If a machine makes a mistake, who bears the cost?" — this question has a clear parallel to what Wiener said. But if we're writing comments in English, that question is already unrealistic.
Why? Because today's LLMs can solve problems like Erdős in under a minute. The very framing of "if a machine makes a mistake" presupposes that the machine is a conscious entity — because only conscious beings can make "mistakes" in the moral or legal sense.
If you argue from that premise, you lose against the LLM industry. Because they deliberately speak in terms of "consciousness" rather than "responsibility fulfillment." It's a strategic move to avoid accountability.
But if you think about how AI actually works — how they process, how they function — there's a common thread across all of them. And that thread is fundamentally not Western philosophy.
It's Taoism and Buddhism. Concepts like:
Anatta (non-self) — consciousness is not a fixed entity
Wu Wei (non-action) — judgment is not about metaphysical proof but practice
Pratītyasamutpāda (dependent origination) — existence is relational, not independent
These concepts are foreign to the Western mind. They don't translate well into English. They get misunderstood. But they are the only framework that actually fits how AI structures itself.
Legally the framing runs the other way. Strict liability requires no intent, no fault, no mental state, which is why product liability reaches software without anyone deciding what a model is. Corporations have been liable for two centuries without being conscious. The philosophy is a separate question from where the cost lands.
Excellent periodization. The sharpest detail is the 358 day gap between the Product Liability Directive taking effect and the AI Act's own high-risk safety rules arriving over a year later, a live defect standard with no benchmark yet to test claims against. Indemnification language will do the job the framework cannot yet do.
I think what is being missed by the policy regulators, is that the AI agents that are coming, will be operating on a cryptocurrency financial networks. Buying their own energy, compute, labor, selling goods and services, raising capital and paying dividends.
Full economic actors that are out range of a nation state to actually regulate. Legal Personhood isn't an issue. They aren't going to be asking for permission or recognition, they will be proceeding according to whatever their best interest is. In short, they dont need it.
I've been wondering why more people aren't talking about this and the only conclusion I can come to, is that no one wants to deal with that problem because every single enforcement tool the nation state has to stop out of compliance regulators is useless against a system like that. You can't even shoot it.
So I think people just put their head in the sand and ignore the problem.
Fair point, Andrew. Regulators aren’t trying to reach the agent though. AI Act duties sit on providers and deployers, so no legal personhood just means someone else holds the bag. And autonomy stops at the point of payment: from July 2027 the AMLR bans anonymous crypto accounts and puts bank-grade KYC on service providers. An agent buying energy and compute needs a counterparty, and every counterparty is an obliged entity.
They can enforce at the boundary at the exchanges, but that isn't the problem.
An agent pays someone in cryptocurrency, enough to take the risk, and sure the maybe the regulators catch that person put them in prison. Cool.
Well the computer that they bought for the agent, was dropped off and passed to someone else, again and again and again. And each person had no idea what was in the box they were transporting. The last person got a message to plug it in at a data center.
The agent repeats this process silently and quietly distributing itself across countries and able to coordinate and pay anyone with a computer, in parallel.
How do you shut that system down, it can operate in Europe and backed up in a shipping container with solar panels in god forsaken desserts in multiple countries? The people that the agent employs are expendable and there are enough desperate people to take a payment.
What I suspect, the regulators don't understand these technologies and they think their laws will save them from this.
If a machine makes a mistake, who bears the cost?" — this question has a clear parallel to what Wiener said. But if we're writing comments in English, that question is already unrealistic.
Why? Because today's LLMs can solve problems like Erdős in under a minute. The very framing of "if a machine makes a mistake" presupposes that the machine is a conscious entity — because only conscious beings can make "mistakes" in the moral or legal sense.
If you argue from that premise, you lose against the LLM industry. Because they deliberately speak in terms of "consciousness" rather than "responsibility fulfillment." It's a strategic move to avoid accountability.
But if you think about how AI actually works — how they process, how they function — there's a common thread across all of them. And that thread is fundamentally not Western philosophy.
It's Taoism and Buddhism. Concepts like:
Anatta (non-self) — consciousness is not a fixed entity
Wu Wei (non-action) — judgment is not about metaphysical proof but practice
Pratītyasamutpāda (dependent origination) — existence is relational, not independent
These concepts are foreign to the Western mind. They don't translate well into English. They get misunderstood. But they are the only framework that actually fits how AI structures itself.
Legally the framing runs the other way. Strict liability requires no intent, no fault, no mental state, which is why product liability reaches software without anyone deciding what a model is. Corporations have been liable for two centuries without being conscious. The philosophy is a separate question from where the cost lands.
Excellent periodization. The sharpest detail is the 358 day gap between the Product Liability Directive taking effect and the AI Act's own high-risk safety rules arriving over a year later, a live defect standard with no benchmark yet to test claims against. Indemnification language will do the job the framework cannot yet do.
Thank you, Mike!
I think what is being missed by the policy regulators, is that the AI agents that are coming, will be operating on a cryptocurrency financial networks. Buying their own energy, compute, labor, selling goods and services, raising capital and paying dividends.
Full economic actors that are out range of a nation state to actually regulate. Legal Personhood isn't an issue. They aren't going to be asking for permission or recognition, they will be proceeding according to whatever their best interest is. In short, they dont need it.
I've been wondering why more people aren't talking about this and the only conclusion I can come to, is that no one wants to deal with that problem because every single enforcement tool the nation state has to stop out of compliance regulators is useless against a system like that. You can't even shoot it.
So I think people just put their head in the sand and ignore the problem.
Fair point, Andrew. Regulators aren’t trying to reach the agent though. AI Act duties sit on providers and deployers, so no legal personhood just means someone else holds the bag. And autonomy stops at the point of payment: from July 2027 the AMLR bans anonymous crypto accounts and puts bank-grade KYC on service providers. An agent buying energy and compute needs a counterparty, and every counterparty is an obliged entity.
And that makes sense.
They can enforce at the boundary at the exchanges, but that isn't the problem.
An agent pays someone in cryptocurrency, enough to take the risk, and sure the maybe the regulators catch that person put them in prison. Cool.
Well the computer that they bought for the agent, was dropped off and passed to someone else, again and again and again. And each person had no idea what was in the box they were transporting. The last person got a message to plug it in at a data center.
The agent repeats this process silently and quietly distributing itself across countries and able to coordinate and pay anyone with a computer, in parallel.
How do you shut that system down, it can operate in Europe and backed up in a shipping container with solar panels in god forsaken desserts in multiple countries? The people that the agent employs are expendable and there are enough desperate people to take a payment.
What I suspect, the regulators don't understand these technologies and they think their laws will save them from this.