Revolutionizing Brain Cancer Treatment: Hospitals Trial Google's Med-PaLM 2 AI Chatbot

0:00Parler tout français. Hablas espanol. Parler l 'italiano. If you've used Babbel, you would. Babbel's conversation-based techniques teaches you useful words and phrases to get you speaking quickly about the things you actually talk about in the real world. With lessons handcrafted by over 200 language experts and voiced by real native speakers, Babbel is like having a private tutor in your pocket. Start speaking with Babbel today. Get up to 55 % off your Babbel subscription right now at babbel.com slash wandery.

0:41Today on the podcast, we are talking about healthcare in AI. One of my favorite topics. As you know, I love finding different ways that AI is helping to revolutionize healthcare for drug discovery, for identifying tumors, all sorts of really interesting things. So there's two big stories that I want to talk about today. So without further ado, let's jump into them. The first story I want to talk about is a report that came out of the Wall Street Journal recently that said that Google's medical AI chatbot is already being tested in hospitals. So the Mayo Clinic has actually reportedly been testing the system since April.

1:14And this is Google's MedPalm 2. So essentially, this is an AI tool that was specifically designed to answer questions and give, you know, responses around medical information. So this is something that's been tested. I believe we've covered, you know, AI and the Mayo Clinic in the past because they've been working on this for a little bit. But essentially, this is going to be tested for research with the Mayo Clinic in hospitals and something that's been going on since April. But the Med Palm 2 is a variant of Palm 2. So Palm 2 is what Google is currently using for Google Bard. They've created a medical version, which was announced at Google I.O.

1:50back in May of this year. So Palm 2 is, you know, kind of the specific general one that everyone uses when you're using BARD. And then now this is the specific one. And I think this is actually really interesting because we had research out of Stanford earlier this year that essentially they were able to create a PubMed GPT where they took all the publicly available medical journals and they created a GPT model based off of them. and it was pretty good at answering medical questions, but they said if you wanted to make this thing even better, you'd go more specific and you'd make a heart, you know, like a heart disease GPT and a breast cancer GPT, et cetera, just based off of data from that specifically.

2:30So kind of taking that same model, it looks like Google has done the same thing. And of course, Google's is still fairly generic. They're not going quite down so granular, but, you know, Google is taking that same sort of approach where they have their general Palm 2 and then they're going down to a MedPalm2 specifically for medical information. So I think this is really interesting. And this all kind of comes from a report. Well, the information this is being integrated comes from a report of an internal email Wall Street Journal saw where Google essentially believes that its updated model can be particularly helpful in countries with, quote unquote, more limited access to doctors.

3:08and the PalmMed 2 is actually trained on a curated set of medical expert demonstrations, which Google believes is actually going to make this better at healthier conversations than the general chatbot on BARD being chat GPT, etc. So literally the exact same thing that the researchers at Stanford were saying. So that makes perfect sense. In the paper, it also mentions a research Google has made public back in May that kind of showed that MedPalm 2 still has issues with accuracy. It still, you know, technically can hallucinate or make things up. But what's really interesting with that, I mean, this is an issue obviously industry-wide with all large language models, but in the study they did on that, a bunch of physicians found more inaccuracies and irrelevant information in answers provided by Google's MedPalm and MedPalm 2 than those of other doctors.

4:06So, you know, that's obviously not a great sign if, you know, MedPalm and MedPalm 2 are making more mistakes than doctors. Because I think that's another thing that's important to remember is just the fact that, you know, people are like, oh, you know, Chai Chifty's not perfect, X, Y, and Z, AI model's not perfect. But neither is like advice you'd get from a human. You'd like to have that perfect. You'd like to have, we'd like to lower those inaccuracies, but that also does happen in real life. And so it's very, very hard to get 100 % perfection, which is still the goal with these AI models. So I think that's something noble to push towards.

