Is symbolic AI more relevant than ever? (www.heise.de)

The discussion surrounding artificial intelligence (AI) has increasingly centered on large language models and data-driven techniques, often overshadowing the relevance of symbolic AI, which has a rich historical background spanning from the 1956 Dartmouth Conference to the rise of expert systems in the 1980s. The article raises critical questions about whether the current reliance on neural networks might be leading the AI community into a local maximum, potentially overlooking fundamental shortcomings inherent in data-driven approaches, such as compositionality, generalization, and the so-called Symbol Grounding Problem, where AI struggles to attach real-world meaning to symbols it processes. The author advocates for a renewed exploration of symbolic and neuro-symbolic AI, proposing a thought experiment for an AI being built on hardwired needs rather than amassed data. This being would integrate neural perception with symbolic decision-making, experiencing the world through its senses and developing knowledge organically. The piece suggests that the future of AI may not hinge solely on scaling up models, but rather on innovating structures that allow for meaningful learning and interaction, potentially marking a shift in the ongoing AI research landscape towards this third wave of neuro-symbolic AI. This could lead to more capable systems that fulfill human-like understanding and reasoning.