Think Differently. Solve the Impossible.

My guest today is Dr. Christian Yeasted. Dr. Christian Yeasted is Chair of Ethics at St. Elizabeth Hospital in Youngstown, Ohio. After seven years of researching the ethics of artificial intelligence, He wrote a science fiction novel titled, “The Final Invention.”

Did you know - there in addition to medical ethics, there is a type of ethics called Bioethics? According to John Hopkin’s Institute, new medicines, biomedical procedures, and ways of altering plants and animals are bringing benefits to millions of people. However, these same innovations have the potential to bring harms and raise ethical questions about their appropriate use.

The characters in this story experience strange malfunctions of technology that were previously thought to be safe. The result is destruction and chaos. At a construction site something goes wrong with the AI foreman, killing human workers. The characters must discover the source of the malfunctions before they turn more deadly.

Dr Yeasted maintains a clinical practice where he instructs medical students and residents in primary care.

In this conversation we talk about:

• The ethics of artificial intelligence in health care
• Bioethics and medical ethics
• How AI might be used to determine allocation of scarce health care resources such as organ donation and pandemic care
• Life extension: determining when someone should be allowed to live or die
• How to know if AI is being used at your hospital or by your doctor
• Potential unintended consequences of AI in health careHow
• How AI can improve health care and be a useful tool for doctors and patients.

Show Notes

Get the Book: The Final Invention: The Ethics of AI in a Near Future

Kindle Edition

Audible Audiobook – Unabridged

Connect with Dr Yeasted:

Website: authorcyeasted.com

Transcript

My guest today is Dr.Christian Yeasted.

Dr. Yeasted is chair of ethics at St. at Elizabeth Hospital in Youngstown, Ohio.

After seven years of research in the ethics of artificial intelligence, he wrote a science fiction novel titled The Final Invention.

The characters in the story experienced strange malfunctions of technology that were previously thought to be safe.

The result is destruction and chaos.

At a construction site, for example, something goes wrong with the AI foreman.

I think this is a robot-like character killing human workers.

Characters must then discover the source of the malfunctions before they turn more deadly.

Dr. Yeasted maintains a clinical practice where he instructs medical students and residents in primary care.

Welcome, Dr.

Yeasted.

Hi, Daniel.

Thank you for having me.

So to start with, I'm an engineer.

I know a little bit about artificial intelligence in terms of pattern recognition and algorithm.

I had no idea it was this specific discipline called bioethics.

And for listeners, real quick, I'm going to, I'm going to...

When I Googled it, I found there's literally this covers stuff like organ donation.

Maybe AI decides you should get one or not.

And maybe that's a good way of figuring out who it will work with and who it won't.

But we're talking about cloning, gene therapy, life extension.

Should you live or die?

Genetic engineering, human experimentation even.

What if the AI wants to decide to do a surgical procedure on you that hasn't been done before, but decides this is the thing to do?

Am I going in the right direction with this doctor?

In terms of explaining what I've...

So there's a lot of moral issues here.

And I'm not sure if even doctors are the people to decide this, but it's the experts, of course, that make decisions for us.

It's the Supreme Court that decides is a law legal or not.

So the doctors are involved in this.

And now I'm going to let the doctor explain from start what in a doctor's viewpoint is AI and ethics.

Go for it.

Yes, thank you.

So I appreciate the introduction.

So bioethics, we'll start with that.

Long history from Hippocrates and Galen using leeches and whatnot, all the way through the Nuremberg Trials, which is where bioethics really got its start.

We started to recognize, okay, well, we can't do trials on people without their explicit consent.

We started, and then there was a bioethics book that came out, that laid down all the foundations of what we think of as modern bioethics.

Bioethics and medical ethics are a little interchangeable, but bioethics includes pretty much the ethics of dealing with all life forms.

Could even include plants in there.

Some people have written books about bioethics regarding plants and stuff.

Genetic engineering, right?

Genetic engineering, is that what you're speaking of?

Well, the genetic engineering hadn't really been conceived of fully.

At the time this book was written, it was more like, it was a bit of a side book, but like how do we treat humans, animals, plants, everybody?

Everything that's alive.

Then now we come to today where ethics is completely taken off bioethics as well, because of all the new emerging technologies.

So artificial intelligence is one of those technologies and we are trying in the healthcare field to keep up and hopefully stay a little bit ahead of the new AI technology that's come around.

So I like to think of artificial intelligence as a set of tools, and a lot of this you know, I'll say it for the listeners.

You have your machine learning, deep learning, and out lately, you have this generative AI.

That's been what a lot of the buzz and where a lot more of the ethical concerns have come around.

I just learned that word the other day.

I didn't know there was an AGI versus an AI.

An AGI for those who don't know, is the super AI, one that actually can rationalize versus a basic AI, which will just recognize a pattern of a broken bone.

For example, AGI can actually truly think, is that correct?

Yes, and not only think, but if it's artificial general intelligence, the idea is that it might actually be able to think a lot better than humans.

We do have to keep in mind though, that there's thinking and there's wisdom.

We like to imagine that humans have both, but we have more wisdom than machines are going to have for a while because machines are very good at computing and they are very, like you said, good at pattern recognition.

But the two main types, they've got the symbolic AI, where it's more like an algorithmic thought process and an if-then.

And then you've got the deep learning full of neural networks and that's mostly input, output.

And unfortunately, it also means garbage in, garbage out.

I had thought that AGI, we already have AI.

I thought AGI wasn't quite yet there, that we had a lot of good AI applications, check GBT, but nothing AGI they could truly think.

Is that, am I incorrect?

No, yeah, that's what the overall consensus is.

Artificial general intelligence, AGI, is not officially here yet.

However, it's coming, it's coming.

That's the story in your book, and that's where we're going with this conversation, right?

If it comes, what could go wrong and how do we prevent it?

