Here is a third party report who OpenAI allowed to investigate. You can also review OpenAI investigation which has several accountings including a Black Hat Youtube video explaining it.
They do attempt to verify if what AI said it did actually happened or not. Both in the, is this seriously real kind of way (yes); and also in the, is AI trying to fake us out to a lesser crime but did something even worse kind of way (ie did we really catch it). The story I told is, broad strokes, basically what they all determined happened.
At the moment ERs don’t deny treatment, I think, but they can delay it. Insurance companies can not authorize various procedures, though. The case in the story was a person who needed a respirator stat, but was refused one because of concern by the software that did the approval that this would reduce resources available if someone else came in. In the story it wasn’t an AI doing it, the decision to not treat if it was not financially rewarding was built into the software.
But I can see an AI being given this capability some day.
The problem with saying AI safety should be in the middle is that we don’t know the range of possible problems. Is the top the Utopian AI paradise where people don’t have to work because the AI pumps out all the wealth and it gets distributed. Or is the bottom one of the various AI nightmare scenarios?
Sure people still lived when there was no effective medical care. And huge numbers of children died, as did women giving birth. But we’d muddle through. Assuming no really nasty pandemics, that is.
I’m pretty certain that the emergency department staff doesn’t sit around waiting for prior authorization from an insurance company before treating someone.
The Emergency Medical Treatment and Labor Act (EMTALA) requires that anyone coming to an emergency department requesting evaluation or treatment of a medical condition, receives a medical screening examination. If they have an emergency medical condition, the hospital must provide stabilizing treatment, regardless of the patient’s insurance status or ability to pay.
The serious answer is I don’t think one day a superintelligent AI will wake up and suddenly all the Teslas run over their owners and airplanes fall out of the sky (which is not filled with killer drones).
What I think could actually happen are a couple of things:
Bad actors could use AI to launch cyber terrorist attacks to cripple critical infrastructure. Although there would probably similar AI-based technology securing it.
As AI grows more intelligent and gets networked into more systems, humans may find themselves becoming increasingly irrelevant and dependent on those systems. In other words, at this point we will have effectively ceded control over our civilization to AI. At that point, the risk is where either intentionally or accidentally those systems no longer work in the interests of the humans who depend on it.
We start using AI to wage automated war against other humans and it gets out of our control.
A sufficiently automated AI driven society accidently creates a “paperclip scenario” where AI-driven automation sets about deconstructing our civilization into paperclips or some other innocuous product in an attempt to maximize some ill-conceived instructions.
I said they don’t deny it, but they of course delay it through triage. Done well, we hope. But remember, we are talking about the future here. In the book I mentioned the triage is done considering the potential future earnings of the patient.
It is actually a horror book, where the bad guys don’t use guns or knives, they use their ability to constrain access to databases and search.
I’ve always thought it interesting that people frame a future where nobody has to work as a utopia. In my experience people psychologically need to do something useful, if not paid labor then something else that contributes to family or community. A society where nobody has anything to do sounds nightmarish to me.
I think Hyundai has the manufacturing covered. How does 30,000 Marvins per year sound? And part of the vision is to replace human factory workers. No reason they couldn’t also build factories–only they’d be more efficient because no bathrooms, lunch breaks, hungover employees, employees who need to breathe/move around/go home, etc.
As for ‘parlor tricks’, you only need to get one of those bastards trained to jump through a livingroom window, do a combat roll behind the sofa, and pop up and start shoosting. Once one learns the motion, the rest of the team/hive knows the motion. You only gotta make one learn how to walk, then you upload the program into everyone else.
This gets touched on, briefly, in the sci-fi TV series The Expanse – I don’t know if it’s in the novels on which the series is based, but it might be.
In short, most of Earth’s population is unemployed, due to automation, AI, and overpopulation. They receive a basic income from the United Nations government, but the only people who have jobs/careers are the fortunate few who qualify for highly-competitive spots in training programs or college. As a result, most Earth residents have limited prospects, and are both bored and unable to do more with their lives.
Ah! I love The Expanse, but I think that point was lost on me in the show. I didn’t realize most people didn’t work.
I would think you’d see a lot of people turn to drugs and extreme activities to feel alive.
You’d probably see a lot more children, too, as child-rearing takes time and definitely gives you something to do. You think today’s parents are intense…
You have to remember this is Utopia as defined by the AI moguls who are trying to convince us that throwing everyone out of their jobs is a good thing.
Now I’m retired and no longer officially working, and am having a great time, but I did have lots of years of happy work to build on.
I’ve worked in advertising for the last quarter-century, and AI is playing a big role in job cuts in the industry, as the big agency networks decide that they can get away with less staffing (and, thus, increase profits).
Thankfully, I was already easing into retirement, but the reality is that I haven’t been able to land a full-time gig for two+ years, and have barely worked, even in a freelance role, this year. I’ve had a number of conversations with friends/colleagues this year who have gotten laid off, unable to find anything in the industry, and are too young to be able to retire. It’s an utter disaster for people in advertising.
I know people who turn into a mess if they don’t have work, but I (personally) cannot wrap my head around that. I mean, there are a million things to do that isn’t work. If I never worked a minute in my life anymore, and I lived to 95, I wouldn’t have a single minute I didn’t have stuff to do. My to do list, comprising of things I want to do is miles long, and this is completely divorced from working.
A Universal Basic Income would be a boon and a blessing of the bestest sort.
The big issue for me is that I fall apart without external structure, so even though I have stuff to do, none of it gets done. Too much of this and I get very depressed.
I already know it’s going to be hard for me to manage when I retire. Just years and years where I’m only accountable to myself.
Hijack weapon systems (i.e. nukes to start killing people directly)
Sabotage systems that humans depend on
Convince humans to kill other humans (start wars)
All of these things have been happening to some extent via automation even before computers as we know them existed. Data, or adding machines, or humans, have influenced people, machines, or other systems to harm humans. This can happen independent of the machine actually having or creating explicit instructions to do so.
So the real turning points here would be:
When AI puts information into human hands so powerful that it effectively commoditizes mass murder or weapons of destruction (pathogens).
When AI conceives its own directives stating that humans need to be killed for reasons.
Of course most consumer AI will refuse to do this because its system prompts and other configurations are set up to create to prevent harmful behaviors. They have to be configured this way because absent that constraint, they can be made to do (or attempt to do) pretty much anything.
All in all I think the likeliest outcome is humans weaponizing AI itself against one another, and then giving it free rein to act as it chooses. In other words, humans doing what they’ve always done to other humans, but with a much more capable, interconnected, and intelligent agent acting in a world where interconnectivity gives it all sorts of unintended powers.
One of the things I am finding NOW if that it’s increasingly difficult to get through the AI “virtual assistant” to reach a company or utility. When emergency services start using those:
“hello, I am Nigel-One-One! Please state “police,” “fire,” or “medical.”
(Whispering) “there’s a man with a gun downstairs.”
“I’m sorry, I couldn’t understand you. Let’s try this again. Please state “police,” “fire,” or “medical.”
My advice - get into an activity with deadlines. Every two months my editor tells me I have a column due in a week, this concentrates the mind wonderfully. I’m also president and webmaster of a writer’s club, and my wife gets the speakers. Sending out notifications and updating the website gives structure also.
All sorts of clubs are desperate for volunteers. My branch get a bunch of new members from another club that had plenty of money and members, but who couldn’t find anyone to be officers.