JOHN YANG: Artificial intelligence is finding its place in all sorts of scientific fields, and perhaps none holds more life savings promise than healthcare programs are learning to answer patient's medical questions and diagnose illnesses.

But there's still some problems to be worked out.

Earlier, I spoke with Dr. Isaac Kohane, the editor-in-chief of the New England Journal of Medicine AI, and the chair of Harvard's Department of Biomedical Informatics.

I asked him about AI's potential in medicine.

DR. ISAAC KOHANE, Editor-in-Chief, New England Journal of Medicine AI: Doctors can definitely use AI as an augmentation, so they'll remember or be reminded of all the things that they should know about their patient, their specific patient, and all other similar patients like them.

At the same time, patients are all too aware that in the United States, we don't have enough primary care doctors.

And so we have very little time with doctors if we can get any time.

And so providing another, if flawed resource to get them medical advice so they can decide whether or not to actually seek medical advice is also probably going to be transformative.

When a doctor sees me, then I can tell you they've forgotten a lot of details about me.

What if you could actually just have and it's doable today a summary that says, what are all the important things that Isaac Kohane has medically that you should know about today?

And then saying, what are all the other patients who are like Isaac Kohane?

What are the right treatments?

Is he on the right treatments?

Are there any screening tests that we should do on him today during his visit?

Medicine is getting increasingly more complex.

There are more things that we know how to do for preventive medicine, but in fact, prevention is getting less and less attention in many ways because doctors are so busy.

These AI programs are not infallible.

In fact, they can make mistakes, but at least it's a conversation you can have and then you can go decide, should I actually discuss this with my doctor?

JOHN YANG: How far off are these things?

How soon before they're realized?

ISAAC KOHANE: Right now they are highly imperfect, but they're being used today.

The reason they're being used today is because there is such a great need.

We've all heard of the so called Dr. Google.

The fact is that people use search already to obtain medical advice.

When ChatGPT was released by OpenAI this past winter, patients started using it already.

We know that for a fact.

We know that doctors started using already.

People don't realize that 30 percent of the costs of healthcare are administrative overhead, the billing procedures, getting authorization, deciding whether to do reimbursements.

They involve some judgment medical judgment about how appropriate the care is, whether patients should be allowed or not.

So that is happening right now, today.

JOHN YANG: You've said a number of times that it's flawed, it's imperfect.

What should the patient be looking for?

Are there red flags that the patient should be looking for?

ISAAC KOHANE: Sometimes it can be quite subtle, but I think the obvious ones are it makes a reference to an authoritative source.

Check that source.

These programs have been known to make up citations.

I know that Google and Microsoft are working hard to eliminate those by having independent programs.

Second, I would never actually do anything that was at all risky without first checking with the medical authority.

So if it told you to change your medication, which it won't, but if it says this medication may not be appropriate, that's useful in the sense that it should allow you to have a conversation, I would not do it.

The other thing is that using human common sense turns out to be an incredibly useful filter.

JOHN YANG: Looking farther down the road, what's the promise of AI?

What do developers and doctors and others in healthcare hope that it will be able to do in the future?

ISAAC KOHANE: So I think that doctors are not spending enough time with patients.

They're spending too much times as bureaucrats.

Having AI take care of that bureaucracy will allow doctors, we hope, to interface more with patients.

That's a short term hope.

The longer term hope, which is when I say long term, I don't mean 20 years, I'm talking about five years, is that these programs will actually be able to look at all our data under the right privacy provisions and actually come up with new biomedical insights, new potential treatments, groups of patients who could benefit from these treatments and actually accelerate the drug discovery process as well.

Because again, the same limitations that were talking about human beings as doctors also afflict human beings as life science researchers, they can't know everything.

They can't know of all the discoveries that are being made at one time.

These programs are pretty good about knowing about everything.

JOHN YANG: Is that the biggest potential pitfall?

People putting too much reliance, too much trust in it?

ISAAC KOHANE: I think that is absolutely the absolute pitfall.

First of all, I would rely on our common sense.

I am not convinced, nor are others, that it has common sense at all times.

And even if it did have a kind of common sense, in the end, we have to be true to our values.

I pretty sure that at this point we cannot rely on these programs, on sharing our values.

But I think the most important thing that these companies could do would be tell us which data were these trained on, these models trained on?

We don't know what these models were trained on.

That tells me, for example, that I don't know, does it correctly represent the problems that American patients have, or did it see a lot of data from India?

And so not knowing which data were used to train these models, a lot of uncertainty around quality and applicability to different populations.

JOHN YANG: Dr. Isaac Kohane of the New England Journal of Medicine AI, thank you very much.

ISAAC KOHANE: Thanks.

It's been a pleasure talking to you.