Yeah, so I'll start with talking about some of the things that we are seeing in the healthcare field with what's called narrow AI, that's good at certain tasks, like you said, and then what the concerns are gonna be, if, you know, as these narrow AIs start to each get used and synthesized together to make basically a general artificial intelligence.

So, with the narrow AIs, so you've got those that are used in patent recognition for radiologists or histopathologists looking through the microscopes, and they're very good at recognizing cancers.

They're very good at recognizing what's on the x-rays and CT scans, etc.

And as of right now, a radiologist or a histopathologist working with the visual assistants, the combination seems to do better than just the visual assistants on their own.

Just the AI application on its own.

Yes.

You have a human shadowing the doctor, making sure things look valid.

Yes.

And then you've got artificial intelligence also used for work.

It's a little more boring, but it's meant for enhancing workflow.

So you've got the patient who checks in, and the nurse takes them back and does the vitals and all that.

But then there's also, even before that, the patient often will sign in, make an appointment for themselves just over the charts online now, and enter in information.

And just the AI is brought in to help with streamlining that workflow process.

And also when it comes to the hospital, patient registration and being taken back to where they need to go.

Is that kind of like now where you get a recording?

Well, it's not a recording.

It's push one for this, push two for this, push, and none of them are what you want.

But some of that's AI driven nonetheless, and it's not really great because there's no option you're looking for.

Is that what you're speaking of?

That's part of it.

That's what we used to think is more like brute force, you know, pre-programmed responses.

The AI is helpful even in just overseeing the other machines that are answering the phones, and like just saying like, okay, here's where all of your issues are, here's where the holdups are, this is how we can, so it's all pattern recognition, and trying to figure out, okay, I see the patterns of where you're getting held up here in the patient flow process.

Let me refine that.

I just thought of a good one.

I don't think it does this, but this is what I would expect to do.

It's like calling 911, and you're afraid to call 911.

You should have called 911.

Instead, you're going through this push the button thing.

It wouldn't be great if that system could go, I'm going to direct you to someone who can help you.

It's 911, or something like that, but it's not doing that.

Yeah, and you may be onto something there.

I bet you that with better decision-making tools, they could better lead you where you need to go.

Yeah.

And then we've got AI.

I won't be able to get all the different fields that AI is involved in, in the healthcare field, I can tell you, but one of the things we're working on right now is that there's ambient listening.

So, a doctor is talking to a patient, and instead of the doctor, you know, turn to the computer and just staring at the computer the whole time, the idea is the doctor is supposed to be able to just sit and talk to the patient while the ambient listening is running.

And when you're done...

Are you speaking of...

I notice this sometimes when...

And the dentist does it too.

They'll give notes to the assistant.

Is that what you're speaking of?

Yeah, it'd be like a virtual assistant.

Yeah, and I'm kind of like, wow, they're really keeping track of what I said, what's happening.

So if I need to do follow up, they know that does seem like a task that could be done in the background by something else.

Absolutely.

Well, now, with generative AI, ChatGPT, that type of technology, yeah, now it can take all of the text that it's recorded and consolidate it all into just a simple note.

And the doctor is supposed to go and check that note.

That's where we get into a bit of the questionable.

I was about to ask, where's the good stuff?

Where's the chaos and destruction in all this?

I'm not seeing it yet.

So, I think the chaos...

Well, I'll mention one more example of how area AI is used.

And we kind of talked about it.

So, a patient assistant, so a digital assistant, it's kind of like a chat bot, but it's meant to direct the patients to a certain place.

So, some healthcare systems are using those.

And then there's what's called clinical decision support tools.

So, long story short, it will give you a list of possible diagnoses that might be going on with the patient.

So, maybe you thought of diagnoses 1 through 11, but you didn't think of that 12th one.

And AI did.

That sounds helpful.

Because you can't remember everything as a doctor or rare diseases, rare diagnoses.

Yeah, so the chaos would come from if AI is allowed to make the actual decision on the health care.

And this is something that, you know, health care systems are actively working to avoid.

Sounds like a great idea.

It seems like a great idea to help the doctor, don't be the doctor.

I think so.

At this point in time.

I'll tell them a little more about that, and I have a little ethical question to bring up.

I mean, yeah, so as of right now, humans have the responsibility to take care of the patient.

And your provider is supposed to know you as a patient and care about you a whole lot more than a machine ever could.

So it's our responsibility to take care of you.

And we want that responsibility.

In fact, if we see that responsibility to the machine, then you ended up with a few problems.

Well, one is there's always the issue with the black box.

Sometimes, often we can't really tell why a machine is making the decision that it's making.

And if you can't tell why it shows to give a certain treatment to a patient, well, then maybe that makes you not trust it as much.

Isn't there a way, wouldn't there be a way for the designer of such an application to give you as a doctor at any time you could push a button and say, tell me why, computer, before I say yes or no, kind of like you go to the post office, are you shipping anything dangerous?

And it makes you say yes or no.

Before this AGI continues, there should be stop gaps, right?

Where it says, if a doctor wants to say pause or tell me why I want to learn mobile, why go continue.

Maybe it sounds like it's not doing that right now.

Not right now, as far as I know, there may be some apps out there that might give you some explainer.

But that is definitely one of the things that the companies are working toward, is creating explainable AI.

So one thing that quickly came to my mind, I'm not an AI expert.

But from what I know, for example, with copyrights with AI, the people trying, like Nashville, the Songwriters Association, they're suing some of the AI companies because they're using copyrighted stuff.

They got off the Internet.

But as a lawyer, you have to know, okay, what stuff and where, and they don't keep track of that.

Meaning, it could be possible with this algorithm in the hospital.

This algorithm actually can't keep track and just tell you, oh, by the way, blah, blah, blah, because it's a million computations it just did statistically, and however, it's done, there is no simple, this is why flowchart.

It would, is that kind of correct?

