Gender, bias and stereotypes: the ethical dilemmas in voice

As synthetic voices are getting better and better, the sort of “is it or isn’t it human” is interesting, and clearly will become more of a challenge when people feel like they don’t know whether they’re talking to a synthetic voice or to a real voice. I mean, part of a study that BBC R&D are doing right now on public perception of synthetic voice. And I’ve had my voice synthesised. And so you can hear a synthetic version of me. Obviously I think it’s quite cool. But not if, in four years’ time, I stumbled across a synthetic version of me saying something I’m not very happy with. So there’s all sorts of legal and ethical considerations around synthetic voices.

Amazon have released a Samuel L. Jackson version of Alexa. You know, people know that’s not Samuel L. Jackson. And it’s a sort of gimmick. It’s a bit like having your satnav as a kind of celebrity. I wouldn’t be surprised if it’s one of those things that we feel like people may have a problem with, and in fact they won’t. 9 times out of 10, maybe 99 times out of 100, the aim isn’t to convince somebody that they’re speaking to a human. In fact, if they believe that they’re speaking to a human, they might have expectations that are by far and are way above what the technology can actually deliver right now. Most people just care about getting something done.

They care about accessing content. They care about accessing services. They care about getting an answer to a question, or playing music, or turning off their lights, or getting an insurance quote, or taxing their car. Whatever it might be, they don’t care about the interaction. They care about the end result that the interaction is there to serve. There’s a guy called Oren Jacob who is the founder of PullString. They were acquired by Apple this time last year. And we’ve had him on the show in the past.

And he gave one of the best analogies that I’ve ever seen applied to this space, which is that if you are an animator, the aim isn’t to recreate a realistic scene to fool the watcher or the viewer into believing that this is the actual, real thing. It’s to just get across the idea and the concept, so that when you see a waterfall in an animated picture, you understand that it’s a waterfall. And the same thing is true when it comes to conversation design. We’re not designing actual conversations with real humans.

We’re just trying to find the essence of the conversation and some of the techniques that we use in conversation, like confirmations, like grounding strategies to recover when a conversation veers off track. We’re using the techniques that we use in conversation to make the interface more intuitive and to help people get whatever it is that they need doing, or to get to the point, the whole purpose of them interacting in the first place. It’s just a means to an end. And so they don’t need to be humanistic. And if they are totally lifelike, if we get into that kind of uncanny valley, then there is a risk that people have inflated expectations that the system and technology just can’t manage.

I do think conversational assistants should have a personality, because that’s what defines them from another personal assistant. And I think we’ll get to a point where there will be a number of personal assistants. And we’ll decide we want to have this one for our music, and this one for our entertainment. And it might be that some of them, we can personalise to the people or the voices that we choose for those individual tasks. But it will help to have some distinction. And just because conversations are with people, people are personable, it doesn’t mean they have to become your best friend, or you fall in love with them, or all these sorts of things. But to have character makes it work.

When chatbots don’t have any personality- it’s just functional, it’s just utility- I think they are not so effective. How you define what personality your persona should be is really interesting. Because then it’s sort of like a brand thing. It’s like any brand strategy. So if Greggs had a chatbot- maybe it does- if Greggs had a chatbot or a voice assistant, what is the Greggs personality? Well, we kind of know it’s kind of cheeky. And how they translate that to a chatbot or a voice assistant is how their brand strategists do it. But I do think it’s important. We can’t help but anthropomorphize when we hear a voice.

And that means that we can’t help but assign human characteristics to something, and perceive it as if it was a human. So for example, people call Alexa “her” all the time. It’s not a female voice. It’s a text robot voice. But it’s in the same frequency range as a female voice would be if it was a real voice. In that respect, there is ethical considerations. And a lot of those ethical considerations are often on the side of the large technology providers. For example, there’s studies out there that show that conversational AIs can be racist, prejudiced.

And some of the theories are down to, if you take it far enough back, the people who create these technologies, largely, if you generalise the whole thing, it’s typically white males. And so there is the theory that there is unconscious biases that seep into this technology. There’s a really fantastic book called Wired for Speech by Clifford Nass and Scott Brave. In that book, they delve into human psychology as it relates to speech. And the studies that they cite in the book suggest that, when we hear a voice, we cannot help but, in our minds, attribute certain attributes to it. We can’t help but give it a gender.

There’s a really interesting project at the moment called Project Q, which is attempting to create a gender-neutral text-to-speech voice. However, if you ask 10 different people, five of them will say it sounds male, five of them will say it sounds female. And that’s just because our psychological nature over hundreds of thousands of years worth of being human, we’re just programmed to assign those attributes to it. There’s been quite a lot of studies around the fact that, currently, voice assistants are female, and you’re asking them to do things, and they are serving you up with information. And is it a dominant relationship that you have with, effectively, your kind of computer slave in the living room, or your kitchen, or wherever.

So maybe you’re thinking, well, OK for my Greggs voice assistant, I should have it male. And that’s like, well, OK, then it’s male. Does that mean everything’s going to be male, because we don’t want it to be female because female means that it’s an assistant? So therefore is all of it going to be male? And then what happens to women? We want it to be a kind of balance. Well, hold on a sec, why do we have to have just male or female? Should we have gendered voice assistants at all? Do men prefer to have a male voice assistant? Do women prefer to have women.

There’s all sorts of interesting- you can imagine you open up that Pandora’s box of ethical concerns. And at some point, you’ve got to make a decision. And it may be the decision is we offer a number of different alternatives, and you choose.