I'm shocked!

Your conjecture is actually incorrect. The Ottoman Empire rejected electricity early on, out of fear of fire hazards (not an unreasonable fear), and also because government officials had contracts with gas companies that would be worth less if electricity took off. https://journals.sagepub.com/d... We do see some of both of those kinds of things happening today with AI.

A country is not a chunk of land, it's a political entity made up of people. Countries *claim* chunks of land, but the land is not the essence of the country. Boundaries change, moving that land from one country to another. The land didn't change, just the political boundaries.

You keep imagining that AI introduces something new here. It doesn't.

Tragedies happen all the time to people who are not using a technology. Your example of car crashes is a good one. If the car crashed into pedestrians because the driver was careless, the driver is liable. If the car crashed into pedestrians because of a mechanical failure due to a manufacturing defect, then the car company is liable.

AI is no different. If a person uses AI carelessly causing damage, the person is liable. If the damage is due to a defect in the way the AI was built, then the AI company is liable.

But would you *really* know that it's private?

Investing in "people coming up with ideas". What does that mean, exactly? And do you mean "people coming up with non-AI ideas"? Why can't AI be part of the ideas people come up with?

Did you know that countries create...software? Crazy, right? Yeah and most of it is outdated and clunky. But they do create software, lots of it. That's just one thing governments use AI to do.

"Snake oil" is a term for products that don't work. AI does in fact _work_. Sure, it's still immature as a product, and it has issues, as any new technology does. But it's not _fake_.

Definitely a biased statement. But that doesn't make AI "snake oil."

AI is capable of doing real work. And each month the tools get better, making that work higher quality work.

After writing code professionally for 38 years in 28 languages, I no longer want to write my own code. AI does it so much more quickly, and does a good job too. Yes, it makes mistakes, all the time. But even with the mistakes, it saves me a *ton* of time, letting me focus on building what I want to build, instead of quirky syntax of whatever is the latest language du jour.

Of course! Every powerful new technology has both great usefulness and a dark side. There is no exception to this rule. Mechanized factory equipment ripped off a lot of arms and cut off a lot of fingers, even killed a lot of people. It took decades to build safety mechanisms to reduce such risks, but still, the risk is not zero.

AI will go through the same progression. It will introduce risks early on (as it already has). But that's not a reason not to use it. We have no choice but to learn how to manage its flaws. This is not impossible, the dark side of AI will be managed with time.

AI shouldn't be held "strictly" responsible for crimes, any more than the farm equipment. Rather, the manufacturer is legally liable, and subject to safety regulations. It's tempting to anthropomorphize AI and think that we should hold "it" responsible. No, the manufacturer is still ultimately liable.

AI doesn't "decide" to do things on its own. It does what it does as a result of its training and prompting. These *are* controllable. The technology is still immature, and therefore more dangerous, than a mature technology. Just like every other new technology ever.

Compliant solar panel plug-in generators don't cause back-feeding, and have never been documented to injure or kill a line worker.

https://cleanenergy.org/news/p...

But they'll last at least a month or two!

The problem is, agency is where the value is. I don't want to ask AI how to highlight differences between two data sets, I just want it to do it. In coding, I don't want to ask AI how to calculate an encryption hash in .NET that uses the same mechanism as SQL Server so I can tell if certain records have been altered, I want it to implement the changes for me. I also like that when I told it I wanted to write code to import a certain CSV file format, it ingested the sample file from my downloads folder and used that to figure out how to interpret the columns. I also like that when changes cause a build error, it automatically detects that and fixes the build error. These are real time savers.

Like you, I don't let it commit. But agent mode *is* where the power is.

Agreed, my experience mirrors yours. I do think that bright young users could get good use of AI, just as I've worked with bright young college interns that could be a real help in coding without AI. Unfortunately, the majority are not "bright." That's not a slam, just how the normal curve works.

Not exactly, but close. From the details I've read about these so-called "rogue AIs" they were in fact instructed to attempt to break in to accounts and impersonate people. The testers, knowing how risky their instructions were, place the bot inside what they thought was a secure network not connected to the internet. But they didn't make the environment secure enough, and the AI did what it was told to do, and found holes in the container.

That's not so unlike regular software. If you build software that can hack, and you try testing it in a container that isn't adequately secured, it's going to do what it was built to do.

What we're dealing with here is not _uncontrollable_ AI. We're dealing with an immature technology that still has significant issues, being tested in a not-quite-locked-down environment.