In the world of artificial intelligence and robotics there is a paradox that has baffled scientists and experts for decades. It is known as the “Moravec Paradox”, and it poses the following question:
Why can machines perform extremely complex tasks, such as calculating advanced mathematical equations or playing master-level chess, with relative ease, while struggling in everyday situations that humans consider simple?
For example, industrial robots are amazing at repetitive, precise tasks, such as assembling products on an assembly line. However, A simple robot that walks on two legs and picks up objects from the ground is a considerable engineering feat.
To understand this paradox and its implications, it is time, with the help of an artificial intelligence expert, to at least try to provide an answer.
Getty
So why does this apparent paradox occur?
The answer lies in the evolutionary history of intelligence, as noted by Hans Moravec, one of the pioneers in the field of robotics and artificial intelligence. He suggested that these two modern fields They were inspired by the functioning of the human brain, but did not take into account the nature of evolution.
Moravec argued that skills considered “intelligent” today, such as abstract reasoning and complex problem solving, are relatively recent in evolutionary terms. These abilities developed in the newest part of the brain, the neocortex, which is particularly large in humans compared to other animals.
On the other hand, skills that are apparently simple, such as visual recognition or locomotion, are rooted in the oldest and most primitive structures of the human being. These abilities are the product of millions of years of evolution and are located in regions such as the limbic system and the cerebellum.
“You could say that while mathematical tasks offer a structured and well-defined playing field, visual perception is a minefield of ambiguities and complexities,” explains Josué Pérez Suay, specialist in Artificial Intelligence and ChatGPT, in an interview for Computer Hoy. .
An inverted pyramid explains how machines work
To better understand this hierarchy of intelligence, imagine an inverted pyramid. At the top of the pyramid are high-level skills, such as solving complex mathematical equations and strategy in chess games. These are the skills that modern computers and robots can master.
At the center of the pyramid are intermediate skills, such as natural language and planning. These skills are complicated for machines, but artificial intelligence has made enormous strides in these areas.such as with the arrival of ChatGPT.
At the base of the pyramid are the most primitive skills, such as visual recognition, locomotion, and tactile perception. These are the skills that humans develop naturally in their childhood, but are often difficult to teach machines.
“Humans can easily pick up on tone and context in a conversation, allowing them to understand irony or sarcasm. For AI, this remains a considerable challenge due to the lack of contextual and emotional cues. While a human can easily recognize a chair, regardless of its design or the position it is in, the AI needs to be trained with numerous examples to achieve similar precision,” comments the expert.
“Humans often make decisions based on intuition or “gut feelings” due to a multitude of chemical, emotional and even faith reactions, something that is difficult to program in an AI system that depends on data and algorithms,” he adds.
Addressing this paradox: advances in computer vision and nature-inspired robotics
As artificial intelligence and robotics advance, it is essential to address this paradox so that machines can function more efficiently in the real world. This may require an interdisciplinary approach that combines artificial intelligence, neuroscience and engineering.
In the case of computer vision, researchers are developing deep learning algorithms that allow machines to learn to recognize objects in various situations and postures. This is more like the human visual learning process.
“Deep learning and convolutional neural networks have shown great potential in image and pattern recognition. These models can be trained to better understand the context and subtleties that humans intuitively grasp,” explains Josué Pérez Suay.
“Combining data from different types of sensors (visual, auditory, tactile, etc.) could help AI better understand the context in which it operates, similar to how humans use multiple senses to interpret their environment,” he adds.
In robotics, biomimicry plays a really important role. By designing robots that mimic the anatomy and behavior of living organisms, such as insects or mammals, great advances can be made in locomotion and manipulation in the environment.
Additionally, develop more advanced sensors that give machines more detailed and contextual perception of their environment or implement reinforcement learning techniques, where machines can learn through experimentation and feedback, similar to how humans acquire skills. motor and cognitive, could be key.
As you see, The solution to Moravec’s Paradox involves a combination of technological advances, scientific research, and an interdisciplinary approach to better understand intelligence and cognition. Until then, this will remain a huge challenge for scientists and a reminder of the great complexity of the human mind and its nature.
“The growth in the use of AI has been exponential and investments are being greater, so it is to be expected that in these next 5 years the leap will be more than we can currently imagine. Work is being done on machine learning, neuroscience, psychology and ethics so that AI not only has more context, but in the end can make a decision more consistent with the human factor, although this will also cause errors when created by people,” concludes the expert.
