03.04.18

Prototyping a collaborative tool for human-machine ideation

Written By Leonardo Amico and Mike Shorter
6 mins read

As part of our ongoing exploration of AI and creativity, we decided to look into the role computers can play in idea development. It’s a key step of the design process and probably the hardest one to crack; but we believe that intelligent technologies could play an interesting role in the task.

Our goal was to see if it was possible to use AI to create moments of “wonder” which would stimulate our imagination during idea generation sessions. We were particularly drawn to this concept through inspiration from two very different examples.

The Background

Through Discover Weekly, Spotify offers their users a new playlist every week of music they might like, by combining data from user playlists and the analytical qualities of music. At times the algorithm recommends songs that would hardly fit in our music collections, yet they become unexpected favourites of ours – creating this AI-driven moment of “wonder”.

David Weinberger’s article for Wired’s Backchannel discusses the implications of machine-learning technologies in scientific discovery. Modern science often deals with complex systems and huge quantities of data, making the task of creating a mental model often impossible. “Rather than reducing phenomena to fit a relatively simple model, we can now let our computers make models as big as they need to” – in other words, machines can understand our world better than we do. And we can still perform effective tasks of classification, prediction and so on through them – despite the fact that we don’t fully understand how it works.

In both these cases, the software creates something. Whether that’s knowledge of the world or just a pleasantly perfect song recommendation, it stirs in us the very human sense of “wonder” – with no human directly involved.

The question was, could we use the same approach in a tool to foster creative work?

The Prototype

Our research began with a one-day design sprint with other creatives. One of the ideas – almost certainly influenced by our environment, the aptly-named Cork Room – was about the use of AI technology in visual mapping.

Visualisation is a powerful tool for helping us come up with ideas, and we can achieve great results using Pinterest and other image search engines to find related images. But when used as a creative aid, we find ourselves viewing a series of similar objects, reinforcing rather than expanding our initial vision; far from ideal when we are looking for an inspiration boost.

The aim was to create a system that could find and show visually similar but contextually different images. Our creative task would then focus on coming up with ideas that try to make sense of, and build upon, that loose connection. Finally, the system would have to be brought into a shared environment for collaborative use during group sessions.

By using open-source machine learning software and training it with different data sets, we started to experiment with the idea, until eventually, we arrived at a process that we were happy with.

As an example, here is how we used the system to come up with a concept for a home speaker.

The first stage was about finding wild associations. We started with images of speakers and people at home, using the system to find similar images from a database of items and people in context. Finally, for each loose association, we discussed ideas that connected the initial image with the suggested ones.

Finding wild associations using images that look like, but are not, a speaker.

In the second stage, we set out to explore concepts related to one particular direction we had found. We asked the system to find images related to the suggested image, using a model trained with pictures of music speakers.

Exploring concepts with speakers that look like the wild association image from the previous stage – in this case, a motorbike.

From this experiment came two insights. The first is that the model should be trained with images from different contexts (in our case, motorcycles with speakers) in order to result in unpredictable and inspiring associations. The other important aspect is that the output images should be comparable in style to the input, ensuring that the associations between input and output images are somewhat meaningful.

Finally, we moved to the corkboard itself. Making a prototype using a computer-vision system that would allow us to use the tool collaboratively, we turned one of our familiar corkboards into an interactive platform. The projector-camera captured the images pinned to the board and then displayed related images as found by the AI system onto blank sheets of paper.

The prototype in action.

The Augmented Corkboard is our idea of a system to stimulate creative sessions between people and machines. This approach could be used to boost creativity during brainstorming sessions, and in particular for coming up with interesting stories and ideas for areas such as advertising campaigns, photography, product design and invention. In the future we plan to test and refine this computer-aided approach to idea development in actual creative briefs.

We’d like to hear what you think about our prototype and the idea of using AI as an ideation aid. Get in touch through Twitter and Facebook to share your thoughts!

 

If you enjoyed the article then please share