Collaborative AI provides inspiration for designers
Researchers at Aalto university have developed a machine learning algorithm that collaborates with designers to create visual collages, called mood boards. The AI is integrated in a design tool, and makes recommendations, explains its choices, and adapts to the designer’s problem-solving process over time.
Inspiration can be hard to support using a computer. Creative processes evolve quickly and are exploratory in nature, so computers can struggle to keep up. An AI algorithm that gives suggestions that are weird and random is not useful and likely to be ignored by the designer, and an AI that is too conservative and only returns suggestions that match what has already been used is also not creatively useful. However by using a novel machine learning technique called a Cooperative Contextual Bandit (CCB) method, engineers and designers have developed an AI that strikes this creative balance.
Janin Koch, the researcher behind the work, says, "The advantages of the CCB method is that it is able decide whether to follow the current visual idea, or to explore new ideas depending on the current mood board. The ability of the new AI to work alongside designers encourages a form of human-computer collaboration in the design process."
The AI also learns from the current designer’s selection behaviour, and uses this behaviour to refine its suggestions. As part of their study the researchers asked a group of 16 designers to create mood boards with, and without the AI helping them. All designers use mood boards as inspiration tools in their work, from design fields including architecture, fashion, and graphic design. In the example on the right, the designer used the AI in the top image while working on a brief for a Health Store, while in the second worked without the AI on a banking firm's youth program.
The key finding was that the adaptability of the AI allowed it to help designers both when they were “on a roll” and when they were “stuck”. The AI can decide when to offer more explorative suggestions, and when to stick more closely to what’s already chosen. The designers therefore perceived the suggestions as novel without being too eccentric or useless. Even if the AI surprised the designer with unusual suggestions, as these were not completely outlandish, they were more inclined to find them original and useful. The majority of the designers therefore used the AI’s suggestions in their mood boards, and the AI-suggested images made up over a quarter of their final boards.
Further details about the study are available on the research group's website. The details of this successful trial are being presented at the CHI 2019 conference in Glasgow, and the team are now looking to apply the collaborative AI systems in professional context. “The next stage of our study will involve getting the AI out into the working practice, like being used in design agencies over the summer so we can see how it works over time,” Janin says.
Contact details
Name: Janin Koch
Languages: German, English
Email: [email protected]
Name Professor Antti Oulasvirta
Languages: Finnish, English
Email: [email protected]
Name Professor Andrés Lucero
Languages: Spanish, English, French
Email: [email protected]
- Published:
- Updated:
Read more news
14 projects selected for seed funding to boost collaboration between Aalto, KU Leuven, and University of Helsinki
The funded projects lay the groundwork for future joint research endeavors, reinforcing the strategic partnership’s goal to fostering impactful and interdisciplinary collaboration.
Two new assistant professors started at the Department of Electronics and Nanotechnology
Two new assistant professors were appointed in the Department of Electronics and Nanotechnology at Aalto University's School of Electrical Engineering. Read the interview with Kim Kwantae and Paul Verrinder.
In a first, physicists show how to use the Helmi quantum computer in Finland to design topological quantum materials
A team of Aalto researchers demonstrated how the Finnish quantum computer Helmi can be used to understand topological quantum materials, providing a stepping stone to designing exotic quantum materials with quantum computers