The state of the sector, through covid and beyond
Covid-19 has brought an increased urgency to the digital transformation agenda. In just a few weeks, we’ve witnessed the attitudes of many organisations change, from a sometimes only theoretical interest in technological innovation, to making solid commitments to innovation projects.
It’s not difficult to understand why. In the month of April alone UK GDP fell by 20.4 per cent, reflecting the closure of entire industries during the coronavirus lockdown. Many sectors – such as the airline industry, travel, and hospitality – continue to struggle due to ongoing restrictions, and organisations across the board are tightening their belts in anticipation of difficult times ahead.
Under these circumstances, any solutions that enable cost savings or increased efficiencies have soared to the top of the priority list. Automation, and conversational AI in particular, is notable for its ease of implementation, scalability and demonstrable return on investment. Under challenging circumstances, it’s no wonder that organisations are increasingly turning to these tools.
From ‘should do’ to ‘have to’
For Tim Deeson, CEO of GreenShoot Labs, covid-19 has turned conversational AI from something that organisations should do, to something they have to do.
“These days, there is much more of an appetite for rolling out conversational AI solutions,” he says. “The hospitality sector, for example, is running at greatly reduced revenue compared to last year. It’s not enough to justify the high people cost of things like customer services – particularly when there are technology-based solutions available.”
Another factor is increased service and support needs around the virus, for both staff and customers. Organisations are receiving many more enquiries, and must deal with complicated guidelines which could change at any moment.
“The best way of dealing with high volumes of enquiries is through conversational AI,” Deeson says. “For example, a large hotel chain we have been working with came to us so that they could provide booking support. They are receiving all their normal booking enquiries, plus a high volume of inbound voice calls due to uncertainties around covid.”
| “The best way of dealing with high volumes of enquiries is through conversational AI,” |
“Our solution can fulfil basic functions on behalf of the customer or escalate more complex queries to the right human agent. It also enables the business to centralise their call system, rather than having customers phone up individual hotels which may not have the available staff.”
An enthusiastic outlook
One of the advantages of conversational AI as a technological solution is its accessibility. It’s intuitive, everyone can use it, and it is possible to see it in action. This makes it an exciting new tool for organisations to implement, as they can easily understand how it can make a difference to their teams.
Like all technologies, advances in conversational AI are being made all the time. We’ve recently seen developments in the form of Google Duplex, the Meena chatbot, and GPT-3 from OpenAI – a language model with 175 billion parameters, that is capable of tasks such as summarising text, text generation and translation. Considering the previous GPT-2 model had only 1.5 billion parameters, this seems like a huge leap ahead. But what does it mean for the industry in practice?
“It’s been interesting to see the release of GPT-3, because it has generated headlines and curiosity from much wider groups than we would normally expect,” Deeson says. “The model emulates some specific aspects of human intelligence in much more realistic ways than before. You can feed it some starting content and it will then create more of that kind of content – a use case might be something like the automatic generation of dialogue in a video game.”
“That said, it is quite a specialist, narrow technology. I’m always a little bit cautious that people will have unrealistic expectations for the technology or want to apply it in ways that it doesn’t have a relevant foundation for. It can certainly appear intelligent in ways that it’s not: looking at previous content and generating text from that is very different to understanding the world from a logic based perspective…”
Beating the bias
Any discussion of AI, conversational or otherwise, should always acknowledge the issue of bias. It may never be possible to eliminate bias from data and decisioning systems, but we should be actively trying to understand and minimise it.
For Deeson, this is another consideration for advances such as GPT-3.
“The AI model inherits bias from the data sources it consumes, which can make the results very problematic,” he says. “These tools work in ways we can’t predict. When you combine that with the invisibility of bias in data and content sources, it’s difficult to know what they will do. The right question to ask is therefore not just ‘does this technology work?’, but ‘does it work in an appropriate way for this context?’”
| “The right question to ask is therefore not just ‘does this technology work?’, but ‘does it work in an appropriate way for this context?’” |
As newsworthy as a new 175 billion parameter model may be, it’s more important to think about how it could be used in a real world context. This may be reliant on adjacent factors – secondary developments (perhaps not even technological) that help the tool to be implemented, or which help to address its bias.
C is for conversation
Ultimately, conversational AI is an exciting technology because it enables the user to interact with it in a completely natural way. This provides inherent opportunities around usability.
“There’s nothing inherently natural about using a keyboard or a mouse,” says Deeson, “but we are all intrinsically verbal. Good conversational AI is when people don’t have to modify their expression in order to interact with a computer.”
“The user experience of conversation is therefore extremely important. How do you model and design successful conversations? The technology will increasingly become more sophisticated, but we see in the market a lack of sophistication around how to design. This is something we will continue to develop as a service at GreenShoot Labs.”