Developmental AI is a bootstrapping approach where embodied AIs develop abilities along a bio-inspired trajectory. They start with innate competences and learn more by interacting with the world including people. Developmental AIs have been demonstrated, but their abilities are like those of pre-toddler children.
In contrast, mainstream approaches have led to impressive feats and commercially valuable AI systems. The approaches include deep learning and generative AI (e.g., large language models) and manually constructed symbolic modeling. However, manually constructed AIs tend to be brittle even in circumscribed domains. Generative AIs are helpful on average, but they can make strange mistakes and not notice them. Not learning from experience, they sometimes lack common sense and social alignment.
The first paper below frames an opportunity for a leap in capabilities and widespread adoption of AI and robotics technology.
The “Bootstrapping” paper lays out prospects, gaps, and challenges for developmental AI. The goal is to create data-rich experientially based foundation models for human-compatible AIs. A virtuous multidisciplinary research cycle has led to developmental AIs with capabilities for multimodal perception, object recognition, and manipulation. Computational models for hierarchical planning, abstraction discovery, curiosity, and language acquisition exist but need to be adapted to an embodied learning approach. The competence gaps involve nonverbal communication, speech, reading, and writing.
Aspirationally, developmental AIs would learn, share what they learn, and collaborate to achieve high standards. They would learn to communicate, establish common ground, read critically, consider the provenance of information, test hypotheses, and collaborate. The approach would make the creation of AIs more democratic, enabling more people to train, test, build on, and replicate AIs
The “Roots” paper reviews our growing understanding of human-AI teaming broadly and in AI research. It describes the competence requirements for creating AI collaborators and .
Publications
Stefik, M. (2024) What AIs are not Learning (and Why) (2024) arXiv coming soon
Stefik, M., Price, R. (2023) Bootstrapping Developmental AIs. arXiv http://arxiv.org/abs/2308.04586
Stefik, M., (2023) Roots and Requirements for Collaborative AI. arXiv http://arxiv.org/abs/2303.12040