- The Next Phase of AI: Embodied AI
- The Concept: While current AI (LLMs) dominates digital tasks like writing and coding, the next frontier is the physical economy.
- Definition: Embodied AI is any machine that collects data (photons), perceives the world, learns, navigates, and manipulates 3D space.
- The “Cambrian Explosion”: We are on the verge of a massive proliferation of robotic form factors, from cars to humanoids, similar to the biological explosion of life forms.
- The Value of Physical Data
- “Fat Tuna” Analogy: Just as a valuable tuna is worth nothing if you can’t catch it, the world’s visual data is worthless without the means to collect it.
- The Race for Photons: Tech companies are racing to collect real-world visual data to train “Vision Language Actuation Models.”
- Biological Efficiency: The video uses the fruit fly (Drosophila) to illustrate how nature solves complex navigation with tiny “hardware” (a poppy-seed-sized brain), suggesting AI needs to emulate this efficiency through massive simulation.
- Simulation & The “Sim-to-Real” Gap
- Digital Twins: Robots “dream” in hyper-realistic simulations (physics engines) to learn skills safely and quickly.
- Examples: Tesla cars aren’t just driving; they are collecting data to train FSD (Full Self-Driving). Meta glasses aren’t just for recording; they are training avatars to perform human tasks like cooking or cleaning.
- Key Sectors & Players
- Autonomous Vehicles (Tesla & Waymo):
- Tesla: Highlighted for its “moat” consisting of data (7 million cars), in-house robotics, energy storage, AI (Dojo), and vertically integrated manufacturing.
- Waymo: Google’s bet on autonomous taxis, forecasted to grow from ~1,500 units to 23,000 by 2030.
- Meta (Reality Labs): Smart glasses are a “Trojan horse” to collect first-person data on human hand movements to train future humanoid robots.
- Amazon: Transforming from “human-heavy” to “robot-heavy.” Predicted to potentially launch “Amazon Bot Services” (ABS) similar to AWS.
- SpaceX: Creating the “connective tissue” of the AI ecosystem via Starlink and reducing launch costs by 10,000x, enabling a new internet for connected devices.
- Defense (Drones): A shift from expensive, manned platforms to swarms of cheap, autonomous, “attritable” drones.
- The Economic Opportunity (TAM)
- Labor Market: The global labor market is ~$40 trillion. A humanoid robot costing $5/hour could replace human labor costing $25/hour.
- Impact: Every 1% substitution of human labor by robots in the US is worth ~$300 billion.
- Re-Industrialization: The video argues this technology offers a chance to reverse the 80-year decline in US manufacturing as a percentage of GDP, countering Carl Sagan’s 1995 prophecy about the US losing its industrial base.
- Investment Strategy: “The Humanoid 100”
- Morgan Stanley introduces a framework to map equities across three categories:
- Brain: Semis, AI models, software.
- Body: Actuators, batteries, sensors, materials.
- Integrators: Companies putting it all together (e.g., Tesla, Amazon).
This is an eye opener for me, as I had limited my interest to the integrators at the expense of all those other areas of industry. I would point out, though, that vision is just one thing out of several that allows me to walk with a full cup of coffee without spilling it, or balance a spinning basketball on my fingertips. The more humanoid robots there are, the better they will perform.
If we are to realize the world in this infomercial it’s going to need a market to serve. If labor and other human functions are obviated by AI and intelligent bots, how will humans purchace or access their services without remunerative exchange. This innovation may solve one set dilemmas but creates another much larger dilemma.
We’re going to need a completely different type of economic model in order for all this inovation and automation not to implode on itself. It’s a bit of conundrum. There is no doubt in my mind that in order to overcome this paradox we’ll have to adopt a form guaranteed universal basic income and find meaning and purpose outside and beyond traditional exchange and remuneration. This all could happen, but not under the current paradigm.
One of the main things that makes life worth living is figuring out things, and doing things to help ourselves and others. If we build machines that figure out too many things for us, and do too many of the things we’ve evolved to do for ourselves, will a lot of people atrophy to the point where a lot fewer people find life worth living, due to the lack of self-agency brought about by relying too much on our machines? Sure we’d like to think that nothing would stop us from using our machines in the ways we’re trying to design them for, to help us lead good lives in which we’re no longer scrabbling for survival, and instead that we’ll have the time and resources to learn all kinds of things and to do good and interesting things, and it’s likely some, maybe even most people would do that to varying degrees, but we really don’t know the percentage of people who will be able to wisely use our machines for these goals, as opposed to those who would just become overly dependent on them.
