Peter Norvig is a director of research for Google and contributed to a fascinating Dreamforce Panel this year, entitled "How AI Is Transforming the Future Of Business", alongside Suchi Saria of Johns Hopkins and Richard Socher of Salesforce.
This is the second part of our interview with Norvig, a true AI Trailblazer. Part one — covering the state of play in AI today, and how companies can leverage AI for better decision-making, is available here.
We’re beginning to see AI algorithms — via chatbots and voice technology — becoming the face of your brand. Indeed, that’s the focus of a recent Harvard Business Review piece, co-authored by Paul Daugherty, whom we interview here, and Nicola Morini-Bianzino, who contributes to this article.
Companies are having to think about how to design their own experiences and interactions with customers over these new conversational interfaces. That leads to a lot of big decisions about whether to make an AI program be transparently human,or be transparently a technology.
What advice would you have for companies taking their first steps into designing experiences and interactions over conversational interfaces?
That’s a tough question, because ultimately this is all new. No one really knows how to do this yet.
As a result, I think it's better not to pretend that AI is a human because you're never going to make it there. You’re always going to fail to some degree with the current technology available. The more a program or app says, “Hey, I'm a person. Talk to me just like I'm a person,” the more people are confused. People are confused because they don't know what they can say that the technology will understand. They don't know what works and what doesn't work. And when the technology fails, they’re disappointed and frustrated.
With current technology, it is safer to aim for an interface that is more functional, and doesn't try to be human-like or have a personality. Perhaps you're not exactly sure what the boundaries are of what you can and can't say are, but you do have a model of the kinds of things that'll work, and can use sentences to replace the mouse clicks your users would have otherwise done.
Having that type of model of “here's what you can say and here's what you can't say” for your users is important, whether you use a voice system or a text-based chatbot. It's a hard challenge. Part of the problem is that you train people to do one set of things with your AI program, but then your system gets better. There’s a whole new set of capabilities, and they don't know all the new things that will now work.
I think that as an industry, we're going to have to figure out a way to convey those new capabilities to users as they start to come online. I'm not sure how that's going to work out right now.
There’s a growing consensus that AI is more likely to replace specific tasks, not entire jobs. So is there anything we can say about the type of tasks humans will be doing in the future, versus those that'll be done by machines?
A lot of what we'll see is a mixture of humans and machines working together.
And part of the human role will be in exercising the right kind of judgement. Computers are better at data-heavy tasks, like predicting how much money a retailer will make by suggesting a particular product to a particular customer. But humans are better at figuring out if a recommendation will be embarrassing or just odd. In short, humans are better at being humans.
As a result, it seems like we're ending up in a position where we're dividing up the roles in a way where the computers can do the more rote tasks, and humans will do the tasks that require higher-level understanding, or empathy.
So what are the sort of skills that companies should begin to hire for, or develop in their existing workforce?
Creativity, empathy, and communication will all be key, as will be the ability to interact fluidly with other people and with computerized tools. We'll see automation throughout various service industries, but I think of it in terms of tasks that will be automated, rather than in terms of jobs.
For example, decades ago, the task of counting out bills in a bank was replaced by ATM machines, and the tasks humans completed became higher skilled — approving loans, and talking with customers about complex financial products. We'll see that kind of change across industries, where the routine is mechanized, and employees will need to be flexible and creative in dealing with the non-automated demands of the job.
Peter Norvig was one of several AI luminaries to speak at Dreamforce 2017, alongside Vivienne Ming of Socos, Suchi Saria of Johns Hopkins, Paul Daugherty of Accenture, and Richard Socher of Salesforce.