From Text to Life: On the Reciprocal Relationship between Artificial Life and Large Language Models

TL;DR

The paper investigates the reciprocal relationship between Artificial Life (ALife) and Large Language Models (LLMs), arguing that insights from ALife can inform the design and training of more adaptive LLMs, while LLMs offer powerful tools for ALife research. It surveys how LLMs can function as intelligent mutation and crossover operators, environment generators, exploration aids, models of human behavior, and scientific collaborators, and it details how ALife concepts—such as internal states, autonomy, self-replication, self-organization, emergence, regulatory mechanisms, 4E cognition, collective intelligence, evolution, and cultural evolution—can shape LLM development. The authors propose a framework of five LLM-for-ALife threads and a parallel ALife-for-LLMs perspective to guide cross-disciplinary work, emphasizing open-endedness, evolvability, and embodied cognition. The work highlights potential benefits for both fields, including more adaptive and capable LLM agents and deeper lifelike understanding, while acknowledging challenges related to data quality, energy consumption, and ethical implications that require ongoing dialogue between communities.

Abstract

Large Language Models (LLMs) have taken the field of AI by storm, but their adoption in the field of Artificial Life (ALife) has been, so far, relatively reserved. In this work we investigate the potential synergies between LLMs and ALife, drawing on a large body of research in the two fields. We explore the potential of LLMs as tools for ALife research, for example, as operators for evolutionary computation or the generation of open-ended environments. Reciprocally, principles of ALife, such as self-organization, collective intelligence and evolvability can provide an opportunity for shaping the development and functionalities of LLMs, leading to more adaptive and responsive models. By investigating this dynamic interplay, the paper aims to inspire innovative crossover approaches for both ALife and LLM research. Along the way, we examine the extent to which LLMs appear to increasingly exhibit properties such as emergence or collective intelligence, expanding beyond their original goal of generating text, and potentially redefining our perception of lifelike intelligence in artificial systems.

Figures (3)

• Figure 1: Number of papers published at NeurIPS that incorporate different lifelike properties.
• Figure 2: Selection of papers on LLMs where some ALife-related concepts are incorporated as key components.
• Figure 3: Can we view LLMs as ALife systems? (from left to right) (a) Self-rewarding LLM agents are able to assess their experiences and fine-tune themselves on the best samples yuan2024self. Forms of intrinsic motivation, such as surprise, are employed in robotic swarms hamann2014evolution. (b) Under Reflexion, an LLM agent evaluates its own experiences and consolidates them in a long-term memory shinn2023reflexion. In ALife, Neural Cellular Automata act as a model of morphogenesis that displays robustness to perturbations sudhakaran2021growing. (c) When granted web-access, LLM agents can learn how to use APIs, thus acquiring access to the digital substrate of human society schick2024toolformer. Traditionally, embodiment in ALife refers to the ability of digital organisms, such as Xenobots blackiston2021cellular, to acquire a biological substrate. (d) The factuality and reasoning abilities of LLMs seem to improve when they debate with other LLMs du2023improving. In ALife, models such as Boids exemplify the emergence of complex patterns in a collective where individuals follow simple rules reynolds1987flocks. (e) Model-merging is a technique for crossing over multiple LLMs into a single one model with improved performance which can be incorporated within an evolutionary loop akiba_evolutionary_2024, a dominant paradigm in ALife sims1994evolving (f) By participating in the generation, transmission and selection of cultural artefacts (e.g. texts, images), LLMs and other Generative AI techniques are participating in human cultural evolution brinkmann_machine_2023. Artificial creativity and human-AI collaboration in digital evolution have already engaged ALife secretan_picbreeder_2011lehman_surprising_2020 (g) By viewing an LLM as the composite of its training data and its implementation code, we can envision a future where it replicates, akin to ALife systems such as Tierra Ray1992EvolutionE.

Theorems & Definitions (8)

• Definition 1
• Definition 2
• Definition 3
• Definition 4
• Definition 5
• Definition 6
• Definition 7
• Definition 8