Will humanoid robots take over blue collar jobs?

Key Concepts

• Humanoid Robots: Robots designed to resemble the human form.
• Large Language Models (LLMs): AI models trained on massive text datasets, like ChatGPT, capable of generating human-like text.
• High-Dimensionality: The complexity of robotic control requiring simultaneous coordination of numerous joints, contrasting with the one-dimensional nature of language processing.
• Data Scarcity in Robotics: The limited availability of comprehensive datasets for training robots compared to the vast textual data available for LLMs.
• Dexterity: The skill and ease in using one’s hands or body, a key human capability difficult to replicate in robots.

The Gap Between AI and Robotics: A Discussion with Ken Goldberg

This conversation with Ken Goldberg, Professor of Engineering and President of the Robot Learning Foundation at UC Berkeley, centers on the significant disparity between the advancements in Large Language Models (LLMs) and the progress in robotics, particularly humanoid robots. Goldberg argues that despite the impressive capabilities of AI like ChatGPT, replicating even simple human tasks with robots is a far more complex and distant goal.

Defining Humanoid Robots & Their Appeal

The discussion begins by establishing the definition of a humanoid robot – a robot designed to physically resemble a human being. Goldberg explains that the current interest in humanoid robots stems from a perceived need for workers in jobs humans are unwilling to do, with the expectation that a human-like form would allow robots to seamlessly integrate into existing human workplaces. He notes the historical fascination with creating mechanical beings, dating back to ancient Egypt.

The Fundamental Difference: Dimensionality & Data

A core argument presented by Goldberg is the fundamental difference in the learning processes for language models versus robots. He highlights that language learning is essentially “one-dimensional,” dealing with a sequence of characters or words. In contrast, robotics operates in a “much higher dimensional” space, requiring the precise and synchronized control of potentially 50 or more joints in a humanoid robot.

This difference is compounded by a critical issue: data availability. Goldberg estimates that approximately 100,000 years of textual data exists on the internet for training language models – the equivalent of the time it would take a person to read all online text. However, “we have no data for robotics,” meaning the field is starting from scratch in collecting the necessary data for effective robot training.

Progress & Timelines: Assembling, Driving, and Beyond

Goldberg acknowledges ongoing research into robot assembly and expresses cautious optimism. However, he emphasizes that significant breakthroughs are not imminent, predicting that achieving practical robotic capabilities may take “not within a year, within 5 years, [and] might not even happen within 10 years.”

He contrasts this with the progress in self-driving cars, explaining that driving primarily involves avoiding obstacles, a comparatively simpler task than the dexterity required for tasks like arranging flowers. He notes that self-driving car development has benefited from 10-15 years of data collection and is becoming increasingly prevalent in cities across the US and China. He also acknowledges recent algorithmic adjustments making self-driving cars more “aggressive,” raising safety concerns for pedestrians and cyclists.

The Enduring Value of Human Skills

Goldberg confidently asserts that certain professions, such as plumbing, electrical work, car repair, and restaurant work, will remain secure for the “foreseeable future.” He attributes this to the “incredible dexterity” of humans – the ability to tie knots, cut cables, and fashion objects – a skill set that remains exceptionally difficult to replicate in robots. He reflects on 40 years in the field, stating that attempting to reproduce human dexterity in robots has only deepened his respect for human capabilities. He states, “We’re going to need them because humans are incredibly dextrous.”

Artistic Exploration & AI’s Potential

Goldberg discusses his artistic projects, including an upcoming exhibition in San Francisco exploring AI within historical and cultural contexts. He emphasizes a “critical perspective” on AI, focusing on the importance of “human beings.”

Addressing concerns about AI taking over and potentially harming humanity, Goldberg expresses a lack of fear, aligning with the views of many of his colleagues. He believes that AI’s limitations are well understood and that science fiction often exaggerates its potential. He views AI as a force with a potentially “much more positive effect on society and culture than negative.”

Historical Context & Nostalgia

The conversation concludes with a nod to the history of robotics in science fiction, referencing Robbie the Robot from Lost in Space and the robot from Metropolis. Goldberg emphasizes the importance of considering AI within a broader historical and cultural framework, ultimately returning to the central theme of what truly matters: “which is which is really human beings.”

Technical Terms:

• Joints (in robotics): Points of articulation in a robot’s structure, allowing for movement.
• Algorithm: A set of rules or instructions that a computer follows to solve a problem.
• Synchrony: The coordination of events or actions to occur at the same time.
• Dexterity: Skill and ease in using one’s hands or body.

Synthesis/Conclusion:

The discussion with Ken Goldberg provides a grounded and realistic perspective on the state of robotics. While acknowledging the rapid advancements in AI, he convincingly argues that replicating human-level dexterity and adaptability in robots is a significantly more challenging endeavor, hampered by data scarcity and the inherent complexity of physical manipulation. The conversation underscores the enduring value of human skills and offers a cautiously optimistic outlook on the potential of AI, emphasizing the importance of critical engagement and a focus on human well-being.