AI-GAs: AI-generating algorithms, an alternate paradigm for producing general artificial intelligence

By Jeff Clune. 2019

154 citations recorded by the source corpus.

Perhaps the most ambitious scientific quest in human history is the creation of general artificial intelligence, which roughly means AI that is as smart or smarter than humans. The dominant approach in the machine learning community is to attempt to discover each of the pieces required for intelligence, with the implicit assumption that some future group will complete the Herculean task of figuring out how to combine all of those pieces into a complex thinking machine. I call this the ``manual AI approach.'' This paper describes another exciting path that ultimately may be more successful at producing general AI. It is based on the clear trend in machine learning that hand-designed solutions eventually are replaced by more effective, learned solutions. The idea is to create an AI-generating algorithm (AI-GA), which automatically learns how to produce general AI. Three Pillars are essential for the approach: (1) meta-learning architectures, (2) meta-learning the learning algorithms themselves, and (3) generating effective learning environments. I argue that either approach could produce general AI first, and both are scientifically worthwhile irrespective of which is the fastest path. Because both are promising, yet the ML community is currently committed to the manual approach, I argue that our community should increase its research investment in the AI-GA approach. To encourage such research, I describe promising work in each of the Three Pillars. I also discuss AI-GA-specific safety and ethical considerations. Because it it may be the fastest path to general AI and because it is inherently scientifically interesting to understand the conditions in which a simple algorithm can produce general AI (as happened on Earth where Darwinian evolution produced human intelligence), I argue that the pursuit of AI-GAs should be considered a new grand challenge of computer science research.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Learning to Continually Learn (2020, methodology): “That is, to solve a challenging problem they try to identify the required building blocks and how to combine them, and then let machine learning tune parameters within this complex assembly [5].”
Inductive biases for deep learning of higher-level cognition (2020, background): “In fact we could have more than two time scales (Clune, 2019): think about the outer loop of evolution, the slightly faster loop of cultural learning (Bengio, 2014) which is somewhat stable across generations, the faster learning of individual humans, the even faster learning of specific tasks and n”
Open-Ended Learning Leads to Generally Capable Agents (2021, methodology): “Finally, the use of the world-agent co-evolution process to create the base distribution for training and evaluation for the remainder of our learning process is an example of AI-generating algorithms (Clune, 2019).”
MetaFSCIL: A Meta-Learning Approach for Few-Shot Class Incremental Learning (2022, background): “In addition, the aforesaid approaches involve hand-engineered heuristics (e.g. saving exemplars [3, 7], decoupled learning [32], prototype refinement [33]) which are defective compared to the learned solutions [6].”
Meta-Learning with Sparse Experience Replay for Lifelong Language Learning (2020, background): “A recent trend in continual learning, as well as machine learning in general, is to directly learn generalizable solutions via meta-learning (Clune, 2019).”
Multi-task learning for natural language processing in the 2020s: where are we going? (2020, background): “A long-term goal that will continue to persist is to develop general experts which can compete with their single-task counterparts (Clune, 2019).”
Unity: A General Platform for Intelligent Agents (2018, background): “It has been argued in recent work that generating eﬀective and diverse learning environments (often as a co-evolutionary process involving agent and environment) is a critical component for developing artiﬁcial general intelligence (Wang et al., 2019; Clune, 2019).”
Improving Generalization in Meta Reinforcement Learning using Learned Objectives (2019, methodology): “Several works have proposed an alternative framework based on meta reinforcement learning (Schmidhuber, 1994; Wang et al., 2016; Duan et al., 2016; Finn et al., 2017; Houthooft et al., 2018; Clune, 2019).”
Fast and Slow Learning of Recurrent Independent Mechanisms (2021, background): “…rapidly with a few training examples, and has demonstrated success in both supervised learning such as few shot image classiﬁcation (Ravi & Larochelle, 2016) and reinforcement learning (Wang et al., 2016; Santoro et al., 2016; Duan et al., 2016; Botvinick et al., 2019; Clune, 2019) settings.”
AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors in Agents (2023, unknown): “These advancements edge us closer to realizing the concept of artificial general intelligence (AGI) (Goertzel & Pennachin, 2007; Clune, 2019) that can generalize across a broader range of tasks.”
Discovering Reinforcement Learning Algorithms (2020, methodology): “The goal of this work is similar to the second pillar (meta-learning learning algorithms) in AI-GAs [7].”
Genie: Generative Interactive Environments (2024, background): “…from unseen action-free videos of simulated reinforcement learning (RL) environments, indicating that Genie may hold the key to unlocking unlimited data for training the next generation of generalist agents (Bauer et al., 2023; Clune, 2019; Open Ended Learning Team et al., 2021; Reed et al., 2022).”
Fiber: A Platform for Efficient Development and Distributed Training for Reinforcement Learning and Population-Based Methods (2020, unknown): “We expect it will further enable progress in solving hard RL problems with RL algorithms and population-based methods by making it easier to develop these methods and train them at the scales necessary to truly see them shine (Clune 2019).”
Meta-learning curiosity algorithms (2020, background): “Related work on meta-learning (Schmidhuber, 1987; Thrun & Pratt, 1998; Clune, 2019) and efforts to increase its generalization can be found in appendix B.”
Enhanced POET: Open-Ended Reinforcement Learning through Unbounded Invention of Learning Challenges and their Solutions (2020, background): “…environments for meta-learning algorithms (Duan et al., 2016; Finn et al., 2017; Wang et al., 2016), and eventually act as a stepping stone towards AI-generating algorithms (AI-GAs) (Clune, 2019) that could one day bootstrap themselves from simple initial conditions to powerful cognitive machines.”