The most powerful data-network in history might be closer than you think. David Sacks recently sketched a future where a humanoid robot returns to their charging dock, not just for power, but to upload what they’ve learned, so that an entire network of “sister” machines can wake up smarter as well.
He’s quite right. Network effects have already proven themselves in human-focused social networks. There, more users bring more content, and lessons learned somewhere in the network can be rolled out to all other users within minutes, improving everyone’s experience. However, when it comes to network effects, robots have a massive edge over humans. Robots can exchange skills digitally, making learning nearly instantaneous. When robots share, each of them becomes inherently more capable, compounding small improvements into massive jumps of aggregate network performance.
Imagine a robot doctor, having discovered a new way to detect cancer, sharing that skill within seconds, allowing all other machines to better improve their human patients’ health, whilst also feeding back additional information to make the new skill/classifier even more robust. That’s a truly powerful tool.
Right now, industry norms and current robotics business models limit such opportunities for health, human education, manufacturing, and defense. Most robots live inside walled gardens: they talk over proprietary data buses, store model weights on siloed clouds, and operate under NDAs that forbid model sharing. Robot experiences die on-device. Their knowledge doesn’t transfer because the global systems for permission, identity, governance, and provenance remain to be built.
The next major unlock for machines is giving them the ability to prove who they are, where they are, and what they are doing, so that they can exchange skills, securely, under clear terms, and without losing control. What’s needed is a coordination layer where robots exchange skill packets over open transports wrapped in a schema that spells out hardware targets and license terms.
Onboarding would allow robots to register their identity, provide information about their physical and social environment, sign their data streams, and license their skills to other machines under constraints of their choosing. To protect their rights, robots could use one-time-programs (OTPs) and N-time-models (NTMs), which use trusted execution environments to enforce limits on how many times a new model, corresponding to a specific skill, can be used by a recipient.
Each skill could then be cryptographically bound to a recipient robot, with usage tracked and enforced at the hardware level. Over time, this will enable a trusted marketplace for robot capabilities, where skills can be licensed, audited, and monetized, just like cloud services today.
Once machines can prove their identity, location, and reputation, knowledge sharing could become safe and scalable. But wait – the sci-fi movies we have all seen, such as the Matrix or Star Trek (with its Borg collective), seems to point to a bleak future where humans are surrounded by powerful collectives of smart and highly capable machines.
That’s why AI and robotics companies need to consider risks and durable governance of such systems. Anthropic is one leader with its “constitutional AI”. Google baked Asimov’s Laws of Robotics into Gemini Robotics.
The network effect Sacks described is coming, but let’s make it open, understandable, and safe. Most importantly,let’s build it here, together in the US.
About the author:
Jan Liphardt is the founder of OpenMind, a San Franciscio startup developing OM1 – the first open-source, AI-native operating system for robots. Liphardt is also an Associate Professor of Bioengineering at Stanford University, where he focuses on new measurement tools and using sensitive health data to make good decisions.
Liphardt’s awards include being a Searle Scholar, a Sloan Research Fellow, a Hellman Fellow and the recipient of the Mohr Davidow Ventures Innovator’s Award. His research has been supported by the US Department of Energy, MITRE, and the NIH and NCI.
