The idea of representation and reasoning is crucial in the development of trustworthy and explainable AI, as sometimes there isn't enough data to learn from scratch, and we want to understand where these systems work and where they don't.
The host and guest discuss the evolution of AI methods, particularly the success of deep learning for a specific set of problems, and speculate on its place as a chapter in the bigger view of AI in future decades.
The idea of representation and reasoning is crucial for AI, but we need to do a better job with one-shot learning and programming for representation, according to an AI expert.
The use of neural networks and learning from data can lead to brilliant success and frustrating failures, with limited introspection into how the network operates. While better quality and organized data can improve the issue, the atomic nature of symbols in AI can still present a challenge.
The need to trust and understand AI systems before using them is critical, and having a conversation to understand their decision-making process is preferable to simply receiving an explanation.