Google AI: Lyra

In 2018 I became a Principal Designer at Google AI, and began a project looking at potential personal AI near-futures called “Project Lyra“.

Google AI: Lyra

My original whiteboard

Roles: Principal Designer

In 2018 I became a Principal Designer at Google AI, and began a project looking at potential personal AI near-futures called “Project Lyra“.

I directed this over 3 consecutive years, using a small budget to work with many of the UK & Europe’s most interesting design and prototyping studios including SuperfluxSpecial ProjectsNordNormally and Tellart amongst others.

Diagram of a spectrum of human–AI collaboration

It took as it’s starting point the concept of a truly-personal AI, that while ‘manufactured’ by Google – was secure and private to its owner once adopted, and worked on that owner’s behalf as a agent with larger systems.

An image I used to playfully explain the relationship between a Personal AI (pai), it's owner and the gap between that and the tools they could both engage with as a hybrid 'team'

The exploration lasted for 3 ‘seasons’ between 2018-2021, with each ‘season’ comprising of interlinked projects carried out by different external studios - who we would convene at points to share ideas and work.

At the end of each ‘season’ there would then be a share-out event with the wider community within Google Research and Google product areas, as well as wider dissemination of deliverable materials, protoypes and applications for patents.

I also commissioned essays from academics and writers - often after exposing them to new research and ideas from the computer scientists and engineers working on the cutting-edge.

Concept image from Project Lyra

Their writing imagined the techniques and breakthroughs made through the unique lens of their own research and experience. The final pieces by Professor Andy Clark, Richard Pope, Adam Greenfield and Cesar Hidalgo circulated internally catalysing debate and further exploration.

We made working prototypes in software and hardware both with the partnered studios – and a small internal team of myself, Rob Marchant and Henry Holland.

Using a technique called Tensor Concept Activation Vectors or TCAVs - allowing subjectivities like ‘adventurousness’ to be trained into a personalised model – then drawn onto a dynamic control surface for search. A constantly reacting UI that allows a kind of ‘2-player’ game between the human and their agent searching a space together. We built this prototype in 2018 with Nord Projects.

Some scrappy form-factor prototyping with green duct-tape for adding UX concepts as overlay in video, 2018
Form-factor and UX explorations for a personal AI device (with Map Project Office / Approach Studio) 2018

The work, which centred on how to relate to personal AI agents continues to have impact within Google – as well as resulting in published patents.