Yeah, that is one of the definite risks.

I think well said.

It might not even be able to tell you why.

Yeah, figure out why it provided the answer that it did, because it doesn't really know why it's doing anything.

That's one of the issues with the AI is it doesn't understand anything.

It's simply input output and programmed to perform a task.

There are a lot of reasons for the why, millions or billions.

What it can't do is like chat GBT or maybe it could if we tried.

Okay, computer, take that billion different ideas you thought of before you gave me the why, and solid it into something I can understand in one sentence because I'm operating and I need to know now in one sentence.

Then I will make a decision yes or no on the checkbox and take the risk of continuing.

I like that future scenario there.

Yeah, as long as it's still acting as your assistant and not being the doctor.

At the risk of sounding like I'm just trying to protect my job.

Otherwise, I'll unplug it.

Like when my computer is not working and I unplug it.

But this happened with my emergency device in the backcountry recently.

I didn't need a rescue.

I pushed the button accidentally.

It was in my backpack.

And I just, what do you do?

I'll just take the battery out, turn it off.

No, it'll tell you SOS in progress.

You can not shut power off until SOS is canceled.

But there's no satellite, so I can't do it.

Anywho, unplugging the computer might not work.

Right.

And you're on the surgery table or in this is mission critical.

Okay.

You know, that actually just brings me to a really good ethics plug here.

In order for it to be ethical, to use a tool in the health care field, at least, it has to be a good tool that works really well.

And if it's a crappy tool, then it's probably unethical to use it.

So that's number one clearance criterion for do we use the, is it ethical to use this new piece of technology?

Let me ask you this then, how if I'm a rocket scientist, we just launch another rocket.

If it fails, it's a couple million dollars.

We just eat the money.

These are people's lives though.

So you can't experiment with them the same way.

How does that work?

Like who volunteers?

Do you volunteer to be the first person to have a procedure in the hospital with some of this new technology?

Yeah, oh, that is such a deep question.

Well, let me say that we would like to think that the technology that's used on the patient population has been vetted rigorously and that it has undergone every stage of trial possible.

I don't think we can say that about all the medical technology that's used.

But yes, people often will volunteer to be the first to either try a medication, and use these clinical trials, or they will be the first to try some kind of new technology.

But I'd have to look into this, but I believe there may be something where with technology, as long as it resembles a piece of technology that's already been cleared by the FDA, that is acceptable.

So it has to be just similar enough to...

And then, this may not be the case anymore, but as I understand it, a piece of technology could actually be pulled off the shelf.

But that doesn't matter if it was originally cleared by the FDA, you can still use your new technology, even if the old technology that you're similar to was found to be unsafe.

So day-to-day, day-to-day in your hospital, do you get a choice as a doctor, or is it the hospital administration that says, you're going to use this or you're not?

As far as the doctor using the technology?

Yes.

If you decide, you know what I did my research in, I don't like this anymore, I'm not going to use it.

But then you have procedures and cost associated with those procedures, and then you have insurance who probably wants you to use cheaper procedures.

Are your hands ever tied where, you know what, I don't like this new technology, I don't want to use it, and you have to?

Yes.

Maybe not every time.

But there are certainly scenarios where, so you use the tools that are made available to you through the healthcare system, the hospital and the offices and what have you, that are part of the bigger healthcare system.

The healthcare system often is in league with technology companies or software companies, and in the case of AI, and I guess we could dive into this a little bit, and then if we need to come back to the previous topics, we can.

The medical legal part of this is quite fascinating.

There's three major players, the provider, the healthcare system, and the software company.

These are the responsible parties for using said tools.

AI, lipstick AI, yep.

So if you are the soft, so these are the three parties that can get sued basically.

Therefore, they're the ones that are responsible, or vice versa.

So if you're the software company, you should be held to, and these are where the solutions come in, because we like solutions here, right?

You should be creating AI in an ethical way, and that usually means make it transparent, make it explainable, as we talked about, and you want to make it equitable, make sure that it's good for all of humanity, and it's not going to divide the rich and poor gap even more.

Safe, and they're actually like 23 or 27 Asilomar principles.

You go through this every day in your ethics committee at the hospital.

This sounds like a full-time job by itself.

It can be.

So the software company has to produce it in an ethical way.

Whether they do or not, I don't know, but you can get auditors to go and make sure that it's being created in an ethical way.

Then, how would they not?

What's an example of a software company not doing something ethically?

Because I'm a good guy.

I'm a good person.

I can't even think like what would be bad or good.

Just do the right thing.

Give me an example that is more common in your hospital right now, where not your hospital, but hospitals in general where software company might try to do something non-ethically.

Well, I can tell you, well, first of all, pay someone enough and you'll see someone do unethical things right away.

But so obviously there's a lot of money to be the company that produces the next big AI technology that is going to look appealing to the healthcare system.

And there's a lot of examples of companies getting in trouble because they promised AI that they didn't actually have.

Maybe what they had was some older brute force programming or something, and they advertised it as a false advertising.

So another way that it could be unethical is if it's just, like I said, a crappy tool, if it just doesn't work well, if it's going to put people in danger because it's unreliable.

So it's not robust.

So kind of like your automobiles where they know they should have not let them out and they have to recall them, and they should never let them off the assembly line with the defects because cars are dangerous.

That's mentioned in the book.

Basically, what kind of beta testing was done.

Basically, people are the guinea pigs out there, and you end up with a woman getting run over in Tampa, Arizona.

Okay.

This is great because we're talking about AI, which I just assumed was so intelligent and supercomputer.

We didn't need to worry about this, but in fact, we do.

Yeah.

It's not to the point where we can take our hands off of the tool you get.

Absolutely.

It's because the software companies are trying to make it seem better than more advanced technological advanced than it is.

Often.

Then also, even if they presented as advertised, maybe car companies, health care systems, whatever, they might just try to use it.

