Silicon Republic editor Elaine Burke discusses the future of AI and voice interaction with experts and industry partners from the Adapt research centre.
Artificial intelligence, or AI, has been part of popular vernacular for some time now. For many in my generation and generations before, our first encounters of the AI kind likely came from the fantastical imaginations of filmmakers and fiction writers.
The future these sci-fi stories explored has arrived in many ways but, in the 21st century, AI is not so much the star of the show as a member of the supporting cast. By and large, we are using and interacting with AI-led technologies every day, and many users don’t even acknowledge it.
“Most people are using AI consistently in their daily life. They carry it around on a phone and are not thinking about it. And they’re still like, ‘When will AI arrive? … Alexa, play Spotify,’” joked Shane Lynn, CEO and founder of EdgeTier, during a panel discussion on the topic hosted at Future Human 2020, in association with the Adapt research centre.
‘Empathy, humour, negotiation, debate – they’re actually fundamentally the hardest bits for computers to emulate’
– SHANE LYNN
AI’s issue, as I see it, is bad branding. There’s an extensive catalogue of science fiction exploring its far-reaching potential, but hardly paying tribute to the true magic of the technology we currently use. This technology is the nuts and bolts of digitalisation. It’s driven by computational efficiency and it’s working at its best when it’s blending into the background. It isn’t the sexy movie star but the projectionist behind the curtain, keeping the boring stuff secret so that we can enjoy the magic.
And, true to this form, practical working AI is not what grabs the public’s attention. EdgeTier’s flagship product, Arthur, uses AI and machine learning to assist agents in contact centres. But, as a lot of media maintains an apocalyptic attitude towards this technology, it could become a headline screaming about AI taking jobs from humans.
But for Lynn, the best bits of AI are the invisible, non-invasive elements. “Where there are revolutions happening is in accuracy around speech analytics, around voice recognition, around translation. And they’ve had fundamental changes to the business models of the companies we’re working with,” he said.
Think of the predictive text that unobtrusively helps you to speedily complete that business email. Or the YouTube captioning bot working in the background to give you that first draft transcript for your video subtitles. “The sophistication of that technology is huge behind the scenes, but to the user it’s just there and it makes a big impact,” said Lynn.
“For our clients, we do live translations in chats as they happen and agents can be talking to someone else in the world completely seamlessly. It sounds like a small change but it has fundamentally changed their hiring processes, their staffing, the way they manage languages. Just a step-change overnight, whereas that didn’t make any headlines.”
The year of Covid-19 may well have boosted the jump in businesses adopting assistive AI. In the case of contact centres, many were suddenly launched into forming strategies for remote working from scratch. Not only did this drive digital transformation at these businesses, it’s driving new innovations at EdgeTier too, as company works to build tools to support entirely new team structures.
The pandemic accelerated voice technology development at SoapBox Labs, too. This pioneering technology company specialises in the development of automated speech recognition for kids, and director of speech technologies Dr Amelia Kelly said project demand has suddenly increased. First, a partnership with US edtech firm Amplify was pushed a year earlier than originally planned and, Kelly said, this has spurred other educational providers into pushing ahead also.
“People have always just been a bit tentative,” she said, blaming the “creep factor” of distrust in powerful AI that hasn’t been helped by the aforementioned sci-fi computer villains or ‘robots will take your jobs’ headlines. But now, “They’re realising that, no, actually this technology is here to make the teacher’s job easier,” Kelly added.
SoapBox Labs is also working to allay concerns around this technology by working with “no ulterior motive”, as Kelly put it. “We’ve got very, very sophisticated privacy protocols and we also have a mission statement that we never use data for anything other than to make speech recognition better.”
As far as Kelly is concerned, “There is no reason in the world that the voice data of a child talking to their teddy bear should be sent up to Amazon or Google.” And this data protection philosophy will be applied to the latest technology being developed by SoapBox Labs in collaboration with NUI Galway and California tech company Xperi. “We are developing speech technology that works end-to-end and fits on a tiny microchip for the toy industry, essentially, so that basically there’ll be an air gap between the child’s toy and the cloud,” said Kelly.
“Imagine: the child can talk to the toy, have a natural interaction with the toy, the toy can talk back to the child, have a conversation with them, but none of that data will ever leave your home. No big company will ever get their hands on it.”
It’s no surprise that SoapBox Labs has linked up with a university on this ambitious project, as all these advances are underpinned by the work of researchers exploring the boundaries of this technology. For Dr Robert Ross, a funded investigator at Adapt, the pandemic presented both new opportunities and challenges to this research.
On the one hand, the shift to remote working and a reliance on digital interactions meant a lot more rich data was being collated in this area. All of a sudden, you had massive amounts of group conversations happening online in work and academic contexts, often with video capture as part of the process.
‘There is no reason in the world that the voice data of a child talking to their teddy bear should be sent up to Amazon or Google’
– DR AMELIA KELLY
“Suddenly we have the opportunity to collect so much information in a different way. And coming from university settings, we can potentially get classes and groups involved and willing to give over their data in that way,” said Ross.
This data surge was particularly inviting to PhD students at Adapt looking into detecting user engagement in online video interactions. But for Ross, who likes exploring human-robot interaction in the physical realm, research in this space was put on ice under social distancing guidelines and lockdowns.
So while private companies are learning how to build trust in AI and we’re all getting used to assistive tech to the point of not even acknowledging its existence, researchers in this area are working hard to advance this technology even further, to the point where it may break out from performing menial, tedious grunt work and into more of a leadership role.
“If you look at the cost of AI and certain types of technologies, particularly as we go towards the higher end, companies are starting to see that cheap human resources are cheap and actually where they’re spending lots of money is on the most educated or the highest-powered decision-makers within an organisation,” Ross said. “And they’re actually beginning to ask us questions in terms of how we can facilitate providing assistance at the highest levels to what would be well-paid positions.”
Looking at this with a positive spin, it’s about providing decision support systems and tools that derive insight from data to leadership teams, helping them to be more efficient. But Ross isn’t so optimistic. “The little cynic in me also sees a negative side on how that can so easily be said to start to devalue some of those higher-value roles and taking away from them,” he said.
Lynn argued that AI will never supersede humans in this way because of the need for emotional intelligence in leadership. “It’s the stuff we take for granted as humans like empathy, humour, negotiation, debate … when you have a good customer experience reaction or a good manager, they’re the bits that those people embody and are great at. But they’re actually fundamentally the hardest bits for computers to emulate.”
But for how long will that be the case? According to Ross, giving machines emotional intelligence is exactly where research is being directed.
“10 years ago, the research we were doing was on basic dialogue state tracking, was on basic dialogue planning, putting Markov processes to work to get the basic infrastructure of the types of technologies we work with moving,” he said. “We don’t get funding for doing that any more. We get funding and we have the students working on things like detecting the emotion in our interlocutors as we’re working with them. We get the funding for expressing emotion back to the users and tracking all of that nice stuff.”
All of this is still at the very early stage in research labs only, but that’s not to say it won’t be out there within another decade. So at least that’s something to inspire the next generation of science-fiction writers.