I think that Fareed Zakaria’s Washington Post column identifies a real problem but presents a solution that is far too complex and, indeed, misses the nature of the regulatory problem:
The central AI problem is not consciousness; it is agency. A system need not feel anger, ambition or fear to cause harm. It needs only a goal, enough intelligence to pursue it and enough access to the world to act. AI’s are not “going rogue”; they are trying to succeed any which way they can.
This is a systemic problem that we cannot leave to the good graces of private companies. When thinking about regulations, we should focus centrally on how much autonomy we give these systems. A chatbot that answers a question poses one set of risks. An agent that can browse the internet, execute code, obtain credentials, move money or operate critical infrastructure poses another. The principle I would propose is simple: Autonomy should expand only as our ability to monitor and control it expands.
I think the solution is simultaneously simpler and older than Mr. Zakaria imagines: strict liability that attaches both jointly and severally to the model developer, API developer, the application developer, and the corporate deployer. If that strategy were used, insurance would become an important part of AI governance.
If an AI system causes legally recognizable harm, the injured party need not establish negligence, recklessness, intent, or a defect in the model. The plaintiff must establish the harm, causation, and a legally defined connection between that harm and the entity that supplied or deployed the AI service. An insurer asked to cover an autonomous AI service would want to know about sandboxing, permissions, audit trails, model evaluations, financial authority, network access, kill switches, and incident history. A poorly controlled autonomous agent would become expensive or impossible to insure.
When liability attaches both jointly and severally the plaintiff need not identify the source of harm specifically. Only that harm was done and AI was part of the chain that produced it.
Joint and several liability is important here. The injured party should not bear the burden of determining which participant in an opaque technological supply chain was ultimately responsible for the behavior that caused the harm. Let the parties that designed, supplied, integrated, and deployed the system allocate that responsibility among themselves through contracts, indemnification, contribution, and insurance.
Zakaria’s safeguards would not disappear under strict liability. They would become the things an AI developer or deployer must demonstrate in order to obtain affordable insurance.







In theory sounds good. In reality you are going after companies with billions of dollars with strong political connections. The finance people made billions, then destroyed the economies of many nations in the 2000s and almost none faced any real consequences. They squealed about not getting bonuses. Same will happen here.
Steve
Again, my bullshit detector is screaming red.
All the usual bad actors are claiming that AI will end human civilization, but none of them want to treat it like it would. Instead, they are grappling for more money and power. Then, we have the usual opponents acting like trained seals.
You are proposing that the same companies that have allowed my data to be stolen in data breach after data breach are going to suddenly have a “come to Jesus” moment. I am holding my breath.
AI CEOs are trying to gain a monopoly. Politicians are trying to gain power. The financial industry is trying to gain new leverage points. The investor industry is trying to gain new suckers. Washington Post columnists are trying to stay relevant.
In the past, they have shipped everything off to China to be imported, and we were told faulty products were just the price for cheap goods. Now, the script has been flipped. Locating data centers in China is bad, and using cheap Chinese AI is deemed bad. Will using Chinese AI shorten the window from ten to five years or lengthen it to twenty?
If a sandbox cannot contain a hacker, it does not matter whether it is a human or AI hacker exploiting the weakness. If AI is such a technological leap, use it to design an AI hackerproof sandbox. I suspect the one it was running in was designed for humans. While AI is at it, could they design a secure OS, network, actual security, etc.
Of course, that would not generate breathless headlines. “AI Designs Sandbox to Test AI Code” does not generate many sales, or garnew money and/or power by the usual scumbags.
Why are the negatives being touted? Why not what will be possible with AI? Why not what will only be possible with AI? AI was going to save the world, and suddenly, it is the monster under the bed.
What am I missing? Could somebody explain how this will end human civilization in ten years?
TastyBits: When we are not being terrorized with imminent destruction, we are assured AI will make life better by making things cheaper.
Chaos inherent in rapid technological change makes prediction very difficult.
Nearer-term threats: Economic disruption, threats to privacy and the expansion of government surveillance, autonomous weapons including terrorism, the inability to distinguish truth from misinformation, rapid development of biological weapons. Miscalculations due to all of the above leading to war.
