Title: Why Physical AI Is Making Waves and Hailed as the Next Major AI Breakthrough
In this column, I'm diving into a sizzling AI trend commonly referred to as Physical AI, or sometimes Generative Physical AI—a fusion of generative AI and its rumored additional physical competence. This evolution in AI technology allows it to become even more useful, extending its reach into countless physical mechanisms and devices, such as robots, automata, humanoids, autonomous vehicles, and virtually any AI-powered machine capable of moving within the concrete realm we inhabit.
This analysis is part of my ongoing series for Our Website, which covers the latest in AI, shedding light on its complexities and identifying significant trends. For more details on how generative AI, like ChatGPT, Claude, Llama, Gemini, et al., is merging with robotic arms and similar technology, check out this link.
Navigating the Physical World
Before we delve into the AI side of things, let's take a moment to ponder the nuances of the physical world: our tangible, three-dimensional environment that we navigate with our senses and bodily movements.
A captivating example to illustrate this concept is the journey of a baby or toddler as they master the art of standing upright for the first time. By wobbling, falling, and attempting to right themselves, they actively engage with the physical world, unconsciously learning intricate lessons about gravity, balance, and motion.
These foundational lessons occur almost instantaneously, well before children are explicitly taught physics or math concepts. In this manner, we all grapple with the physical world, shaping our actions and responses based on our unique encounters and experiences.
The AI Perspective
Translating this to AI, we must consider the current state of its understanding of the physical world. Today's generative AI, including conversational models like ChatGPT, tables mathematical calculations, along with a semblance of human-like writing, but remains a mathematical pattern-matching tool devoid of sentience.
To demonstrate this, let's ask ChatGPT about a simple physical process, such as dropping a rubber ball.
- User Prompt: What will happen if I drop a rubber ball from shoulder height?
- ChatGPT Response: When you drop a rubber ball from shoulder height, it will fall due to gravity, striking the ground. The ball will then rebound due to its elasticity, Employing a scientific perspective, this bounce is the result of energy being stored during impact and then released when the ball collides with the earth once again. However, with each successive bounce, the height decreases due to energy losses such as heat, sound, and deformation.
While the answer is accurate, it's essential to recognize that AI isn't "knowing" the physical world in the human sense. It's merely processing patterns from its training data—learning to mimic human responses based on the books we've written and the stories we've told.
The Dawn of Physical AI
This gap in AI's capabilities highlights the need for further advancements. Two paths forward in the realm of Physical AI are:
- Meadow of Words: Enriching generative AI's knowledge base by incorporating more text-based content about the physical world and its complex scientific principles. This "book learning" approach will help improve AI's command of physics and physical processes, enough that it may eventually gain a semblance of physical intuition.
- Garden of Experiences: Encouraging AI to have real-world experiences by being integrated into physical environments, where it can directly observe and interact with physics in motion.
The label coined for AI equipped with physical comprehension is Physical AI—or Generative Physical AI for AI powering this new kind of machine intelligence.
The AI community is racing to unlock the secrets of Physical AI. With applications ranging from robotic surgery assistants to manufacturing automation and self-driving cars, the potential for reshaping how we live, work, and play is enormous. However, as we embrace this brave new world, ethical and regulatory considerations arise that require careful navigation.
In the words of Isaac Asimov, "A robot may not injure a human being, or, through inaction, allow a human being to come to harm." It is society's responsibility to ensure its AI creations remain ethically aligned as they become intimately entwined with the physical world.
- The 'embodiment intelligence' concept in Physical AI involves integrating generative models like OpenAI's ChatGPT (o1, o3, pro plus, gpt-4o), Anthropic's Claude, Mistral AI's Llama, Google's Gemini, and Microsoft's Copilot with robotic arms and similar technology, as explored further in this link.
- The endeavor to develop 'agentic physical AI' aims to bridge the current gap between AI's mathematical pattern-matching capabilities and its understanding of the physical world, paving the way for AI systems to demonstrably 'know' and interact with the world like humans do.
- Advancements in 'physical AI' could significantly impact various industries, including healthcare with robotic surgery assistants and manufacturing with automation, as well as disrupting transportation with improvements to autonomous vehicles and self-driving cars.
- In the pursuit of Physical AI, it's crucial to consider ethical implications, emulating the words of science fiction author Isaac Asimov: an 'agentic physical AI' should prioritize avoiding harm to humans, both through its actions and inaction.
