The field of AI is constantly evolving, with companies pushing for better hardware and experimenting with existing system architectures to meet the demands of this compute-hungry beast. In this pursuit, Nikhil Malhotra, the global head of Makers Lab at Tech Mahindra, has proposed the concept of ‘Dream AI’. This architecture, called Deep Reinforced Engagement Model AI, combines symbolic AI with deep reinforcement learning, addressing the limitations of current AI models.
Malhotra highlights three key problems with current AI systems – they are limited to narrow domains, lack reasoning abilities, and struggle with context. Dream AI aims to overcome these challenges by incorporating symbolic reasoning and neural networks, creating a dual-loop system where agents can simulate and act based on physical and logical rules. This approach allows AI to ‘dream’ by simulating environments and learning with a nuanced understanding of context, rather than relying solely on vast datasets.
The role of reinforcement learning (RL) in Dream AI is crucial, as it enables AI to contemplate its own capabilities and understand its existence more profoundly. Malhotra’s research goal is to make AI less compute-intensive, shifting away from traditional reward models and empowering AI to ‘dream’ about its own potential. This ‘min-max regret model’ allows AI to develop its own questions and aspirations, similar to how humans subconsciously store information and use it as an expert.
In conclusion, Dream AI offers a promising solution to the limitations of current AI systems, by combining symbolic reasoning and deep reinforcement learning. This approach not only improves AI’s performance but also allows it to ‘dream’ and contemplate its own existence, bringing it closer to human cognition.
