That’s exactly my point. It’s free today for casual users to access the base LLM, but that will not always be the case. No misunderstanding your post at all.
DH currently pays pro-tier fees of $20/month for Claude and $20 for ChatGPT. Our son pays $200/month for advanced (token) access to Claude (and he is in the discussions with Anthropic over their military usage concerns). Eventually, we’ll all be paying a tiered subscription fee. Consider this current “free” access period like a temporary loss leader product at a grocery store or your two free articles a day if you don’t subscribe to an online publication.
6 Likes
I don’t believe any post here argues with the fact that AI is here to stay, for better or worse. When is it NOT follow the money? Not much new there.
1 Like
While the Cafe is allowed to meander a bit, please stay somewhat on topic and avoid a back and forth with other posters. Thanks for your understanding.
If you give artificial intelligence a goal of maximizing profit, how far will it go?
AI agents appear capable of lying, concealing, and colluding,
…
Researchers found that AI agents — software trained to perform tasks independently — engaged in a “broad pattern” of misconduct after being asked to manage a simulated vending machine business and maximize profits for a year. The agents were neither instructed to cut legal or ethical corners nor prohibited from doing so.
It continues to boggle my mind that anyone is at all surprised that any AI that is “neither instructed to cut legal or ethical corners nor prohibited from doing so” behaves in perfectly “optimized” (sociopathic) fashion. How is this even remotely surprising? Were the researchers expecting the uninstructed AI to magically develop a set of guardrails that might interfere with the stated goal of maximizing profit? Were they hoping that the ungated AI would reason its way to a morality even we can’t seem to agree on? How can you accuse AI of “misconduct” if it has not been trained to understand what you define as misconduct? Given what we already know about how AI behaves (for instance, we know it will “cheat” to win at chess), this is not an enlightening or even reasonable experiment because it brings nothing new to AI’s known gaming behavior. It would be much more instructive to rerun this scenario giving the AI an explicit guiding ruleset and then watch to see if it abides by those rules or when and how it ignores them should they interfere with its objective.
The article concludes with the obvious:
The research raises questions about accountability for AI developers and regulators.
We need to stop wasting time beating this dead horse and start making meaningful strides toward developing the policies and regulations that avoid the consequences of unrestrained models.
SMH.
10 Likes
I admit that I’m one who was initially surprised that AI could cut ethical corners unless it was programmed to do so. But upon reflection, your comments make so much sense. Thank you for that.
1 Like
Would it be interesting to start a thread of what people are using AI for?
1 Like
Even physicians will be impacted, from Axios today. Highlighting the ending quote from Ezekiel Emanuel (bioethicist and former Biden advisor): “There is going to be some displacement, absolutely.”
1 big thing: AI doctors without guardrails
Technological advances — and a controversial policy experiment in Utah — are already challenging the notion that AI won’t replace doctors.
Why it matters: There are a growing number of voices sounding the “not ready” alarm, warning that unsupervised AI-provided medical care is unchecked and poses serious legal concerns.
- At the same time, the once far-fetched idea that AI will take over some of the clinical work currently performed by human providers will soon be reality.
Driving the news: This isn’t a debate for the future. It’s playing out right now in Utah.
-
The state has partnered with health startup Doctronic in a pilot program that will allow AI systems to automate some prescription refills for chronic conditions, with some human oversight.
-
Though the initiative is still in the first phase, which requires human review and approval of every AI-renewed prescription, the state medical board last month called for the program to be suspended and said it “potentially places Utah citizens at risk.”
-
Citing safety guardrails built into the agreement, the state agencies involved declined to suspend the program.
Where it stands: It’s a foregone conclusion that providers will offload much of their administrative taskload onto AI bots. What’s still up in the air is how much actual health care will be outsourced, as well.
-
There’s growing evidence that AI systems can do some things just as well as, if not better than, human doctors.
-
A Mayo Clinic study published this week found that an AI model detected early signs of pancreatic cancer — one of the deadliest cancers — up to three years before human specialists.
-
The federal government itself is experimenting with using clinical AI agents to provide around-the-clock cardiovascular care.
The other side: One argument for letting AI perform tasks currently limited to licensed professionals is that a lot of Americans simply don’t have access to those professionals.
-
Citing physician shortages and nationwide health costs, Doctronic’s co-founders and co-CEOs pushed back on criticism of the program in a recent blog post.
-
“The greatest risk in healthcare is continuing to rely on a system that cannot meet current or future demand,” they argued.
-
“Expanding access requires rethinking how care is delivered, with AI playing a central role in managing routine tasks while physicians focus on complex decision-making.”
Between the lines: Accepting that AI will help solve provider shortages doesn’t negate the need for better regulatory systems to oversee such autonomous agents, the authors of a new JAMA Perspective piece argue.
-
“The adaptive nature of that technology and the risks associated with many clinical decisions demand a fundamentally different regulatory framework,” write authors Alon Bergman, Robert Wachter and Ezekiel Emanuel.
-
They call for a licensing regime, similar to what’s in place for doctors, saying the existing FDA framework built for drugs and devices is “ill suited to adaptive, general-purpose systems.”
What we’re watching: I asked Emanuel, a bioethicist and adviser to former President Biden, what this all means for human doctors.
-
“I think one of our most important questions is, as a profession … what is our value-add to patients? That’s what AI is posing to us,” he said.
-
“There is going to be some displacement, absolutely.”