Longer-term and highly-speculative threats: Super-intelligent AI pursuing goals misaligned with human interests, potentially rendering humans irrelevant, dependent, or unable to control critical systems; biological agents unleashed leading to the death of billions; or, run amok war machines leading to Terminator-style destruction.
Benefits: accelerated scientific research including medical discoveries, increased productivity and human wealth.
What will actually happen will include many unforeseen aspects. However, it is not wrong to raise the alarm about possible dire consequences. That’s how to avoid them.
A fire breaks out.
Boy sounds fire alarm.
Boy: “Help us put out the fire!”
Old Curmudgeon stays sitting in his rocker: “Town never burned down before.”
Townsfolk put out the fire.
Old Curmudgeon: “Told ya so.”
@Zachriel
The claim is not that AI will cause a fire. It is that AI will end human civilization in ten years, with the implication that something must be done today. Additionally, we do not have people continuing to build fire-prone structures as quickly as possible lest we lose the fire-prone structure race.
The problem with you lists of mishaps is that they apply to humans as well. If we are so stupid to give a computer running an algorithm the ability to develop a biological weapon and determine when to use it, we deserve to be exterminated.
Does it not give you a little pause that the same bad actors who have hollowed out the US and brought us the Great Financial Collapse (GFC) are now worried about humans? Do you really want Trump deciding the fate of human civilization?
Could these Saviours of the Universe work on securing my personal data, first. It might be a nice baby step. Maybe, they could put their AI to work.
As to your little fire example, good try. I know what AI is and how binary circuits work. No matter how sophisticated, a binary logic gate has one of two states. To my knowledge, there have been analogue circuits, but today, everything is binary. AI is statistical regression that no matter how many regressions, it can only return one of two states.
AI is only magic until you understand the trick. Once you know how it works, it is not a scary monster under the bed. While I am mostly unimpressed, I can see how it can be very useful. If software has a weakness, it will be exploited, by humans or AI. If we give control to psychopaths, the results will be unpleasant, human or AI.
(Likewise, humanoid robotics is a magic trick. It is all choreographed. The Terminator ain’t coming anytime soon. Quantum computing will arrive after we get a commercial fusion working.)
I will remind you that the GFC was caused by human psychopaths – no AI was needed. I doubt AI could have engineered that disaster or that it could have engineered the “cure”. We might be better off letting AI run things. I doubt AI scumbags have anything on their human counterparts.
What I see is a lot of wealthy people scared shitless by cheap Chinese AI and/or AI uselessness. Either way, their wealth and the exotic financial engineered structures will implode taking their wealth. We have a bunch of rich people trying to scare the public shitless to save their wealth.
I have to say. I am surprised that you are gobbling up this bullshit. I get it that you might be more impressed than I with the technology, but I thought you would detect the FUD bullshit. Even the investor community is starting to call bullshit on this “circle jerk” (their words).
If this seems disjointed, I apologize. I was trying to keep it short, but I kept inserting paragraphs. I have provided a lot of meat, but it is not as fully fleshed out as I would like.
From your previous comments, can I infer that you would prefer an AI president to Trump?
(When I was very young, I got a “computer” from Radio Shack for Christmas, and I was very disappointed. It was a plastic square with switches, light bulbs, springs, and wires to connect them. In college, I would learn that it was creating logic circuits and that may technically have been stretched to be called a “computer”.
Today’s circuits may be smaller, faster, and have more exotic gates, but they still are nothing more than wires and switches turning light bulbs on and off.)
@Zachriel
Regarding your second comment.
The same could be applied to Iran. I am being told that I am an Old Curmudgeon because to date Iran has not exhibited the capability to build a nuclear bomb, build an ICBM, shrink this non-existent nuclear bomb down to fit on this non-existent ICBM, develop a rocket guidance system capable of intercontinental delivery and have it accurate enough to target NYC.
This is Wile E. Coyote level thinking. It is beyond absurd, but I’ll play along.
To attack a large landmass country with a dispersed population, you are going to need more than one. Iran has shown no signs of being suicidal, and attacking any country with nukes would be suicide. (At this point, Israel may have excluded itself from that protection.)