Because it seems like it's a better option, that people will buy it, that they're going to be excited about it.

It's a marketing thing.

But yeah, maybe it wasn't ready to be released to the public yet.

If I'm going to the hospital, fortunately, I don't have to go very often.

What are some more common, because this sounds kind of negative, kind of really concerning.

I'd be worried right now.

But what?

Nothing's perfect, right?

Nothing's perfect.

Every tool needs to be sharpened or dulled when it first comes out.

Give me some examples of real stuff that could happen or almost happens.

Kind of like the pilot.

None of the passengers knew that the plane, one of the engines went out, because the pilot made a quick switch.

And then like, okay, that engine, not good on this plane.

Give me some examples, like actual real examples, with this AI again, if you can, in the hospital where people might want to really ask more questions or know, hey, is my AI being used at this hospital on me?

Yeah, that's a very good point to address.

I can tell you that I've never heard anyone ask yet, but that is a big topic where we don't want to just incorporate AI without telling the patients.

Now, granted, there may be AI in, like I said, managing workflow and all that.

I mean, not something that you need to necessarily tell patients about.

But if a doctor is going to use an AI clinical assistant for decision-making, then that would be a legitimate question, how much of that should you reveal to the patient?

The answer is, of course, in full disclosure, you reveal as much as you can.

You give a disclaimer, this is something that some healthcare systems, even maybe some offices are thinking about doing.

You say, hey, we have no choice, though.

I'm injured, I broke my bone, broke my hip, fell down, I've got a concussion.

It's not as if I'm going to just sign all the paperwork.

Sure.

Fix me.

Yeah.

Yeah.

I mean, if anybody even reads it, but it would probably be just like a paragraph.

Yeah.

I don't have time.

Okay.

Well, so if you're describing an emergency, the ethics change a little bit.

It would say, I think what's ethical is to treat as fast as you can with what you can, do what you can emergently.

Obviously, you still have good intentions and you try to do the best job you can, still do no harm, do good, etc.

But it's a little more like whatever happens, happens and you did the best you could because it's such a crazy, a chaotic time.

But for patients who are a full frame of mind and they come in and they're wondering how much AI is being used in the doctor's care of me, yeah, there should be a disclaimer saying this is how much AI we use, this is what we use it for.

Sign here and people will because the younger generations is going to be used to it.

I mean, they're really counting on you.

It'd be like I'm flying the airplane, I'm the pilot, there's a checklist.

It'd be like me giving it to the passenger in the front row going, are you okay with this?

How am I supposed to know?

You're the pilot.

Yes, if you are, I am.

If you're the doctor, I would look you in the eyes and say, can I trust this guy or this man or this woman?

That's how I would go with it.

Yeah.

I'd imagine the scenario would be, you're the patient, you're looking at me dead in the eyes saying, hey, you use AI in my care, can I trust that you're going to be giving me the best care?

If I'm being honest, my answer would be, as of right now, yes, because it's going to give me suggestions, but I still make the ultimate decision or rather, you still make the ultimate decision because we have a discussion here and we respect autonomy in modern bioethics.

I'm starting to understand now why you're the chair of ethics at the hospital and as a group of doctors, you decide these ethics versus it being a bigger moral issue and the general public or someone else working on it.

Because when we go to the hospital, we're in your hands.

You have to decide ethically.

You, this is what you do then, right?

You're working on ethics as the patients distrust you.

Well, we try to be worthy of that trust.

Yeah.

What we don't want is to get to the point where artificial general intelligence has been created and implemented, and it makes better decisions.

Well, it makes the decisions and the humans don't have any input because I guess now's a good time to talk about it.

Yes.

Problem.

Because your research after seven years, this is where your research led you.

You're worried about the future.

Right now, things aren't so, you don't have to be so concerned to go to the hospital, but we're talking about the future, right?

Yeah.

I don't think we want a future where machines are making the ultimate decision and have ultimate responsibility over humans.

I think one of the main reasons why is humans should be responsible for humans.

If you say, well, but if an AI can think better than a human, act better, do better, treat me more accurately, give me better results, why wouldn't I do an AI?

Well, that is one way to look at it.

I think a better way is when AI is used as a very, very good tool by a human who has ultimate responsibility and understands what the patient is going through, the emotional suffering, the pain, the maybe why the patient's smoking or drinking, doing harm to themselves somehow.

These are the things that AI will never understand because it's never gone through any of that.

It doesn't even understand the deeper level of the human condition and the suffering that we've been through over millions of years, etc.

I mean, there is no compassion that is possible with the machine, and it's essentially a completely different species.

So would you want a completely separate, disinterested, different species to assume your ultimate care?

That's a good way of putting it because if I was a machine, I don't want a human working on me.

That's in their best interests.

I want a machine to work on me as a machine.

That's in my best interest.

Yes, machines taking care of machines is one possible future we have.

The other thing, here's an example that maybe is another part of the issue.

Say AI is giving a beneficial goal for the environment, and it decides to accomplish that goal.

I actually have to get rid of the humans, and you didn't see that one coming.

Or even if it's biodiversity, like I want to, I want the AI to take care of protecting the bald eagle.

And then it decides, well, there's some fishermen and they have a little house here on the dock by the lake, and it's not in the zone.

I'm gonna have to take it out.

Kill some humans, kind of like what sounds like what happens in your book.

So talk about that.

That's what you're leading to, right?

In the hospital setting though now.

Yeah, and I think there's only one main way to prevent that.

Prevent the paper clip scenario.

I mentioned that in the book, but Nick Bostrom is often credited with saying, you give a machine with ultimate power a task and say, make as many paper clips as you can, it'll look at you and turn you into a paper clip, and it'll turn everything around it into a paper clip.

So it maximizes paper clip production.

Yeah, that's the idea with an AI that goes off the rails and is not in line with our values.