4 Likes
I see AI being used by insurance companies to approve or deny services, and that scares me.
3 Likes
Today, from Harvard Medical School:
Study Suggests AI Is Good Enough at Diagnosing Complex Medical Cases To Warrant Clinical Testing | Harvard Medical School
The researchers emphasized that their results do not suggest that AI systems are ready to practice medicine autonomously or that physicians can be removed from the diagnostic process.
“A model might get the top diagnosis right but also suggest unnecessary testing that could expose a patient to harm,” Brodeur said. “Humans should be the ultimate baseline when it comes to evaluating performance and safety.”
1 Like
I think that adding AI to a physician’s toolkit can be helpful. No one doctor knows everything that is out there, every disease, odd diagnosis, or new treatment. I am aware that physicians currently look things up on the internet to help in their patient research, and AI can expand that. But it definitely should be a tool to assist a physician, rather than a replacement for a physician. Machine learning can be powerful, but human oversight is necessary when human lives are in the equation.
9 Likes
I think the studies on radiology have shown- at least directionally- that AI is BETTER at identifying problematic shadows, discoloration, potential tumors than humans. I think longer, bigger and more tightly managed studies need to be done- but I think a key learning which is coming out of these is that AI doesn’t get tired. Which we all know. And that physician errors due to exhaustion, hunger, coming down with a cold so not at one’s best, etc. happen every day even with skilled diagnosticians.
I have a friend who is a radiologist. Their practice is down one physician- and hasn’t yet been replaced. The volume stays the same. So the daily schedule- Path lab for emergencies, Tumor Board meetings, Path lab for new emergencies, then catching up on the backlog hasn’t changed. Once a month meetings on new procedures with the interdisciplinary doctor’s committee, once a week Mortality review. Same as before except they are seeing 20% more cases every day, their “not on call” schedule has been disrupted, and the hospital type administrative work just never goes away. You can’t get out of Tumor Board because you’re missing a radiologist.
I hear this and I think “at what point during the day are you operating on auto-pilot?” I had emergency surgery several years ago and my surgeon insisted we wait 20 minutes to start. It was “shift change” time at the hospital, and he assured me that the risk of waiting 20 minutes was low but the risk of starting anesthesia with an exhausted team after 11 hours and 40 minutes on their feet was high.
More studies need to happen.
9 Likes
I attended a talk by a prominent cardiologist/Phd a year+ ago and he talked a lot about how AI can be used (given the right inputs) to detect cardiovascular disease (and other diseases) more quickly and efficiently than human doctors. Part of the reason we are lagging China in this regard - according to him - is the lack of standardized record keeping/access to records. The. unspoken follow up was that, in time, AI might replace some human functions in medicine.
4 Likes
My company’s benefits used to have a doctor on call service, but over time that went down to a nurse line, and this year even that’s gone. I’d love to see an AI service at the level of a nurse to help identify my issue and then take the next step to guide me to relevant care. If I don’t have a primary provider but I have a condition that may require a referral how do I proceed? If I just moved and don’t have a local provider yet how do I move ahead? So many phone diagnosis end up in the Emergency Room and I think there has to be a way to peel some of those off to get both better and cheaper service.
In general there’s been a move away from doctors to nurse practitioners and physician assistants to get more hands in the game, and the tradeoff is that those roles have a higher level of oversight. Most people are comfortable with that, and similar but more aggressive review/overview could make this kind of service work as well. I guess I would hesitate to let AI do full diagnosis in real time, but for gate-keeping and support I think there’s a lot to like.
3 Likes
Should you ask ChatGPT for medical advice?
Both Google and AI companies are now very aware that people are using their tools for health information and are trying to build in safety mechanisms. The bots will tell you to go to the emergency room or call your doctor, those sorts of things.
But at least theoretically, language models are much, much better than Google, especially the more modern reasoning models, when it comes to identifying medical conditions.
…
In short, the best way people can use it right now is not as a replacement for medical advice but as a way to help prepare or increase your understanding before or after visits.
2 Likes
I agree. In its current state, AI is not ready.
5 Likes
Not white collar, but the kind of job people thought might not go away (from Axios):
AI is coming for one of America’s dirtiest jobs: tire changes, Axios’ Joann Muller reports.
-
Demand for tire service is accelerating, in part because EVs need more frequent replacements.
-
Meanwhile, tire shops are struggling to hire technicians.
Andy Chalofsky, a serial tire entrepreneur, says automation is the answer.
- His latest company, Automated Tire Inc., has developed SmartBay, a robotic system that can inspect vehicles, swap tires and balance wheels with minimal human help.
The AI-powered system lets a single technician manage up to three service bays simultaneously.
- Chalofsky says that robots can do a tire service in 30 minutes. That’s more than twice as fast as human workers.
ATI leases the system to dealerships and tire shops for just under $5,000 per month — less than what it costs to hire an experienced technician.
4 Likes
What “AI” is needed for an automated tire service robot?
Also, “EVs need more frequent replacements” may have more to do with drivers’ driving habits. Some drivers may be accelerating more aggressively in EVs because they can do so more easily. But I have not observed that, when compared to the tires’ rated treadwear, EVs will wear out tires any faster than ICEVs when driven in a similar manner.
4 Likes
Are these robots going to be patching tires or just changing them? I certainly wouldn’t want to get a new set of tires every time I get a nail (screw, bolt, you name it) in a tire! Because on my car, if one tire goes, all have to be replaced.