位置データから物体が「生命」か「生命ではない」を判断する「偽生命」認識コンテストが開催されています。当コンテストには研究員のラナ・シナパヤ(Sony CSL / ELSI)が共催しています。
【Fake Life Recognition Contest】
The submissions for the Fake Life Competition are now open, with a cash prize of USD1000, sponsored by AI company Cross Compass Ltd：https://competitions.codalab.org/competitions/20612
The problem of identifying living systems from non-living one is a difficult question, which research fields such as artificial life or origins of life have been trying to address for decades or more. One can probably tell by sight that a walking line of ants is made of living things, while a flowing river is not. Can a computer tell apart living from non-living things as well as we can? The question is not settled in theory, but we had the idea of trying to settle it in practice.
The Fake Life Recognition Contest is co-organized by Lana Sinapayen (Sony CSL / ELSI) and Olaf Witkovski (Cross Labs / ELSI) , and supported by Cross Compass Ltd. The competition has just been announced at the ALIFE 2019 conference, in Newcastle, UK. Concretely, a list of unlabeled datasets are made available online. Some are time series of real living systems, and some are generated by non-living systems. Participants can test their algorithms on these public datasets, but their algorithms will be judged based on unreleased datasets. Along with a working algorithm, each participant is required to submit a description of the underlying theoretical measure (i.e. "What is your theory about to correctly classify these datasets? What is your idea based on?").
We intend this competition as a critical initiative to help Artificial Life research, and drive it in a practical way, just like computational linguistics for example made tremendous progress in the past few decades by setting up well-defined tasks as shared open challenges. Our goal is to contribute to scientific progress in the field. Although the impact on industry is not the main purpose, one may imagine numerous applications in terms of detection of living systems. The contributions from participants will hopefully help progress in refining our mathematical understanding of the nature of living systems, i.e. what life is in contrast to inert matter. This, in turn, is connected to problems of life detection on exoplanets, the detection of extremophiles on Earth, and the origin of life, and the study of synthetic living systems in general. We consider this as a very important yet not much funded research area, relating to fundamental questions about the status and uniqueness of human life and intelligence in the universe.
Concretely, the participants need to provide two major elements: a code in Python that returns 1 if the data is from a real living system, and 0 if it's not, and a very short PDF description of what's the mathematical measure under the hood. The use of learning algorithms is expressly not allowed, for they would not yield an explanation of why they work. In case of any doubt, please contact the organizers at flrc.contact[at]gmail.com.
The datasets consist in time series of trajectory positions in 2D (those are meant as actual spatial trajectories), which correspond to either artificial or real living systems. Each dataset comes in the form of a 20Mb or so CSV file with 3 columns labelled “t“, “x” and “y”, with 10,000 rows.
The competition is volunteer-based and open to all. Head to https://competitions.codalab.org/competitions/20612!