So I think the only way to prevent a scenario like you're talking about taking out the guys who are in the way of the bald eagle is to make AI in line with our values from the beginning.

This is where the software companies are faced with the ethical challenge, and ought to have ethicists on their boards, not to hinder them, but to help them produce the technology in the best, most sustainable way possible, that benefits all involved, to make the best product possible.

That way, you don't end up with an AI that is putting bald eagles priorities over those of humans.

We design it so that it has the same intentions as humans, the same values, and what are those values?

Well, that's where we start to reflect on what we want as humans, what we want for ourselves now and for the future, who we are, what we really care about, and we want to build AI that cares about those same things.

When you chair of ethics at the hospital, what kind of conversations do you have every day with the other members?

Well, I'll be honest, a lot of it is end of life things.

By the way, if you don't have a living will and a power of healthcare, a power of attorney, you should get them.

And if not, what will AI do if you don't have power of attorney and a living will, explain what a living will is to those who might not know.

Well, I'm glad you asked that.

No, I gave like five living wills out to my patients today.

Basically, living will is...

You know, they've got the checkbox, hey, I don't want to be a vegetable.

I don't want to be kept alive as a vegetable.

So you can check that box and say it's okay to pull the plug if you're vegetable.

And then you've got your living, or your healthcare power of attorney, where you write down the name of the person who's going to make medical decisions for you if you're unconscious, essentially.

These are papers that, at least in Ohio, you get notarized, give them to your doctor, you scan in to the chart, and then the family members can't squabble as easily for what to do with you if you're laying there as a vegetable, because you've got it on paper and writing.

You make the decision before you get there.

Now what if you don't have a living, Will?

Well, you know, at least in Ohio, the hierarchy goes, if you're not able to make the decision, it goes to next of kin.

And then eventually when it goes down the line, it's, if you can't find it in next of kin, it's providers.

And then sometimes they'll apply for guardianship from the courts.

But would AI fit into this?

Is AI fitting into this now?

Well, I would imagine that if we were to use AI in any part of this process, it would be education for the patient on the importance of having a living, Will, and in writing.

It would be, you know, talking about what, just getting information about, hey, what are your values?

What do you do?

And probably better to have these questions come from a human to a human.

But a machine can prompt those questions, you know.

When would you say that you no longer have a good quality of life, you know?

Or what about if the AI had access to where AI gets data to make these decisions, it gets some, it learns about your past history.

It takes into account your health, but it kind of is trying to rationalize a bit where another person, should I leave this guy a vegetable or pull the plug?

And I'm thinking to myself, if I say this, would go scrape the internet.

And then how would it do that?

Come back with the decision?

How would it possibly go?

No, you can't pull a plug.

That's killing somebody because I read that on the internet versus having a like a human would do rationalizing.

You know what?

But it's pointless.

The person's, he can't, he's in a coma.

He's, she's in a coma.

Probably will never come out.

Has been that way for a year.

It's the rational thing.

Is, is, am I, am I on the right track here?

How would it figure out?

Oh, would then take all the data on your Facebook channel and maybe go, what kind of life has the guy or the woman be living?

Is it a, is it a happy life?

Is it productive life?

Has she had thoughts of committing suicide?

Am I doing them a favor?

All that, right?

Well, so these are all factors.

Well, there's almost all of them are factors to consider.

Actually, in bioethics, we don't look into what kind of life the person lives if we can avoid it.

That would be way down the line because we treat drug dealers the same way we would treat saints.

It's pretty judgmental.

Yeah, we want to avoid those types of human input judgments there.

Yeah, so it actually, essentially, it really...

But your AGI is a judgmental character.

It's supposed to be so advanced technologically, it can actually make judgments, right?

Eventually, yeah.

I mean, an AGI would be able to assimilate all kinds of information from various fields and hopefully synthesize it all into, well, a decision, maybe.

As long as it didn't explode in the process, like in Star Trek.

But the two things I would point out, I think the last place in the world we would want to use AIs deciding whether to let someone live or die, hold a plug or not, because that is definitely something that is difficult and requires whole committees of ethicists to step in and help out with.

Whole families coming together, having big discussions, a lot of communication, a lot of family meetings.

It's the biggest decision ever potentially, and that's the last thing where AIs should even get involved.

Because compassion is so important there.

What does something else common?

Another one on my shortlist is organ donation.

Who should get an organ and who should not?

What are some of the things you actively discuss in your group in the Ethics Committee at the hospital in regard to AI?

Yeah, I will say that at our hospital, as in a lot of other health care facilities, there are independent companies that are, that take care of all that basically.

So they will come in just at the right time when a patient is, maybe just, was just intubated and maybe found to be brain dead or something, but they still have good usable organs.

And these companies know that they've got their whole list of criteria, and it might be a lot of data that they're using, but they've got it pretty well established already which organs are viable, which are okay to use, which are not.

And again, they send in their humans who are trained to be super compassionate, empathetic, because this is the hardest point of a lot of people's, a lot of family members' lives.

And yeah, the role of AI, I don't even think AI is necessary for that because they know which organs are good already, and they need people with eye-to-eye contact, being compassionate, hand on the shoulder, that kind of thing.

So again, that's probably nowhere close to what AI is going to be used for.

Now, if you're talking using algorithms, as far as who gets what, well, this is something worth discussing.

So when COVID came around, for example, and we were running out of ICU beds, and running out of short of supplies in general, we had to try and figure out how to allocate all the resources.

Allocation, yes, allocation.

So certainly we didn't use computer programs to do so.

So we set up, through our discussion, okay, we're going to use this research-based format, it's called the SOFA or MSOFA.

It's like, how likely is your mortality based on all the organ systems that are failing, your lab results, et cetera?

And we use a lot of validated tools.

Meaning, meaning, if I've only got two respirators, and I can give one to the young kid versus the older person with COPD, who's going to die anyhow, that kind of logic?

