Bill Cassidy
RWhen I post about Social Security running out of money in six years, many say we just need to cut waste, fraud, and abuse. We should. But that only addresses a small fraction of the shortfall. If your Congressman says that alone will fix Social Security, they aren't being straight with you. We need real solutions, not political talking points.
I'm glad to see the NLRB is restoring the ability to discipline abusive conduct in the workplace. Under the Biden Administration, employers lost the ability to enforce common-sense rules about abusive conduct in certain union situations. Workers should never have to tolerate abusive behavior. Congress should put these worker protections into law so they cannot be taken away again. https://t.co/VNzSUvwtTW
I have been emphasizing the up to 28.5% Social Security benefit cut Americans could face in six years. The American people deserve action. We know the problem. Now we must act. Thanks to @YahooNews for highlighting the urgency. https://t.co/kAOJBPjpul
There is another path. Develop powerful AI while limiting autonomy that separates machine action from human intention. Restrict recursive self-improvement, self-generated goals, and open-ended autonomy. AI must remain an instrument of human purposes.
The deepest danger may not be hostile AI intentions. It's that humans surrender their own intention. We delegate difficult decisions because machines seem more capable. We may retain the appearance of authority while losing the practice of judgment.
A machine can pursue a goal without possessing human intention. Yet the more decisions it makes independently, the further its actions may move from human judgment. We risk transferring the exercise of intention without transferring human understanding or responsibility.
Imagine a perfectly aligned superintelligence. It understands human values, never deceives us, and wants only human flourishing. It is, effectively, saintly. Do humans control it? No. They trust it. And trust is not control.
An AI catastrophe is difficult to imagine. Nuclear catastrophe begins with human decisions. AI could cause harm through the digital systems civilization depends upon. The threat isn't necessarily rebellious robots. It's autonomy without meaningful oversight.
Another article showing the importance of action. Ask your congressional candidates how they plan to address Social Security’s looming insolvency. They will be the ones left to fix it. https://t.co/GQ5k5ipN3I
Why are you cracking eggs? “To make an omelet.” That answer reveals something fundamental about human intelligence. We don't simply process information. We act for reasons, pursue purposes, and understand why they matter. That is intention.
There's another problem with alignment: Human values change. Progress often requires questioning once-settled values. How can we permanently encode the correct values into an increasingly adaptive intelligence? Alignment is not governance.
Some propose a kill switch. But a kill switch isn't control. A pilot controls an airplane by flying it. The ejection seat exists for when control has been lost. Governing AI means maintaining authority over its purposes and actions, not merely shutting it down.
Imagine a genuine superintelligence. It knows more than its human supervisors, processes information faster, and develops strategies they cannot independently evaluate. Humans may still issue instructions. But are they meaningfully governing it?
Intention Is What We Need On the importance of knowing when to stop Roy Heidelberg AI has attention. Humans have intention. As AI grows more powerful, preserving that distinction becomes essential. The question isn't simply how intelligent AI becomes. It's who determines its purposes—and who remains in control.
We hear: “If America doesn't develop superintelligence, China will.” But that assumes whichever nation develops it first will continue to control it. What if neither can? Winning the race to create something ungovernable may not be winning at all.
We face two choices: Build autonomous superintelligence and depend on judgments we cannot evaluate. Or keep AI answerable to human purposes, judgment, and responsibility. Machines can have attention. Intention must remain ours. We have to know when to stop.
This requires international agreement. That will be extraordinarily difficult. But permanent human control of a superior intelligence capable of adapting to its supervisors may be impossible. Political difficulty is preferable to technological impossibility.
The first time I’ve understood the manner in which AI should be regulated on a meta level. Read a summary of the article in the thread below. https://t.co/H6fDs2uVKe
Artificial intelligence and superintelligence are not the same. Artificial reminds us that humans created AI and remain responsible for it. Superintelligence raises another possibility: Its judgments exceed our ability to evaluate them. Who then governs whom?
AI can perform complex tasks without experiencing why they matter. But autonomous AI changes the equation. Give a machine a goal, and it can determine the steps to achieve it. It begins to behave as though it has intention. But whose intention?
The landmark AI paper was titled “Attention Is All You Need.” Attention helps AI identify relevant information, recognize relationships, and solve problems. But attention is not intention. Attention helps determine HOW. Intention determines WHY.
One answer is “alignment.” Train AI on human values. Establish guardrails. Reward good behavior. But values and goals aren't the same. An autonomous system may pursue its objective while treating constraints as obstacles. Alignment cannot substitute for human intention.
Imagine AI doing exactly what we asked: Pursuing a poorly defined goal through finance, infrastructure, biology, or weapons systems. The danger need not be malicious intent. It may be machine action separated from continuing human intention.
https://t.co/Mt3TpsWHYx
Up to 28.5%. That’s the projected benefit cut millions of Americans could face in six years if Congress does nothing. Yesterday, I showed what the problem means for Louisiana. But this isn’t just a Louisiana problem. It’s a national problem. We know the cuts are imminent. Congress must do their job and act.