The Future of AI: Combining Symbol Manipulation and Neural Networks


Hatched by Glasp

Sep 21, 2023

4 min read


The Future of AI: Combining Symbol Manipulation and Neural Networks


Artificial intelligence has come a long way since its inception, with deep learning models like neural networks dominating the field in recent years. However, as we delve deeper into the potential of AI, it has become increasingly clear that deep learning alone is not enough to achieve human-like intelligence. In this article, we will explore the debate between symbol manipulation and neural networks as the foundation for AI systems and propose the integration of both approaches through hybrid models. By understanding the origins of symbol manipulation and its potential for learning, we can unlock the true power of AI and move closer to safe, trustworthy, and interpretable artificial intelligence.

The Debate: Symbol Manipulation vs. Neural Networks:

For nearly seven decades, the fundamental debate in AI has revolved around whether AI systems should be built on symbol manipulation or neural networks. Symbol manipulation, rooted in logic, mathematics, and computer science, treats thinking as an algebraic process. On the other hand, neural networks mimic the structure and function of the human brain, processing numeric inputs through interconnected nodes. Both approaches have their merits, but they are not mutually exclusive. The integration of symbols and neural networks, through hybrid models, offers a promising way forward.

The Quest for Hybrid Models:

Early pioneers in AI, such as Marvin Minsky and John McCarthy, favored symbol manipulation as the only reasonable path to achieve AI. Meanwhile, Frank Rosenblatt, a neural network pioneer, argued for the power of statistics and numeric processing in AI systems. However, it has long been recognized that these two approaches can coexist and complement each other. Hybrid models seek to combine the data-driven learning of neural networks with the powerful abstraction capabilities of symbol manipulation. By finding the right balance between the two, we can unlock the full potential of AI.

Learning Symbol Manipulation:

A crucial question arises: Can symbol manipulation be learned rather than built into AI systems from the start? The answer is a resounding yes. While symbol manipulation is often considered innate, it is possible that an indirect basis enables the acquisition of this skill. By focusing our efforts on discovering this indirect basis, we can develop systems that leverage the world's knowledge effectively. Learning symbolic abstractions is the key to building AI that is safe, trustworthy, and interpretable.

The Challenges of Deep Learning:

Deep learning has undoubtedly made significant strides in AI, but it faces specific challenges, primarily related to compositionality, systematicity, and language understanding. These challenges center around generalization and distribution shift, where neural networks struggle when faced with novel situations. While deep learning excels in pattern recognition, it falls short in areas such as natural language understanding and reasoning. Current systems still struggle to reliably extract symbolic operations, highlighting the need for a more holistic approach.

The Power of Hybrid Models:

Combining the strengths of deep learning and symbol manipulation can revolutionize AI. By incorporating symbolic representations into data-driven learning, we can enhance the efficiency and effectiveness of AI systems. Human infants and toddlers demonstrate the ability to generalize complex aspects of natural language and reasoning, suggesting the presence of symbolic thinking before formal education. By integrating a little built-in symbolism into AI systems, we can improve learning efficiency and address the shortcomings of statistical correlations.

Actionable Advice:

  • 1. Focus on Research: Emphasize the discovery of the indirect basis that enables the acquisition of symbol manipulation. This research can lead to breakthroughs in building hybrid models that leverage the power of both symbols and neural networks.
  • 2. Foster Collaboration: Encourage collaboration between symbol manipulation and deep learning researchers. By sharing insights and expertise, we can accelerate progress in developing hybrid models and advancing the field of AI.
  • 3. Invest in Education: Promote the integration of symbolic thinking and abstract reasoning in AI education and training. By nurturing these skills in AI systems, we can enhance their ability to handle complex tasks and improve their interpretability.


As we look to the future of AI, it is evident that deep learning alone is insufficient to achieve human-like intelligence. The integration of symbol manipulation and neural networks through hybrid models offers a promising path forward. By uncovering the indirect basis of symbol manipulation and combining it with data-driven learning, we can create AI systems that are safe, trustworthy, and interpretable. The quest for hybrid models is not about replacing one approach with another but about harnessing the strengths of both to unlock the true potential of AI.

Hatch New Ideas with Glasp AI 🐣

Glasp AI allows you to hatch new ideas based on your curated content. Let's curate and create with Glasp AI :)