Similar.

Yes, that's, that's definitely part of it.

And you would get scores based on how close you were to death or et cetera.

And it'd be like minuses and pluses here and there.

So we did end up following an algorithm to figure out, okay, and we had it all mapped out there.

Fortunately, we didn't have to put it into full effect, but we were close.

And would an AI, the general AI be able to come up with a better algorithm?

Maybe.

I'm not sure.

We, I think we did a pretty good job.

I think the algorithm, it can, it can follow what you give it.

The issue is, it doesn't have the data follow it.

It has data, but is the data good?

How well does it know the patients and their actual state of well-being versus the doctors who have been treating them?

I think that is a brilliant point that is probably the most important point.

Remember garbage in, garbage out.

If you have data that is based on, say, small sample sizes or it's biased in any way, the AI is going to be biased as a result.

What you could end up with, with the biased algorithms, and this has been known for a few years now, which I'm very glad about.

The word is out, that AI is biased.

By biased, define the word biased to those who don't know what it means.

Yes.

I'm sure statisticians are going to just come and try to kill me if I butcher this definition.

You're an engineer, I'm a doctor.

It's okay.

Leaning toward one path over another, leaning toward one view on things over another, and the AI ends up being biased because the data it gets, it might be based on only certain populations, African Americans or Asians or what have you.

It might be biased because the sample sizes were like 20 patients doing this or that.

It might be, and here's the fun and tricky part, the fun if you're an ethosist, it might actually get data that was a large sample size, seems like everything was done right, and it might even be quote unquote true information that the AI gets.

But if it gets this information that says, okay, well, you know, African Americans are likely to have higher levels of creatinine in their blood than Caucasians.

Well, fine, except now you've got the AI dead set on, hey, this information is true as far as I know it, so I'm going to treat all African Americans like they've got higher creatinines.

So it might actually reinforce itself unless it gets other data.

Would it be correct to say the AI isn't a problem?

AI isn't biased, it's the data that's incomplete, and therefore you might say biased.

Like we only asked white people who they want to vote for the president instead of black people or something like that.

And then you have this data set that says, here's who people want to vote for.

And that's biased, but the AI algorithm then looks at the data set and calculates all the voting statistics.

That's not biased.

Yeah, I think that's fair.

But I should point out also something that you mentioned.

So there's a big issue with correlation versus causation.

That is also in the book, where AI might see that more black people, higher proportion of black people, voted for Biden than a portion of white people.

I have no idea.

But then, can you draw a conclusion from that, that, okay, Biden won because he had more black people voting for him?

Well, not necessarily.

Unfortunately, we, as humans, we're just causation machines.

We're looking for causes for everything.

One plus one doesn't equal to all time.

Yeah.

Oh yeah.

You're taking some math classes.

So I'm glad you mentioned your book again.

I didn't realize, I would admit I haven't read the book.

I don't have the book yet.

Perfect.

I'm excited to hear that there is really, that in the book, it's kind of like some of these books, you read the character, the setting, the scene, everything is actually real.

It's a story, but it's really based on, on reality, on what could happen.

What's the way to put it?

It has some factual stuff.

It's based on fact, even though it's fiction story.

Yes.

It's exactly what I was going for.

In fact, if you read something that says it happened, it did happen.

If you read about something that's like in the future, after 2024, well, it could happen.

Interesting.

There's one part where I get into, there's a chapter where I give a long explanation of what led us to this current state of affairs.

Yeah, it's all just backed by the research that I've done.

So I definitely didn't want to, I didn't throw in anything that happened before 2020 that wasn't true as far as I know.

I can't give a spoiler alert because I haven't read the book, fortunately, so I'm tight-lipped.

I hope it has a happy ending, which gets us back to, and I'm sure your book will, the characters and the story will help with this.

Tell us now, what are some of the things people can, how can we society make sure that ultimate destruction and chaos does not happen?

So, I'm happy to say that there are many organizations that, there's religious, scientific, secular, governmental, they're all coming out with guidelines, rules, executive order from Biden.

There's the call for ethical AI from Rome.

It feels like the Rome ethical call for AI.

And then there was the Asilomar Conference back, I think, in 2017.

So, all these different organizations filled with ethicists and leaders and every major Abrahamic religion with other major religions around the world, they're all coming together and they're saying a lot of the same things.

Here's what we want.

We want safe, reliable, transparent, explainable AI that we still have governance over, we can control it and it's aligned with our values and that will keep things as equal as possible for everybody, so equity.

These are like the core principles that a lot of these organizations use.

Oh, well, I mentioned safety, privacy is...

And as a kill switch, I can unplug it.

Maybe.

Yeah.

You know, honestly, that's not actually part of a lot of these principles.

Because who gets to decide, who gets to decide, the doctor or the patient can unplug it.

Yeah, there's been some AI ethicists who have presented this as a problem.

I mean, by the time we probably realize we should...

Okay, part of having control over it is probably being able to terminate it.

But it simply won't be a matter of unplugging, that's for sure.

I mean, despite what the whole cloud strike incident showed us, how easy it is to just turn something off.

Yeah, there's the fear, and what we should try to avoid is that by the time we realize we need to turn off this technology, it's already too intelligent.

It's out of the box, essentially.

It's embedded in everything we do, everything we do our way of life.

There's no simple way to turn it off because it's in...

It's like how many people have a cell phone, a computer on their watch.

There's the Apple built into your phone.

There's sensors everywhere and everything.

It's like you can't really turn it off because it's everywhere.

Yes, that is the concern.

Now, I will say that a lot of that might be human-driven, so I also mentioned this in the writing, where it was basically a competition-based.

I mean, it was organizations trying to get a leap ahead of their rival companies or one country, United States trying to outdo China, etc., that led to basically the AI arms race.

