4 things the co-founder of Cortana told us about digital assistants

Posted 29 January 2018
Cortana, the AI powered virtual assistant

Whilst attending Exponential Unknown, an event hosted by Wired and frog Design as part of the London Design festival 2017, aiweek got the chance to speak to Ed Doran, a Microsoft Researcher in artificial intelligence (AI).

Ed Doran is the co-founder of Cortana, and leads the strategic planning for the evolution of Cortana, so we grabbed the opportunity to ask, “what makes a digital assistant successful?”

Here’s what he told us (subheadings our own):

  1. Successful digital assistants serve a purpose

Ed Doran: “The big, fundamental design question is: what are you going to do with this technology that will be meaningful for people’s lives? You have these capabilities, but how do you apply them, and who do you apply them to in a way that enhances their life?”

Ed Doran is the co-founder of Cortana and an AI Researcher at Microsoft
Ed Doran, co-founder of Cortana

“With voice interaction, for example, we’re reaching an inflection point where it’s roughly as error prone as normal human speech. Now we need to talk about systems that not only understand you but have something to say.”

“What can a bot say that’s helpful to you? I think that’s a design question that we should be addressing.”

  1. Successful digital assistants don’t have to be human-like

“I think you have to think carefully about what you’re trying to deliver. I don’t think we should be trying to deliver AI replacing humans, instead we should be trying to deliver AI that enhances humans.”

“We all have these amazing minds: we can explore, we can experiment, we can do all these things that machines can’t. So, let’s put humans in a position where they can enjoy that, where they can execute that.”

“The goal is to make it easier for humans to do whatever they want to do, to discover things that they didn’t know before. If I secretly love Icelandic Death Reggae and I just don’t know it, recommend that to me, help me find it.”

  1. Successful digital assistants deliver on performance, before personality

“The mark of a successful digital personal assistant is not that it has a persona, it’s that we’ve made a place where you can ask questions and can get whatever you need without digging through menus. It’s that we’ve found a way to sort through your inbox and surface up the things that are most important to you and push back the ones that aren’t, without you even having to ask.”

“These implicit augmentations of productivity feel incredibly useful, because if you think about your work life, it’s very complex. You have a huge amount of people you’re interacting with, you have a huge amount of technologies you’re bridging across.”

“The things that work quietly in the background, and make the tools you already know how to use more productive, easier to use, more effective, really end up being the most powerful.”

  1. Digital assistant builders should consider cloud based training

“I think you can train in a number of places. If you’re willing to be asynchronous, and you’re willing to integrate across several different sources, a cloud-based approach offers unique training capabilities. Now, in some certain circumstances due to issues of latency, or privacy, or network utilisation, you’re not going to have that chance. Or you might not even need that level of power.”

“I think if the cloud can span multiple inputs, and has the ability to leverage that toolchain, then the interesting thing is: can you see things that you wouldn’t have seen with the limited data set that you have from a single instance, like a single device?”

“Perhaps, using the cloud, you could end up with models that are more representative or more predictive, because you can take a more inclusive view of either a person or a population. So, cloud processing is really interesting, and then deployment out to an individual device lets you realise that potential.”

aiweek: So there you have it, some pearls of wisdom from a titan of the industry. Of course, some of you might be thinking, “I’m not taking advice from anyone at Microsoft about digital assistants, they can’t even control their own!”

And of course, 2017 got pretty embarrassing for Microsoft on the chatbot front after their Tay chatbot was trained to become an anti-feminist Nazi and their XiaoBing chatbot, based in China, was shut down for being ‘unpatriotic’.

We asked Ed how that had happened, and how Microsoft would be changing their approach to chatbots moving forwards, but unfortunately we got a very clear response:

Ed Doran: “Apologies, can’t comment on that.”