Ioannis Papaioannou was one of the main developers of Alana when it initially appeared as an entrant in the 2017-2018 Amazon Alexa Prize Competition. He’s been a driving force in defining the technical direction of Alana and continues to encourage the team to push new boundaries with machine learning and AI innovation. This interview tells a little more of the history of Alana, as well as revealing Ioannis’ impressive appreciation of fictional AI.
What is your day-to-day role at Alana?
I tend to wear two hats – researcher and CTO.
As CTO, I make technology decisions on behalf of the company. This involves working closely with our clients to find out more about their needs. I then take this knowledge and work with our team to oversee, steer, and assist in the development and deployment of the Alana conversational framework. You will also sometimes find me training new hires and managing Alana’s technology budget.
As a researcher, my focus is on building long, coherent, and engaging conversations on various topics. I’ve poured this expertise into our current ‘Dialogue Manager’ strategies. This enables Alana to have a more natural conversation with its users, covering a wide variety of topics.
What does this mean for how we experience Alana in real terms?
Imagine the sort of conversation you have with a friend in the pub. You may start by talking about the news or exchanging some interesting and fun facts. You’ll probably jump from topic to topic coherently, while at the same time paying attention to your friend’s interests. This keeps the conversation mutually interesting. The ability to hold this kind of advanced dialogue is what we’re striving for with Alana.
What does Alana mean to you?
I was part of the core team that first perceived, designed and implemented Alana. What we’ve achieved makes me very proud and I have an extremely close bond with Alana. I remember fondly when we deployed Alana into the wild for the first time during the Amazon Alexa Challenge. The contest gave us a unique opportunity to test our technology with an immense number of real people. What this showed us is that once you involve actual humans, they’ll always find ways to throw all of your predictions out of the window… no matter how well you think you’ve modelled a user’s language.
This was an important reality check for us as a team and an eye-opening experience that really accelerated our ambitions to take Alana out of the lab and into the real world.
What are some of your goals for the development of Alana?
There’s cumulative academic experience of around 110 years in the field sitting within Alana. This equips us with the knowledge and skills to really push the boundaries of what’s technically possible with Conversational AI. We’re using this expertise to find more and more unique ways to use our technology to support real people.
The companion role of Conversational AI became more pertinent than ever in 2020. The isolation that for many has become an unavoidable part the coronavirus pandemic has made using technology to combat loneliness and connect people more relevant than ever before. My goal is to equip Alana with the capability to better people’s lives using natural conversation.
What are some of the wider developments we can expect to see in Conversational AI?
Modelling human language is one of the most challenging problems the scientific community has to solve. The human brain is incredible at connecting different concepts together. For instance, when I say the word ‘wine’, you immediately think of the colour, the smell, the taste, a nice dinner you had, and potentially a very bad hangover from the next morning. You do this all at the same time. Our brain makes these links based on underlying accumulated experiences. Although AI can represent words in various ways, it still hasn’t reached the point where a machine can make these more complex associations.
That being said, technology has made huge leaps in recent years. I’m hopeful that in a decade from now, we will see humans interacting with machines in a more intelligent way through dialogue. I believe this will happen as machines develop a much deeper understanding of human language.
What’s your favourite fictional Conversational AI?
That is a tough one. Growing up in the 1980s-1990s, I was always intrigued by Star Wars’ and Star Trek’s AI interfaces. (Especially when Captain Picard would say out of the blue, “tea, earl grey, hot” and then voila, he had tea.)
Other influences include Iron Man’s JARVIS conversational AI, or 2001: A Space Odyssey’s HAL 9000. Interestingly, cultural references such as these still play a huge role in moulding people’s expectations of what Conversational AI is. And for us, they’re inspiration for what Conversational AI could be.