Before you know it, we had technology that was, like you said, fully embedded, and had already surpassed what we are able to control.

Back to your hospital, what are some other examples we have?

We talked about life expectancy, pull on the plug, Oregon donation, oh, the COVID one.

COVID one was a good one.

The AI is just following an algorithm now, which seems acceptable.

What are some other practical scenarios in the near future?

Well, I...

Okay, so we talked about the ambient listening, making the interaction with the doctor-patient interaction a little bit more streamlined.

And is that a bad thing?

Is that one sounds kind of okay?

Well, I was going to mention one potential problem.

The AI...

So notes are still pretty important.

I mean, to be able to keep track of what happened last time, there's so many moving parts in medicine now, you can't rely on the memory of either the provider or the patient.

So you want the note to say, okay, what was our plan five months ago and what is it now?

The problem that you could potentially run into, a lot of user error, I guess, is the AI listens to everything, comes up with a brief note.

Is that note going to incorporate all the things that you as the provider would have written in?

Is it going to talk about, well, this person has this many kids and here's what the kids are up to and all that?

That's something that providers will write in there does sometimes because it helps them feel for the patient.

Well, it's not medically relevant, is it?

So maybe the AI won't write that down.

Maybe the doctor, and this is where the real problem comes in, the press for time because they spend all the time yapping with the patient and getting it all recorded.

They look, and they're like, okay, yeah, it looks fine.

Then they move on and they're satisfied with the note as is.

The note has, however, left out some very important parts, and the doctor hasn't made known to that.

Or it's included some inaccuracies, and the doctor hasn't really spent the time to note those inaccuracies.

Maybe just one or two words that might make a big difference, like patient is taking or patient is not taking.

My phone is always, even with predictive text, typing a word I did not mean to type, and saying what I didn't mean to say.

It has good intentions, it's just not perfect, and some of them are really backwards from what I really meant to type or say.

Yeah, it's the new responsibility of the provider, and perhaps one of the most important ones, if we use this technology, is going to be to look over that note and make sure it is completely accurate, because it's going to affect the healthcare next time the patient comes in.

We don't want to be propagating false information and it's going to be like a game of telephone then.

And how long before it becomes commonplace for an AI piece of software to actually be full-time or more used more of the time that people should really be aware of this?

It's already been rolled out in many offices and healthcare facilities.

And again, maybe it won't affect patient care significantly.

Time will tell.

We'll see how these trials go.

But would we ever know if there are minor inaccuracies in some of the healthcare we give?

I don't know.

I mean, sometimes…

It will be he said, she said, because when something goes wrong as a patient, I will be the one to say, I was in there.

This is what I said.

I didn't say that.

How did they get this from those notes?

And I remember the doctor saying, my kidneys didn't need to be replaced.

And now I have an organ transplant, because those guys did their job, or AI didn't.

Well, anywho, now I am…

It's kind of like you cut the wrong foot off in surgery.

Like, where's the oops?

And that…

I mean, we tend not to think things can get that, go that awry, but that's usually when they do.

Yeah, the protection against that as it has been for hopefully, like, a number of decades is the Swiss cheese model.

At some point, it's going to hit cheese instead of just a bunch of holes.

So we hope that there's enough safeguards in place that you keep your foot and don't get the wrong kidney.

But yeah, I mean, still, but I think that the opportunities for mishap are increased if there is an inaccurate tool recording inaccurate information and the doctor doesn't check it because they're pressed for time or reliant enough on these tools because they've been performing, you know, 85 percent sure is pretty good, 90 percent good enough.

And then moving on.

So or the doctor doesn't think there's anything important that was really discussed with the patient.

They're gonna be like, I got it, I got it.

We'll refill your medicines and move on.

Yeah.

So but so they don't even read it at all.

And then totally inaccurate information was recorded.

And next thing you know, doctor goes back six months later and that's all he's got.

It's just inaccurate information that he doesn't know whether to believe or not.

So yeah, that's one way that I could see it going badly.

So there's a lot of organizations and committees, fortunately, working on this, which is great news.

Is there anything a patient or a person can do to help out the, about the process or should know about like raise their hand?

Hey, I just saw your computer write this in the notes.

That's bad.

I don't think that.

Look, you know, raise the hand.

I don't think that's what you meant to say, Doc.

Is there anything we can do besides just trustworthy?

You know, where that's what you do and you'll take care of us.

Yeah.

So fortunately, in this age where notes may be a question of accuracy, patients are able to read their own notes.

They can look at their own results.

They can access all the information that is recorded.

So it'll be a shared responsibility.

I hate to put any extra onus on the patient here, but I mean, we have to be vigilant and make sure that it's considered accurate by both parties.

So that's one thing patients can do.

I suppose you could, when you go into your doctor, say, yeah, how much AI is used?

And do you guys use any AI?

How much?

What are you using it for?

Ask questions, be curious.

And it might irritate some of the health care providers.

Like, yeah, we got to wait a little bit more time.

I don't know.

I don't know what the answer to your questions are.

Well, if they can't give you adequate answers, then you could consider, well, I'm not sure I necessarily want to be coming here then.

You know, I irritated the X.

The last time I got my foot X-rayed, I irritated the radiologist as they put the big protectors, the bibs on themselves, the lead bibs on themselves.

And they were 10, 15 feet away from the machine.

I'm right there and they decided, only my ankle needs to be X-rayed, that they didn't need to cover other body parts.

And the X-ray itself was a couple feet above my body.

And I'm thinking to myself, well, you know, there's gotta be some scattering and how come I don't have one of those?

So I asked him, like, will we do this every day?

I'm like, Dan, it's just my question.

There's some exposure.

Anywho, they got a little irritated, but I, it's my body.

So I spoke up.

I said, you know, can you put the protector on my body and look?

Sure.

Thank you.

Oh, so they did.

Good.

They did.

Cause that's reasonable then.

