As artificial
as it gets

A reflection on designing for
voice user interfaces

Voice User Interfaces are an emerging technology that has consolidated itself through the main tech companies as digital assistants. These interfaces are activated through voice, calling for the redefinition of the user experience.

Designing has been a visual task mainly, beginning with the print medium and shifting to screens and their different size variations. Once the first digital assistants entered the market, people were confronted with having to converse with machines, and the ones behind this product have to redefine how to best engage with the users.


Humans are social animals

The task of designing for VUIs is if an unique nature. As Clifford Nass and Scott Brave have thoroughly explained in their book Wired for Speech, humans are social animals who draw their most important interactions from social cues implied in speech, without consciously thinking about it.

Once machines are able to interact with the user through conversation, people automatically engage in the social protocol of social interactions. Therefore it is not enough for a person to syntactically and semantically follow what a digital assistant says, but foremost the context pertaining to the exchange is paramount for the user to accept this interaction without becoming creeped out by it.


As artificial as it gets

VUIs are obviously different from GUIs. In Graphic User Interfaces, the User is presented with options that rely on people’s visual capabilities, their sense of organization and priorities that will hint to the wished navigation through the medium presented. Once the interface solely relies on voice, these cues are words, a.k.a. prompts that are usually presented by the digital assistant or bot in question:

“You may ask: the weather in Berlin, the weather forecast for Berlin for the next 5 days.”

A skillful designer will carefully select which prompts are most appropriate for the desired exchange in order to save the user’s time and make this interaction most efficient. Learning the mentioned nature of social interactions, the goal of making this experience seamless seemed unattainable, since in my research most of the attempts at making a machine sound human, fell in the uncanny valley.

Therefore the hypothesis of making the interaction deliberately artificial intents to offer the user with a transparent interaction, and hopefully realistic expectations concerning what they will get out of the machine.


In order to test this hypothesis, an experiment was designed around the interaction between user and machine. Some guidelines and materials were used to assess the exchange among the user and the machines. An interface was built for the users to talk to the machine about five subjects they could choose from. For a detailed account of the methodology and tools used in this experiment check the Medium article I wrote.


The catch in this experiment, was that the user was interacting with a person who was enacting the role of a machine through an interface that mediated the interactions. These computers were connected through a server and google hangout conference, in which the cameras were disabled.

The intent of the experiment was not to prank the user, but to research how machines should behave, through role playing as a research method. The person who is enacting, gets to immerse herself from within, instead of analytically giving their opinion. The exacts tools used in the role play can be found here.

People could choose which profile the machine should have and also the different voice. The desired message is typed into the text box and once they hit enter the message is spoken with a mechanized voice in the user’s end.


Do as I say not as I do

From the information compiled during the experiment through observation, and subsequent interviews, some insights were collected. The hypothesis was proven to be false and the research question was partially answered. The answers and main insights are listed into the following bullet-points:

  • People familiar with Siri spoke slowly in the beginning, but as the interaction progressed and they realised that the machine could understand them well, so they spoke normally.
  • No one was abusive to the machine, quite the opposite. Once the machine acknowledged gray areas topics, people remained respectful.
  • The people who enacted machines were more dominant and wittier, daring the human. In turn the people were welcoming the quirks of this machine’s personalities.
  • If the answers don’t come quickly, the user grows impatient or they think the machine stopped working. That’s the challenge of having people type responses instead of choosing them from a pre-established responses or speaking them directly.
  • The machines who spent a lot of time speaking to the user in a humane level, also say they would rather prefer to invest in interpersonal relationships instead of relationships with machines.
  • The users who mentioned they did not trust the machine, spent the longest talking to it and also shared most intimate content.
  • According to this experiment people would indeed have machines as their as a friend, is that was a possibility. Therefore the hypothesis that having a robotic interface would limit the interaction, is invalid.

Next Steps

This experiment was conducted in a very artificial environment. The next step would be to see if this behavior would be the same in the person’s home environment. The idea would be to have an object that is personalizable by the person. This object would become the vessel of their personal digital assistant. Would people still keep this level of intimacy with a machine, in their own home?

In order to answer this question, I look forward to collaborate with other areas of expertise that could contribute to the next experiment.

Joana Francener

I’m an service and experience designer based in Cologne, currently researching human and machine relationships. I especialized in user experience research and facilitation of design thinking workshops.

Read other case studies on Medium and get in touch at .