My approach to Artificial Intelligence is based on Evolutionary Computing: I develop, study, and employ artificial evolution to produce artificial intelligence.
Specifically, I am investigating embodied AI, that is, AI in a physical or simulated body. This brings me “From evolutionary computation to the evolution of things” as outlined in my Nature paper. You can watch my TEDx talk and my HUB Berlin presentation for an introduction to the general public and read an accessible two-pager about the grand challenges for evolutionary robotics. The technical basics are covered in the original vision paper on embodied artificial evolution, while the articles on the The Triangle of Life and EvoSphere lay the groundwork for possible implementations. A key point of my vision is that the physical incarnation in the real world is essential: In Vivo Veritas.
My long term goal is to develop the science and the technology of physically evolving robot systems. This means robot populations that can reproduce, learn, and develop themselves “on the job” over consecutive generations. In other words, I want to develop EvoSpheres to enable and inspire innovations in applied as well as fundamental research. On the practical side, evolving robot systems represent a (r)evolutionary new way of engineering robots through reproduction and selection. This can take place under supervision on robot breeding farms or autonomously in remote environments, such as other planets. Of course, the related issues on safety and ethics must be addressed, cf. the “How to stop” section in this paper. On the theoretical side, evolutionary robot systems open up new avenues for fundamental research. An EvoSphere forms a novel research instrument to study evolution and the emergence of intelligence. It plays a role akin to that of a telescope for astronomy or a cyclotron for nuclear physics. Conducting experiments with evolving robots we can investigate scientific questions about evolution and the emergence of intelligence in a new substrate, different from carbon-based life as we know it. In words of J.M. Smith, the preeminent evolutionary biologist, “So far, we have been able to study only one evolving system and we cannot wait for interstellar flight to provide us with a second. If we want to discover generalizations about evolving systems, we have to look at artificial ones.” (Nature, vol. 355, 1992)
On 26 May 2016 the VU research team presented the tangible result of a proof-of-concept project: world’s first “robot baby”. The corresponding research paper is published in the journal of Artificial Life. The making-of is shown in the Robot Baby Movie on the Vrije Universiteit Science youtube channel, while the website evosphere.eu provides more info and some scientific background, including related work elsewhere. The Robot Baby Project received much attention in the Dutch media including TV (NOS journaal, Nieuwsuur, RTL Z), radio, and the daily newspaper de Volkskrant. The project got international coverage as well on the technology news site TechCrunch.
… and for fun: the robot baby inspired a cartoon by Hein de Kort in Het Parool (in Dutch)
I was/am involved in various international research projects including:
- ARE (Autonomous Robot Evolution: Cradle to Grave, EPSRC) will start in Q3 2018.
- PHOENIX (EU H2020) is developing methods for evolving small sensor agents for exploring inaccessible environments. See phoenix-project.eu.
- DREAM (Deferred Restructuring of Experience in Autonomous Machines, EU H2020) is investigating robots that post-process daily experiences overnight to become smarter. See robotsthatdream.eu.
- EVOBODY (New Principles of Unbound Embodied Evolution, EU FP7) explored the potential of physically embodied evolutionary systems. See evobody.eu
- FOCAS (Fundamentals of Collective Adaptive Systems, EU FP7) was concerned with all forms of adaptation in all kinds of socio-technical collectives.
- SYMBRION (Symbiotic robotic organisms, EU FP6) investigated modular robotic organisms that emerge by self-aggregating independent robotic units.
- NEW TIES (NEw World models Though Individual, Evolutionary, and Social learning, EU FP6) investigated artificial societies subject to threefold adaptation by evolution, individual learning, and social learning.
- DREAM (Distributed Evolutionary Algorithm Machines, EU FP5) investigated evolutionary algorithms implemented on a group of different physical devices.