Introduction
Artificial intelligence (AI) systems are everywhere, revolutionizing various sectors. Yet, despite their apparent successes, a crucial question remains: why do AI systems struggle to learn autonomously? This article explores this issue by drawing lessons from cognitive sciences.
The Limitations of Current AI Learning
Today, approximately 90% of AI projects rely on some form of human supervision. This starkly contrasts with the autonomous learning observed in nature, particularly in humans and animals. But why this dependency? Current AI systems, although effective for specific tasks, lack the cognitive flexibility possessed by living organisms.
Systems A and B
According to researchers Emmanuel Dupoux, Yann LeCun, and Jitendra Malik, the current AI architecture could benefit from integrating two learning systems:
- System A: Learning from observation, similar to how children observe and mimic adults.
- System B: Learning from active behavior, where the entity explores and learns on its own.
These systems need to switch fluidly through metacognitive control (System M).
Inspiration from Cognitive Sciences
The evolution and cognitive development of living organisms offer an intriguing model for AI. For instance, human children learn by exploring their environment and interacting with it actively and passively. This approach could inspire more adaptive and resilient AI models.
Concrete Examples
- DeepMind has embarked on projects of reinforcement learning without human intervention, exploring the potential of AI capable of autonomous adaptation.
- OpenAI, with its GPT model, demonstrates how AI can learn from vast datasets without explicit labeling, though still far from complete autonomy.
Toward Artificial Consciousness?
A fascinating question is whether this evolution toward autonomous systems could lead to a form of artificial consciousness. While this remains a long-term prediction, the idea of AI capable of understanding and learning like a human is both exciting and perplexing.
The Crucial Role of Interdisciplinarity
To achieve this goal, increased collaboration between neuroscience, cognitive psychology, and AI engineering is essential. These disciplines can provide valuable insights into designing AI models that not only learn but evolve with their environment.
Conclusion
In conclusion, for AI systems to achieve true autonomous learning, they must integrate architectures inspired by cognitive sciences. This approach requires theoretical and practical innovations, as well as sustained interdisciplinary collaboration.
Want to automate your operations with AI? Book a 15-min call to discuss.