4:38But what's interesting is while it did have some inaccuracies compared to doctors in almost every other metric, for example, you know, showing evidence of reasoning, conscientious supported answers, or also showing, you know, no sign of incorrect comprehension, MedPalm 2 performed more or less as well as an actual doctor. So there was a number of areas that actually outperformed doctors. And the Wall Street Journal actually reported customers testing this will have the ability to control their data, which is going to be encrypted and Google's not going to have access to it. So this makes perfect sense.

5:13You want to be HIPAA compliant. There's obviously a lot of compliant issues. There's a lot of privacy issues around medical information. So having that encrypted, having that owned by the healthcare provider makes perfect sense. You don't want that sent over to Google servers. And it looks like, you know, Google, who typically, I would say in most industries, wouldn't want to do that. I think they understand the need here and they're willing to make that play. So according to some Google senior researchers, specifically the Google senior research director, Greg Courand, he said that MedPalm 2 is still in its early stages.

5:51And he says that while he wouldn't necessarily want this to be part of his own family's, quote unquote, health care journey, he believes that it, quote unquote, takes the place in health care where AI can be beneficial and expand them by tenfold. So what's really interesting there is the fact that one of the key researchers on this is saying, like, look, I want to use this for my own family, but it does a lot of things. And I think there is definitely ethical issues and considerations around this. one thing in particular to focus on is the fact that perhaps this is useful in countries where they don't have access to doctors and then you got to say like okay well what's the difference between that and perhaps searching on Google and so there's all sorts of different questions there but I do think it's something to bring up and is something we're gonna have to continue to grapple with.

6:38The second news story I wanted to talk about today that I am super super excited about is the fact that there's a pretty major AI breakthrough. So this was reported recently in Bloomberg, but essentially there is a artificial intelligence that shows a lot of promise for treating aggressive brain tumors. And so what it is, is essentially it's called the cryosection histopathology assessment and review machine that shows significant, it has a lot of promise in essentially assisting doctors to treat this aggressive brain tumors. So its innovative method of identifying critical tumor characteristics has, I believe, the potential to really revolutionize the current surgical procedures and methods that we use today.

7:25And I would say that one of its key advantages is its ability to essentially predict a tumor's genetic profile almost instantly. Now, I will, you know, put a word of precaution. It's not as precise as current genetic testing, right? and so the ability to have this be incredibly accurate is obviously going to be found in genetic testing, but the ability to expedite the whole process, which currently takes days or weeks, I think is really going to make a big difference for physicians to initiate appropriate treatments more quickly because this thing works instantaneously. This thing can look at scans of your brain, instantly tell you if there's a tumor based off of different profiles, what type of tumor etc etc and of course you're going to want to get the genetic testing but if you had some serious issue and you could know a week or two weeks ahead of time and begin treatment two weeks early I think it does make a very big difference and so I think that's really really cool so the ability to essentially just provide quick genetic identification of glioma tumors which is a severe type of brain tumor I think is going to significantly improve patient outcomes and reduce the need for multiple surgeries.

8:35So the speed of the system coupled with its ability to, well it's called CHARM, so that's the system's name is CHARM. So CHARM's ability to distinguish between malignant tumor cells and benign cells as well as identifying a tumor's grade or aggressiveness makes it potentially very very transformative tool in the field of oncology and I'd say the development was really underpinned by machine learning. Like this is where we have to give a lot of the credit because essentially this thing was trained on images of samples collected during brain surgeries. And so the technology really innovated in a way that shows, you know, AI isn't just about creating efficiencies, it's about making a tangible, it makes these really tangible life-saving differences when you start integrating it into these.

9:24I think that the impact of this breakthrough is quite significant, the possibility of having nearly instantaneous results in determining the genetic profile of a brain tumor is, you know, already, that's already a game changer. And potentially this, you know, potentially leading to rapid treatment decisions and improved patient outcomes is going to be amazing. So we're gonna have to continue to watch how this unfolds, how this continues to develop. But overall, this is something I'm really excited about with AI is, you know, the ability to detect tumors, the ability to discover drugs. There's so many incredible use cases for AI and so I'm really all about listening and finding how this continues to impact healthcare and how this continues to help benefit people's lives all around the world.