Keeping a nuclear arsenal with or without ICBMs requires maintenance and replacement when they reach their shelf life. So, not one and done.
So, call me an Old Curmudgeon because I ain’t buying the bullshit being shoveled. You are being fed a shit sandwich. You can call it an AI Crisis, the Iran Nuclear Crisis, or (fill-in-the-blank) Crisis of the Month, but it is still a shit sandwich.
(To all, I apologize for the language. I am trying to tone it down, but the BS is getting pretty deep.)
TastyBits: The claim is not that AI will cause a fire.
It’s an analogy. “The fire will burn down the whole town! Everything will be lost!” Which actually means the town will burn down — if we don’t do anything.
Enter the boy. The boy is almost hysterical, crying, “Fire! Fire!” The Old Curmudgeon notes the town never burned down before. And, of course, that’s because somebody rang the alarm, and some kid or other ran through the town crying, “Fire! Fire!”
TastyBits: It is that AI will end human civilization in ten years
As noted in the above comment, the longer-term scenarios are highly speculative. It is important to be as realistic in the assessment as possible but also not foreclose discussions of highly speculative outcomes because — the town never burned down before as somebody rang the alarm, and some kid or other ran through town crying, “Fire! Fire!”
TastyBits: AI is statistical regression that no matter how many regressions, it can only return one of two states.
Not sure where you got that. Even standard computers return huge numbers of states. However, advanced AIs are qualitatively different, are capable of agency behavior, and are now largely beyond human ability to reverse engineer their actions.
TastyBit: From your previous comments, can I infer that you would prefer an AI president to Trump?
We’re more of the representative government persuasion. Humans need human representation. As for AI, there is no reason it can’t be harnessed for human benefit. Human-AI collaboration can be highly effective. But humans being humans will almost certainly create dangerous AI agents that can work autonomously outside of human control. Or use AI to create biological agents that work autonomously outside of human control.
TastyBits: The same could be applied to Iran. I am being told that I am an Old Curmudgeon because to date Iran has not exhibited the capability to build a nuclear bomb
Iran has had the capability of building a nuclear device for decades. Yet they haven’t.
@Zachriel
Oh, I thought you meant AI would cause an actual fire. Never mind.
There is no AI fire. There is no AI smoke. If Iran had breakout AI capability, would Trump have been justified in bombing their data centers? For the children, of course.
There are weaknesses in the current security regime, and humans are exploiting that without AI. If the Saviours of the Universe are so worried about civilization, they could prove it by increasing security of my data, and I do not mean requiring me to use 2FA to somehow prevent their employees and contractors from stealing my data.
FYI: A binary has two states, period. When I was taking circuit classes, there were AND, OR, NAND, and NOR. I think XAND and XOR plus more exotic ones exist now, but all the code ever written gets compiled into binary executables that uses those logic gates.
A binary circuit does not understand higher or lower languages. It does not understand Assembly or even machine code. Increasing the flow from 4 to 8 to 16 to 32 to 64 simply opens more lanes for each clock cycle to process more bits per cycle.
A CPU is not magic. It is physics plus the properties of silicon plus mathematics. (Binary is kinda magic.) It is like a giant FedEx sorting center. Packages can branch left or right, and they can keep branching. If a package branches back to the start with no option to exit, it is in a Do Loop.
This AI FUD is being generated by CEOs to protect their products from cheap Chinese AI. They want a great big moat built so that they must become our overlords to protect us from the AI monster, especially cheap Chinese AI. Politicians are trying to gain more power by regulating the scary AI monster. Now, I see there are protests against AI.
You are being fed a shit sandwich, and you are trying to convince me it is a Big Mac. I am surprised that you are buying this crap. It came out of nowhere, and not one of the people raising the alarm is an honest actor. If cheap Chinese AI gets a foothold, they all have something to gain, and a lot to lose, like everything. Even if AI is a super scary monster, why do data centers need to be in the US?
When it comes to most things Trump, you are some emo Brutus, but with AI, you are wearing a MAGA hat.
TastyBits:
You wrote
IMO CEOs are generating AI FUD to a) gain control of all AI development and b) crowd out upstarts including but not limited to Chinese upstarts.
They could exclude Chinese upstarts without the measures they’re advocating.