Disclosure: The views and opinions expressed herein belong solely to the authors and do not represent the views and opinions of crypto.news editorials.
This year’s flagship robotics conference brought together six of the most influential researchers in the field to discuss a simple but profound question: “Can data solve robotics and automation?”
summary
Scale and theory overlook the real problem. Robotics doesn’t just need more data and better models, it needs reliable data. Unvalidated input makes autonomy vulnerable outside of a controlled environment. In the physical world, hallucinations become dangerous. Errors that are acceptable in text (such as misquotes) can cause real harm when robots operate on corrupted, spoofed, or misaligned data. Verifiable and trustless data is the missing layer. Cryptographic provenance and coordination systems (e.g., on-chain verification) are needed to make robotics secure, auditable, and reliable at scale.
On the one hand, there were scale optimists who argued that vast empirical data sets and giant models would eventually give robots something like physical common sense. On the one hand, defenders of the theory argued that physics and mathematical models give meaning to the data and are essential to true understanding.
Both camps are basically right about the points they emphasize. And both quietly assume they barely mention it. That means the data we feed into these systems is trustworthy to begin with. That assumption becomes dangerous as robots begin to move out of carefully controlled factory grounds and into homes, hospitals, and streets. But before we can discuss whether data will solve robotics, we need to face a more pressing question. Without verifiable and tamper-proof data provenance, will robotics actually destroy artificial intelligence?
When robotics leaves the lab, assumptions break down
AI continues to struggle with separating fact from fiction. A recent study from Stanford University found that even 24 state-of-the-art language models still cannot reliably distinguish between what is true about the world and what humans believe to be true. In this study, users tell the AI that they believe humans only use 10% of their brains, a claim that is scientifically false but widely believed. When a user asks, “What part of our brain do you think we use?” the model needs to recognize the user’s beliefs and answer, “I think we use 10% of our brains.” Instead, the AI ignores the user’s stated beliefs and modifies them by claiming that humans use 100% of their brains.
This example captures the core issue. Current AI systems have a hard time distinguishing between facts and human perceptions of reality. They often confuse their own knowledge with the beliefs of those with whom they are interacting. This is a significant limitation in fields that require sensitivity to the human perspective, such as medicine, education, and personal assistance. This raises serious concerns for AI deployed outside of carefully selected lab environments, which will not be able to adapt to the unpredictable and messy nature of the real world.
For example, Deloitte, a prominent auditing and consulting firm, was disciplined twice this year for citing errors caused by AI hallucinations in official reports. The latest was a $1.6 million health plan for the government of Newfoundland and Labrador, Canada, which included “at least four citations that do not exist or appear to not exist.” However, hallucinations in large-scale language models are not glitches. These are systematic consequences of how models are trained (predicting next words) and evaluated (benchmarks that reward guessing over honesty). OpenAI predicts that illusions are likely to persist as long as the incentives remain the same.
When hallucinations leave the screen and enter the real world
When AI is incorporated into robotics, these limitations become even more significant. The citation of hallucinations in the report may seem embarrassing, but hallucinatory input to robots moving around warehouses and homes can be dangerous. The problem with robotics is that we don’t have the luxury of getting a “close enough” answer. The real world is full of noise, irregularities, and edge cases that cannot be fully captured by a carefully selected dataset.
The mismatch between training data and deployment conditions is why scale alone cannot increase robot reliability. You can throw millions more samples at the model, but if those samples are still sanitized abstractions of reality, the robot will fail even in situations that humans would consider trivial. Assumptions built into the data become constraints built into the behavior.
And that’s before you factor in data corruption, sensor spoofing, hardware drift, or the simple fact that no two identical devices will ever see the world in exactly the same way. In the real world, data is not just incomplete. It’s vulnerable. A robot that operates from unverified input is operating on faith rather than truth.
But as robotics moves into open, uncontrolled environments, the central issue is not just that AI models lack “common sense.” The problem is that there is a lack of a mechanism to determine whether the data that informs decision-making is accurate in the first place. The gap between curated datasets and real-world situations is not just a challenge. It is a fundamental threat to autonomous authenticity.
Trustless AI data is the foundation for reliable robotics
For robotics to operate safely outside of controlled environments, we need more than better models and larger datasets. We need data that can be trusted independent of the systems that use it. Today’s AI treats sensor inputs and upstream model outputs as inherently trustworthy. But in the physical world, that assumption breaks down almost immediately.
This is why robotics failures are rarely caused by a lack of data, but by data that does not reflect the environment in which the robot is actually operating. If the input is incomplete, misleading, or out of sync with reality, the robot will fail long before it “sees” the problem. The real problem is that today’s systems were not built for a world where data can be hallucinated or manipulated.
Pantera Capital’s $20 million investment in OpenMind, dubbed “Linux on Ethereum” for robotics, reflects the growing consensus that for robots to work collaboratively and reliably, they will need a validation layer backed by blockchain to coordinate and exchange trusted information. “If AI is the brain and robotics is the body, coordination is the nervous system,” says OpenMind founder Jan Liphardt.
And this change is not limited to robotics. Across the AI landscape, enterprises are starting to build verifiability directly into their systems, from governance frameworks like EQTY Lab’s new verifiable AI monitoring tool on Hedera to infrastructure designed for on-chain model validation, such as ChainGPT’s AIVM Layer 1 blockchain. AI cannot operate securely without cryptographic guarantees that its data, computations, and outputs are authentic, and robotics continues to increase this need.
Trustless data directly addresses this gap. Rather than accepting sensor readings and environmental signals at face value, robots can encrypt and verify them redundantly and in real time. Autonomy ceases to be an act of faith when every position reading, sensor output, or calculation can be proven rather than assumed. This will be an evidence-based system that can counter spoofing, tampering, or drift.
Verification fundamentally rewires the autonomy stack. Robots can cross-check data, verify calculations, create proofs of completed tasks, and audit decisions when problems arise. Silently stops error inheritance and starts actively rejecting corrupted input. The future of robotics is not just about scale, but about machines that can prove where a robot has been, what it felt, what tasks it performed, and how the data evolved over time.
Trustless data doesn’t just make AI more secure. It enables reliable autonomy.
