I, Travelbot
Artificial Intelligence has come to the railway in the form of travel assistance in London, writes STEFANIE FOSTER
Using artificial intelligence to provide travel assistance.
“Ipropose to consider the question ‘can machines think’?”
If you ask a computer and a human the same question, can a computer answer in a way that is indistinguishable from a human response?
British mathematician Alan Turing asked this question back in the 1950s. He wanted to test a machine’s ability to exhibit intelligent behaviour by asking a human evaluator to judge natural language conversations between a human and a machine designed to generate human-like responses through text on a screen.
Nearly 70 years later, modern computer scientists are performing their own versions of Turing’s experiments to improve the way companies provide information to their customers.
A new generation of artificial intelligence, in the form of AI chatbots, has emerged. Essentially, they are programmes designed to conduct conversations with humans through an online messaging interface.
Growing numbers of people are already using applications such as Facebook Messenger and WhatsApp in preference to the traditional text message to communicate with their friends (globally, 60 billion messages are sent every day via these messaging platforms - three times the number of SMS text messages).
Already people are moving from looking on websites for the information they want to using apps on their mobile phones. So, some companies are starting to look at how they can use the apps that are already popular with their customers, instead of forcing them to download yet another one.
In June, Transport for London joined this new revolution in AI, with the launch of TravelBot. Designed as a digital travel assistant, the new service is accessed through Facebook’s Messenger App and uses artificial intelligence to provide live bus arrival information, Tube and rail service updates and maps (see panel, page 56, for full details).
To find out why TfL has looked to this new kind of social media to communicate with its passengers, RAIL asked TravelBot Product Manager Charul Gupta to explain the thinking behind the launch.
It feels like you’re talking to a human, so you don’t have to understand technology or download different apps. Charul Gupta, Digital Product Manager, TfL
“If you go by the numbers, research indicates that more than half of the world’s population is now online, and more than 50% of those online are active social media users,” Gupta explains.
“We already had around 400,000 people interacting with our Facebook page, so we know that we have customers on social media.
“The idea was to provide customers with information via their channel of choice. Obviously, if people are there, that’s what they prefer to use, so why not provide information where they want to see it?”
TfL was already providing information to passengers through its Facebook presence, but it was all via customer service agents responding to individual queries. The functionality was not there to automate any response, so any and every question, no matter how small or simple, had to be answered by a human. And that takes a lot of time, both for the staff and for the passenger desperate to know when they’re going to be able to get home. So, what does TravelBot do to change that? “It’s basically replicating human communication,” says Gupta. “TravelBot is based on natural language. It feels like you’re talking to a human, so you don’t have to understand technology or download different apps to get information.
“We’re trying to provide all the information you want, using everyday language. So, someone might get up in the morning before work and want to know what the service status is [on the Tube network]. If I was talking to a friend, I might just ask, ‘do you know what the service status looks like?’ and that’s exactly how you communicate with the ‘Bot’. You write it as you would say it to anyone else and the ‘Bot’ will reply with an update on the lines.”
Impressively, it actually does work with everyday language (see RAIL’s test, right).
But what about questions that it has not been programmed to answer yet?
“If the ‘Bot’ can’t answer the question, there are two different responses it might give. One is where someone might have asked a question in a tense the ‘Bot’ didn’t understand or language it couldn’t decipher. Then it would reply saying: ‘I don’t understand, can you rephrase or provide more detail’? The other situation is where you ask a question the ‘Bot’ understands but we haven’t built the functionality to answer. So, it will reply to say: ‘I’m new and I’m still learning. We are coming up with new features soon’.”
Gupta explains that the important thing is to understand what the intention of the customer was, given the question they asked, to ensure that TravelBot can respond appropriately - even if it does not know the answer to the specific question.
Presumably, TravelBot is ‘learning’ new answers all the time?
“Artificial intelligence robots like this normally have the capacity to learn by themselves. But because we have just started with this and are using a very simplified version at the moment, we are helping the ‘Bot’ to learn. But as we add more functionality and get more understanding about the technology behind it, we want to be at a stage where the ‘Bot’ learns by itself. So, it has the capability to do that but we are currently helping it along.”
By logging all the questions TravelBot receives and categorising them into types of question, TfL can monitor what it is that passengers expect from the new service and what functionality should be prioritised for the next update.
Explains Gupta: “The features we have rolled out so far are based on knowledge of the information customers want from
us. On buses it can help you with the route status, it can give you live arrival information. For the Tube, we currently only have line status information, but we are working on journey planning functionality.”
Just because TravelBot is not a journey planner in itself, TfL does not want the system to appear completely unhelpful when asked a question relating to planning a journey, so the TravelBot will direct people to the correct link on TfL’s website to answer the question.
“From the feedback so far, it looks like journey planning is one thing that customers expect, so it is something we’re looking into. We want to provide customers with what they want.”
And do they want TravelBot? What has the take-up been like so far?
“It was launched on June 12. In the three weeks since then, using only organic promotion [such as word of mouth on Facebook], we have 4,500 unique users and we have had 67,000 messages. This was in line with what we were expecting because we wanted it to grow slowly.”
It is important to Gupta that TfL can learn from the feedback received by users and improve the service before it becomes really popular. They have already fixed some issues since launch to ensure TravelBot provides the service that is expected from it.
The service is still very much in a soft launch stage with feedback being analysed, improvements being sought and next steps planned. It is being promoted through social media but, says TfL, further promotion is still being discussed to give the service time to grow at its own pace. Baby steps.
Says Gupta: “These are really early days for the TravelBot and the main benefits at the moment are really on the bus side of things, but we’re taking feedback as we go along so that we can add functionality.”
For example, many bus stops in London do not provide a countdown to the next bus and a text service is offered to find out when the next bus is. This is the next-generation style of that service. Waiting at a bus stop, sending a friend a Facebook message, you can just drop TravelBot a quick message asking when your bus will arrive and carry on with your personal conversation.
“We really want people to interact with the service in their preferred setting,” says Gupta. “People spend a lot of time using messaging apps such as Twitter, Facebook, WhatsApp and, obviously, people want to have information in the easiest way they can get it. So, if they’re already using Facebook, the benefit to the customer is that they don’t have to switch apps to check the information. Everything’s in one place. A lot of people would say that this is where this type of communication is heading.”
Surely there’s a knock-on benefit for TfL too?
“It frees up more time for our customer service agents to focus on more complex queries, which in turn helps us improve our response rate. Because if someone is calling customer service agents to find out when their next bus is due, it takes up their time. And this is something that we can easily automate with the functionality we have in the artificial intelligence.” ■
■ To find out more about TravelBot, visit https://tfl.gov.uk/travel-information/social-mediaand-email-updates/facebook-travelbot