Computers have come a long way in terms of artificial intelligence, but will they ever reach or even surpass human-level intelligence? This question has been the subject of much debate among experts in the field. The Association for the Advancement of Artificial Intelligence (AAAI) conducted a survey earlier this year, asking its members whether neural networks, the current stars of AI systems, would be enough to achieve this goal. The majority of respondents said no, indicating that a combination of neural networks and symbolic AI would be necessary.
Symbolic AI, also known as ‘good old-fashioned AI’, is based on formal rules and logical relationships between concepts. It involves mathematical principles, ‘if-then’ statements, and coding languages like Python. While symbolic AI was an early frontrunner in the field of AI, it was overtaken by more flexible neural networks in the early 2010s. Neural networks excel at learning from vast amounts of data and power technologies like large language models (LLMs) and chatbots.
The concept of ‘neurosymbolic AI’ has gained traction in recent years as a way to combine the strengths of both neural networks and symbolic AI. This approach aims to create artificial general intelligence (AGI) that can reason and generalize knowledge like humans. Neurosymbolic AI could be particularly useful for high-risk applications such as military or medical decision-making due to its transparency and understandability.
Researchers are exploring various ways to integrate neural networks and symbolic AI. Some approaches involve using symbolic techniques to enhance neural networks, while others use neural networks to improve symbolic algorithms. Projects like AlphaGeometry and logic tensor networks showcase the potential of these hybrid systems in solving complex problems efficiently.
However, challenges remain in effectively combining neural networks and symbolic AI. Ensuring logical reasoning and generalization in AI systems requires a deep understanding of how these two paradigms can work together seamlessly. Research on AI metacognition, or how AI monitors and conducts its own thinking, is crucial for developing sophisticated integration strategies.
While neurosymbolic AI shows promise, further research and development are needed to unlock its full potential. New hardware and chip architectures may be required to run these systems efficiently. Additionally, emerging fields like quantum AI could offer new possibilities for enhancing AI capabilities.
Ultimately, the goal is to create AI systems that can learn and reason beyond human capabilities, inventing new symbolic representations and algorithms. By studying how computers can teach humans, rather than the other way around, we may uncover groundbreaking discoveries and push the boundaries of artificial intelligence.
This article, originally published on November 25, 2025, highlights the ongoing evolution of AI systems and the potential for neurosymbolic AI to revolutionize the field.
