Yes, Amazon Alexa and Google Home could shake things up in voice-based conversational commerce, however text-based conversational commerce is easier to implement, and most consumers today prefer text-based interactions rather than voice interactions. Successful voice-based conversational commerce needs to have an AI that is able to understand a wide range of accents, perform in a range of noisy environments, and have a voice that sounds more human than robot. These are harder problems to solve, and for humans to trust an AI in order to make purchase decision, voice-based conversational AI has a long way to go.

Pepper’s voice interactions today are far from perfect, and the more customers become accustomed to having interactions with an AI that is message-based, the harder it will get for Pepper to compete in a retail environment. For example, a customer who walks into a store might be greeted by Pepper at the door, but is more likely to end up chatting with the store’s chatbot on WhatsApp or Messenger. The point is that AI chatbots seem more human (look at Microsoft’s Xiaolice, used by 40 million users in China) than a humanoid robot like Pepper. Also, Pepper costs €19,900 plus service fees, which makes it a much harder return on investment (ROI) case to make to a retail store. Even if Pepper is used to complement chatbots, the difference in the conversational experience might put off customers, and could end up relegating Pepper to simply being a doll or an entertainment bot, rather than being something useful.

Having spoken with a number of developers and companies at the Pepper Partners Europe event, the limitation around voice conversation was the top issue cited. Today, one needs to hard code sentences and conversations into Pepper, rather than using a smart algorithm to figure out human-like responses. Although SoftBank told me that they are working on these issues and expect progress in these areas, it will be hard to compete with the voice recognition and natural language processing (NLP) algorithms from Google, Amazon, Apple, or more so with the AI messaging chatbots that seem to be finding more traction in the market, attracting a wider audience, and as a result getting better much faster than any other conversational AI medium.

Comments are closed.