In this project, we demonstrate how a generic artificial agent can generate human-level design solutions given multiple design challenges. Moreover, we have created a common environment to test these challenges with the same conditions with humans.
This project is a work in progress and will be published soon. So many information is now
closed for general public.
If the agent is able to create it, are these designs on human-level? We want to understand, given the same challenge, outputs similarities and performance of both human and artificial designs.
Design is mostly a collaborative process, with many individuals interacting with each other. This generates a lot of valuable knowledge that can help with future designs. How can we allow a bidirectional learning transfer and collaboration between artificial and human designers?
By incorporating artificial designers, are current human capabilities augmented or enhanced?
Human capabilities on designing solutions to problems to are based on our experimentation and interaction with the world. Since we are born, we are able to perceive our environment through our senses, learn about complex situations and create mental models about how the world works.
We are able to extract complex patterns, relations between elements, anticipate behaviors and transfer learning to ourselves and others based on previous experiences.
In this work, we explore how a generic artificial agent, without any previous information, is able to learn how to design an object to solve a given challenge through multiple simulations. This first step, will help us to compare agents results with human designs in terms of creativity and efficiency. Morever, it will help us to understand how humans designs and artificial agents can collaborate together, transfering learning between them in any step of the design. process
I am responsible for all process from research definition, to concept creation and final implementation and testing.