I know, but exactly.

You did the right thing.

You asked a question, you stood up for yourself and yeah.

And it wasn't an unreasonable request.

So yeah, I'm glad it all worked out.

You know, really, it's just common sense.

And I'm just sitting there thinking cause it's pretty important to me.

So it's just common sense.

Which is one thing that machines again, do not have not, not yet.

But yeah, like does the robot care that you are exposed to minimal amounts of radiation?

No.

Prudent avoidance is the other word I use, prudent avoidance to the operators of the machine.

When they, they looked at me like, what's the problem?

I said, what's the problem with putting it on?

So with the snow taking, I would probably say the same thing.

Well, how did you know?

Can you, if I was the patient, can I read it?

And if it says something different, speak up and say, you know, Doctor, is this what you meant to say?

Or how would someone phrase that if they did want to object, that they were being polite and not rambunctious or difficult?

If you use your pleases and thank yous, there's almost no wrong way, you know?

Trying to try not to yell and they'll eventually come around.

Please.

When they say, well, please.

Yeah, right.

Exactly.

Exactly.

Like, I insist, please.

With a smile, you know?

Yes.

But that goes for a lot of human interactions.

Let's just throw this in there.

When it's a human interacting with machine, as machines are right now, it can be polite technically, but it doesn't give you that warm smile that really makes you feel like you're talking to another person that makes you bond with the machine in any way.

That may be different in the future though.

If anything, if they're programmed correctly.

Or just repeats the same thing in a polite or different voice.

Yeah, right.

Let's talk about a couple of other things.

Definitely, it's not really application of AI, but we're using a lot of robotics in medicine now.

A lot of implants.

More and more implants are being created.

We've got implants for seizure prevention or deactivation as soon as it comes on.

We've got now implants for sleep apnea, where it'll stimulate a nerve every time you take a deep breath in.

So it has the sensor on it and has a stimulus.

And it's not how you would consider AI.

But if these systems were to be integrated, and as AI becomes more common, why wouldn't you put some deep learning applications into some of these stimulators?

But, yeah, so there's been a lawsuit already where, this is more like medical devices, I guess, but a pain pump put in too much pain medicine.

It just malfunctioned and kept on going.

That brings us back really quickly to...

That's because the sensor malfunctioned and it didn't know how much it gave or how much the patient needed?

Yeah, it was...

I'd have to look at the exact details, but I imagine a disconnect between the sensor and the dispenser for sure.

And what ends up happening there is...

I'd mentioned that they're the three parties, the provider, the healthcare system and the software company.

So if the device malfunctions, maybe that's the software company problem.

Let's bring it to AI, same deal.

If it's the hospital is the one that works with the software companies and says, okay, your technology seems safe enough, we'll use it.

And then if something goes wrong, healthcare system could be in trouble and responsible.

And then you've got your provider who used the technology.

They may also be responsible.

So here's the interesting parts.

First of all, the provider is, they don't really have a lot of safeguards.

The doctor uses whatever technology is provided to them, whatever tools are considered the best, and tries to go with the standard of care of the state or region where the doctor is working.

And if they use the tool that goes wrong or whatever, well, they could be in trouble there.

They have malpractice insurance, but that's about the only safeguard there is.

The healthcare system, they've got the risk management and they've got some safeguards to protect them.

Maybe the patient signed some waivers saying, yes, that's okay if you use this tool or this kind of hip replacement, et cetera.

But the software companies are way ahead of both the healthcare system and doctors with their safeguards to keep them out of legal trouble.

So, one example is Chat GPT.

They say explicitly on their user agreement, do not use this for anything, medical, legal, financial, judicial, whatever.

They list all these different ways that you must not use this technology.

Meanwhile, they know full well that they've put this technology into other companies that contract with the health care system.

And these companies use the Chat GPT technology for whatever tool they're peddling to the health care system.

So, Chat GPT is embedded in these AI tools that the health care system ends up using.

And they're often clinical decision support tools.

And you didn't know that?

You had no idea?

The provider didn't know that.

I'd hope that the health care system knows that they often will employ some people.

I know our health care system has a committee that is meant to work directly with the software companies and usually has a decent idea about what kind of technology is being used.

But that's a few people in a whole health care system and where it gets deployed and used by everyone.

The software companies, OpenAI is not going to be sued for something that went wrong.

It's not the chat GBT engine itself that's going to make a mistake or be cause for error.

It's the fact that it has a limited database.

The database it's getting its chat information from is only so big.

Yes, definitely the data it's pulling from is a bit of an issue.

And it is helpful.

It is.

You don't want to throw the baby out with the bathwater, so to speak.

I think that's why it's in there for a reason.

The trouble it's not perfect.

And when you don't know it's perfect, then it causes the plane to crash.

It's kind of like some of the software in the airplanes when the pilot doesn't know how the software works and he can't hold the kill switch on the software and just fly the plane, then we are in the trouble.

Yeah.

And I think we could summarize by saying, you know, the way to avoid all of the worst terrible mishaps that could come with AI, build it using, making it align with our values using the core ethical principles that have been agreed upon by all kinds of organizations all over the world and have the process overseen by auditors and ethicists, etc.

Then when it's put out, make sure that humans maintain responsibility for using it appropriately and maintain control over it.

And that's probably about the best we could hope for.

That's great.

But I appreciate you taking the time to be on my show.

And I hope the future of humanity is bright and there's a positive outcome to doctors and patients integrating AI into health care.

And I'm sure you're going to continue to work towards making sure that happens, that ethically we're in good hands when we go to hospital and get health care.

For those enjoying this podcast, his book, The Final Invention, is available on Amazon and of course wherever else you buy books, either locally or online.

Thank you, Dr.

Yeasted.

It's been a pleasure.

Really appreciate you having me, Daniel.

It's been a pleasure.

Thank you.
Edit