Picture a company CEO obsessed with building a Large Language Model (LLM), thinking this is the way to win the intelligence race. Monday mornings she grills her team on variables like benchmark scores, token costs and hallucination rates. Every hard-won incremental improvement feels like a triumph. “We are really building something special,” she cheers her staff on.

But in all that time, a single question never gets asked, one that might give our CEO a real edge in the AI Age: how will it learn to pick up a pen?

Are We Witnessing A Physical AI Awakening?

Across corporate America, many executives still think like our fictional CEO. Myopically, they focus on digital AI, the kind that lives on your browser, phone, or tablet. Chris Chen, Co-CEO of Faraday Future , an embodied artificial intelligence (EAI) company, believes that’s the wrong way to prepare for the future that’s fast materializing.

“The whole world is on the cusp of physical AI,” he told me in an interview at the official launch of the company’s EAI Robotics Education Ecosystem in Los Angeles. Here his organization demoed various models of its robotics line, which Faraday Future brands collectively as “EAI robots,” including its second-generation flagship full-size humanoid robot and FX Navi, a micro quadruped EAI robot for family and school-based AI education. The event also offered a preview of two more humanoid models still awaiting full launch: Master Mini, developed for youth education and athletic competition, and Nova, a smaller companion robot for children.

Chen’s point is that digital AI was trained on “a massive amount of data—billions or trillions of words from books, articles, websites, code and other text sources,” according to IBM.com . Over time, models learned to predict, reason, and detect patterns, leading to major gains in artificial intelligence, especially the generative type that has roiled the white-collar sector, prompting Anthropic CEO Dario Amodei to suggest “AI could wipe out roughly 50% of all entry-level white-collar jobs,” per Fortune .

But all that data ingestion has mostly enhanced disembodied AI. Impressive as that is, and it is a stunning feat, there’s a world of difference between two-dimensional digital screens and AI that can act and move in a three-dimensional world.

To put it in more concrete terms, an AI that can describe the sound of an egg breaking using Faulknerian language is helpful. A robot with an AI brain that can crack those eggs to make your fiancé a souffle signifies a sea change executives may ignore at their peril.

Why Embodied AI Is Such A Training Headache

It sounds odd to suggest it, but digital AI learning is almost easy compared to what’s required for embodied AI. “A robot must face countless real-world scenarios,” says Chen. “That's much more difficult, especially when it comes to navigating forces like gravity, balance, friction and uncertainty. Embodied AI must understand the laws of physics.”

Knowing this, Chen suggests the coming bottleneck isn’t intelligence as it was throughout much of the 21 st century while AI came online. Rather, it’s acquiring sufficient real-world data. After all, you cannot build embodied AI quadrupeds that work well enough for the consumer market to embrace them without sufficient data to inform their locomotion and behavior.

To reiterate, ChatGPT, Claude, Gemini, Grok and other frontier models were trained on copious amounts of text and images. The Robotic Revolution that I previewed in a past Forbes article can only take off if companies train embodied AI on the myriad physical activities developing humans like babies and toddlers take for granted: touching, walking, carrying, failing and of course, recovering.

The Sim-to Real Challenge Is Real

Okay, fine, you might be thinking . Then why don’t we just trains robots using simulations? We train human pilots that way .

According to training-data company Claru.ai , the “sim-to-real gap” is real. And gnarly. “Differences in visual rendering, contact physics, sensor noise, and actuator dynamics between simulator and real-world cause policies to encounter conditions they never saw during training, leading to failures that range from reduced accuracy to complete task breakdown.”

This same source suggests robots enjoying a 90% success rate in synthetic environments may only achieve a 30-60% performance rate in the real world.

Physical Data Collection is the Real Gamechanger

“Being first to deliver robots to real users matters because the latter will generate the data that improves future robots,” says Chen. We can see this truism with the example of self-driving cars. A consumer with only a facile understanding of the electric vehicle charging in their garage may be excused for merely appreciating the fact they don’t have to pay for gas anymore.

What they may not realize is that such vehicles are “more than meets the eye,” to borrow a line from Transformers . They are, in fact, data collectors . “Tesla achieves their self-driving features through AI/Machine Learning combined with an incredible system of data acquisition and aggregation from various sources. All Tesla models are equipped with sets of sensors, cameras, and radars, that constantly collect data from individual cars’ detailed surroundings from the outside and driver and passenger gestures within the car,” explains Harvard Digital Innovation and Transformation .

So where might such all-important physical data collection come from in the future? Our youth. To this end, Faraday Future has invested in a strategic two-way partnership with Triple I, an education institution, to promote summer camps where young people learn the types of hands-on robotics skills needed for tomorrow’s workforce, while at the same time students contribute to embodied AI’s physical understanding of the real world.

The Workforce Reckoning We Must Discuss

While there are real-world benefits to such training, it would be disingenuous to suggest that there aren’t economic risks to such technological disruption. Just as digital AI has disrupted the white-collar workforce, as Amodei acknowledged, it is troubling to recognize that some blue-collar human workers are already training their robotic replacements. “The latest upheaval at Amazon comes from the rollout of advanced warehouses outfitted with robots. The company is overhauling older facilities to reduce their worker headcount,” according to Dissent Magazine.

How automation ends up affecting humans working at factories, not just at Amazon, but globally, remains to be seen. For now, the key insight for executives everywhere should be not just to focus on training for digital AI, but also to prepare for the day when embodied AI takes off.

None other than Nvidia CEO Jensen Huang has publicly stated that if he were a 20-year-old today, he would focus on the physical sciences as the next big opportunity. Huang told CNBC, “The next wave requires us to understand things like the laws of physics, friction, inertia, cause and effect.”

Huang’s remarks, coupled with Chen’s, underscore a broader shift in thinking no CEO should dismiss. If the last few years have taught us anything, it’s that AI is fast becoming an essential technology like the internet, and before that electricity. To keep up with the pace of rapid change, executives everywhere need to make the leap from digital to physical, anticipating a world where AI isn’t limited to your screen but is part of your living, breathing everyday reality.