How might we design AI that’s less discriminatory, stereotyping & exclusionary?

Process

To answer the hypothesis, I conducted expert interviews with people working in and with inclusive design, voice user interface design, gender sensitive design. Afterwards, eight cultural probes were conducted with users of voice assistants to understand and analyse users’ perception of different voice assistants. Among other things, the participants were asked to draw the persona of the voice assistant. The result shows the clear perception of this persona as a human-like figure.

I identified three main problems regarding the developed personas:

Through anthropomorphic design, the virtual personas perpetuate gender stereotypes and exclude non-binary and queer individuals while normalising and tolerating abusive language.

Result

The aim of the thesis was the development of an inclusive voice assistant persona to prevent the retention of these problems and to present an emancipated as well as empowering alternative to common voice assistants.

The result of the thesis is the humanising voice assistant Doubletwo. The virtual persona is dehumanised, representing a gender-neutral and robotic entity with its wake word, sound of voice and behaviour. The dialog management was developed to no longer trivialise off-topic, sexual or offensive user commands, but respond accordingly: pointing out misconduct, reminding the user of the robotic nature and breaking up societal gender roles.

In order to present the newly developed persona Doubletwo as a voice assistant alternative, a voice prototype was developed with which users can interact in time and space. See the result of the voice in the video below.