Reflections by Ramon López de Mántaras on the Current AI Boom

In early October, Ramon López de Mántaras was honoured with a lifetime achievement award for his scientific contributions to Artificial Intelligence at the AI & Big Data Congress in Barcelona 2024. Upon receiving the award, Ramón López de Mántaras, Emeritus Professor at the CSIC’s Artificial Intelligence Research Institute, emphasized his balanced stance, steering clear of both techno-pessimism and techno-optimism. In his speech, he presented a fascinating reflection on the AI boom through a metaphor that I would like to share in this article.

The AI Metaphor Featuring Isaac Newton

López de Mántaras used a captivating metaphor to illustrate the complexity and perception of advanced technology. He imagined Isaac Newton being transported to the present and confronted with a modern mobile phone. Despite being one of the brightest minds in history, who invented calculus, explained optics and gravity, and formulated the laws of motion that revolutionized physics, Newton would see this device as “magic.” This analogy highlights how Artificial Intelligence, with its seemingly impressive results, can appear magical to many.

In this context, he recalled Arthur C. Clarke’s famous quote: “Any sufficiently advanced technology is indistinguishable from magic” underscoring that Artificial Intelligence is a clear example of this technological magic.

What is the Real Situation of Artificial Intelligence?

Artificial Intelligence (AI) has been stalled for over 50 years on one of its most fundamental problems: endowing machines with knowledge and common sense. These non-specialized knowledge, which humans acquire through our experiences, are essential for a deep understanding of language. Although large language models can display a certain superficial level of understanding, there are many reasons to doubt that they possess deep comprehension and reasoning capabilities.

Subbarao Kambhampati, former president of the Association for the Advancement of Artificial Intelligence, states that large language models perform “approximate retrievals of text patterns contained in the model’s data patterns.” A recent study showed that these models cannot detect illegal moves in chess more than 50-60% of the time, something even a 5-year-old child could do. This highlights the data contamination in published AI benchmarks and its effects on model evaluation.

AI leaders like Demis Hassabis and Yann LeCun have acknowledged that large language models alone are not the path to general Artificial Intelligence. The way forward involves acquiring a model of the world, not just a language model, and equipping AI with a multisensory body to interact with the world. This is crucial if we want to achieve general AI someday. Some tech giants have already started projects in this direction, such as DeepMind’s PLATO system, which learns basic physical concepts directly from visual information and interacting with the world.

Inspired by Jean Piaget’s developmental psychology, Ramón López de Mántaras has been researching learning in cause-effect relationships with a humanoid robot for the past nine years. He believes this is an essential step to learning the model of the world and beginning to understand what happens with physical objects, as well as on a social level. This approach could be the key to endowing AI with the deep understanding and common sense necessary to achieve true general Artificial Intelligence.

Conclusion

The metaphor and subsequent reflection by Ramon López de Mántaras remind us that despite the impressive advances in Artificial Intelligence, there are still fundamental challenges to be resolved. Deep understanding and common sense are essential elements that machines have yet to achieve. The path to general AI involves not only improving language models but also equipping machines with a multisensory body that allows them to interact with the real world meaningfully. Only then can we advance towards a general Artificial Intelligence that truly understands and reasons like humans.