Discussion Groups
DG A: Open Endedness with Automata Chemistries
| Facilitator: Susan Stepney | Time: Monday 13:00 - 14:30 | Room: 4A05 |
I will discuss how we can use automata chemistries in ALife research to generate computational systems that exhibit Open-Ended behaviour. First, I will describe the framework of models, metamodels, and metametamodels (it stops there!), and use it to give a precise definition of three levels of novelty: variation, innovation, and transformation. I will explain why most agent-based evolutionary systems can exhibit only variation (level 0 novelty). Second, I will define what automata chemistries are, and give some examples (Tierra, Avida, Physys, nanopond, Stringmol). I will explain how these systems exhibit open-ended innovation (level 1 novelty). Third (the research challenge) I will discuss what is needed to exhibit transformation (level 2 novelty), and how automata chemistries might be extended to implement this.
DG B: Emergence of hierarchical intelligence in simulated agents
| Facilitator: Eric Medvet | Time: Monday 13:00 - 14:30 | Room: 5A09 |
Complex tasks require the agent to act on two levels simultaneously: locally, to perform low-level behaviors and manage close environmental events; and globally, to strategize and achieve the task goal. These local and global may exist both in time or space, shaping a hierarchy in the agent intelligence. For example, while navigating an arena, a differential-drive robot might first plan a sketched trajectory (global in time) and then reactively move avoiding obstacles while attempting to follow the planned trajectory (local in time). A modular robot may govern its behavior by first dictating sketches of behavior to subsets of the modules (global in space) and then actuating the single modules based on their role in the body (local in space). In this project, we want to investigate the factors enabling and favoring the emergence of hierarchical (embodied) intelligence in simple simulated agents subjected to optimization: what are the brain/body structures which mostly favor this emergence? What are the best objectives for driving the optimization? How to detect/monitor the emergence of an hierarchy?
DG C: Why do living organisms exist?
| Facilitator: Stefano Nichele | Time: Monday 13:00 - 14:30 | Room: 3A07 |
Why do living organisms exist? This question was addressed in 1953 by Nils Barricelli, one of the founding fathers of artificial life, using numeric symbioorganisms (one dimensional cellular automata). In his CA models, reproduction and mutation were not sufficient to explain the origin of an evolutionary process (and therefore the origins of life). He proposed that the missing ingredient was symbiogenesis, the creation of a new entity out of a mutually beneficial relationship between two pre-existing entities. Barricelli’s work has not been fully appreciated, however it is still very relevant today. In this discussion group we will review Barricelli’s ideas on symbiogenesis, recent ideas, and identify future directions and open questions.
DG D: Environmental Buffering and the Constraint-Function Loop
| Facilitator: Alyssa Adams | Time: Monday 13:00 - 14:30 | Room: 4A09 |
This topic explores how agents restructure theit environment to reduce the need for internal information processing. While standard control theory treats the environment as unmodifiable, the idea of “environmental buffering” suggests that agents “offload” information processing to their surroundings by directly modifying their environment. By turning a complex environment into a set of stable physical constraints, agents shift the energy burden from internal information processing to the construction of external stability.
Some possible questions to start with:
- Does shaping an environment make agents more autonomous by freeing up resources, or less autonomous through external dependence?
- What if the upfront buffering cost is too high?
- If an environment is changing too much, constantly eroding any constructed stability, is it more efficient to evolve better sensors than to try to buffer?
- If an agent’s function depends on the constraints it created, where does the agent end and the environment begin?
DG E: Biocomputation: Programming bacteria to perform computations
| Facilitator: Ángel Goñi-Moreno | Time: Tuesday 10:30 - 12:00 | Room: 4A05 |
The design and implementation of genetic circuits in living cells is a major activity within biocomputation. Genetic components are arranged generating pre-defined regulatory cascades where some inputs are turned into outputs according to predefined instructions. This allows us to build biological computations, for instance, in the form of Boolean functions where the combination of inputs turn ON or OFF a specific reporter protein. Although this conceptual mapping has been very successful, there are stability and robustness issues that cannot be overlooked. I argue that the performance of biological circuits in living cells cannot be reduced to Boolean functions. Any other model that is more powerful than Boolean logic? What computing primitives within the cell could we use? Of course, by using DNA alone we cannot go that far… but this is a very reductionist approach. Let’s embrace complexity. Let’s use biological noise, the context, evolution… all of them. Any computation more complex than combinatorial logic now?
DG F: How can we use ALICE concepts for creative purposes?
| Facilitator: Kyrre Glette | Time: Tuesday 10:30 - 12:00 | Room: 5A09 |
In this group we will discuss the advantages and challenges of using ALIFE systems for artistic and creative processes, and their role in the era of deep learning-based generative AI. We will discuss how to interact with such systems and how to map the processes to artistic output. We will also cover evolutionary methods as tools for exploration and creativity. Quality-diversity, open-ended evolution, stepping stones, innovation engines, and interactive evolution are interesting topics in this context.
DG G: TBA
| Facilitator: Alexander Mordvintsev | Time: Tuesday 10:30 - 12:00 | Room: 3A07 |
