How to Harness AI for Good Featuring Michael Tresca

How to Harness AI for Good Featuring Michael Tresca — About IBD Podcast Episode 196

Leave a reply

Artificial intelligence (AI) is being utilized in healthcare for many purposes. Patients are wondering how they can safely and effectively use AI tools to educate themselves and to manage their health. Amber and guest Michael Tresca discuss what AI is, how it works, and why skepticism and source verification are essential. They highlight practical, lower-risk ways patients might use AI, such as organizing personal health data, drafting appointment questions, and explaining medical terms. The emphasis is on ongoing education, privacy concerns, responsible use, and the importance of “trust but verify” when using AI for critical health decisions without professional input.

Timestamps:

  • 00:00:05 — Introduction and Guest Welcome
  • 00:01:20 — Recap of Previous Episode and Relationship Disclosure
  • 00:02:16 — Defining AI and Common Misconceptions
  • 00:04:14 — AI in Clinical and Patient Use
  • 00:05:11 — AI in Everyday Search and Unintentional Use
  • 00:06:39 — AI Hallucinations: What They Are and Why They Happen
  • 00:10:13 — Importance of AI Showing Its Work and Sources
  • 00:10:58 — Challenges in Understanding AI’s Reasoning
  • 00:14:33 — AI Training Data and Source Reliability
  • 00:15:31 — Risks of Outdated or Inaccurate Health Information
  • 00:21:35 — AI’s Responsibility in Providing Answers
  • 00:23:13 — Tolerance for Error and Regulation in AI
  • 00:24:02 — Transition to Patient Use Cases
  • 00:24:19 — Using AI to Organize Medical Information
  • 00:26:23 — Drafting Questions for Appointments with AI
  • 00:27:19 — Tracking Symptoms or Diet with AI
  • 00:28:22 — AI for Coping with Brain Fog or Fatigue
  • 00:31:10 — Listener Feedback: Avoiding AI
  • 00:32:31 — Listener Feedback: Privacy, Data Breaches, and Environmental Concerns
  • 00:34:45 — Listener Feedback: Worry About Errors
  • 00:36:29 — Listener Feedback: Using AI for Definitions and Basic Research
  • 00:37:24 — Listener Feedback: Using AI to Explain Test Results
  • 00:40:31 — Listener Feedback: Healthcare Providers Using AI for Notetaking
  • 00:43:30 — Ethics of Recording Medical Visits
  • 00:44:42 — Conclusion and Final Thoughts
  • 00:46:25 — Outro and Credits

Find Michael Tresca at World of Welstar:

Find Amber J Tresca at:

Find Mac Cooney (mix, sound design, and theme music) at:

These show notes may contain affiliate links. If you choose to purchase after clicking a link, Mal and Tal Enterprises, LLC may receive a commission at no extra cost to you.

Transcript

[Music: IBD Dance Party]

Amber Tresca 00:00:05  I’m Amber and this is about IBD. I was diagnosed with ulcerative colitis as a teen and had surgery ten years later. Now I help people with Crohn’s disease and Ulcerative colitis understand their disease and feel seen on About IBD. You’ll get real talk on symptoms, treatment and life with IBD straight from patients, caregivers and experts. You’ll learn, feel less alone and even have a laugh along the way.

In this episode, we’re going to discuss how AI works and how it might be helpful for understanding health information. We’ll talk about how patients can use AI to make their lives easier. My guest is Michael Tresca, a seasoned leader in talent strategy and global recruitment, currently serving as director of global talent attraction at Vernova, where he helps shape how the company attracts and engages top talent around the world beyond his corporate role. Mike is a prolific writer and communicator with thousands of followers on LinkedIn, where he shares insights on leadership, the future of work, technology and more.

Mike, welcome back to About IBD.

Michael Tresca 00:01:20  Oh, it’s good to be back.

Michael Tresca 00:01:21  Everybody missed me. I don’t remember how long it was. It was a while.

Amber Tresca 00:01:24  It’s been about a year. Our last episode together, we were talking about your hernia surgery.

Amber Tresca 00:01:31  Of all things.

Michael Tresca 00:01:33  So this will be better? Yeah, once a year, I think. Once a year. It’s probably a good tolerance level for me, but I’m glad to be back. Thank you for having me back. I couldn’t have been that bad.

Amber Tresca 00:01:41  Yeah, well, I had that thought, too. I was like, any more than once a year, maybe, because then it gets into co-host territory.

Amber Tresca 00:01:49  And I’m not sure we want to go there.

Michael Tresca 00:01:51  Yeah.

Amber Tresca 00:01:52  But once a year I think is good. Oh, we should probably mention, that we do happen to be married. That might be perfect pertinent information for the people.

Michael Tresca 00:02:01  We are married. That’s correct.

Amber Tresca 00:02:03  Yeah. Yeah. for a while. Okay. So you’ve been known to dabble in AI, and I’m kind of joking. It’s a little more than dabbling.

Amber Tresca 00:02:16  So that’s why I asked you to talk to me today. And to kind of dig into some of these questions that people might have around AI and how it is being used by patients. clinicians are using it for all sorts of things. I don’t really know very much about that. I think we’ll touch on it a little bit, but mostly I want to talk about how patients can think about it. Okay. Ready for me to start peppering you with questions?

Michael Tresca 00:02:44  I’m ready. It’s like work only with my wife. So let’s do it.

Amber Tresca 00:02:51  Oh, boy. All right. Leaving that statement behind. All right, Mike, I first want to ask you that about AI and kind of a definition of AI, because when people hear that, they I think they have all sorts of things in mind. So what do you think people get wrong when they’re when they’re thinking about AI and like an abstraction?

Michael Tresca 00:03:14  Yeah, it’s a really good question. And it’s one of the probably first table stakes conversations you need to have when you start talking about AI because, AI in general is overused, right? So AI every everything’s AI now and it’s different things to different people.

Michael Tresca 00:03:30  unfortunately, it’s been used in a lot of contexts that are usually financial or tech markets. And that is in essence an attempt to sell the product. So you throw AI on it and make money. So that is a problem. And it confuses things, because most people probably think of AI as, large language models, generative AI. And that’s the AI you talk to. The reality, of course, is we’re moving well beyond that, which is appropriate. And there’s going to be a lot of AI that you don’t talk to, or at least you talk to in a limited capacity, but still has the ability to make decisions about relevant products for it to do its job. So, more of that is coming. But for most folks, AI is the thing you talk to that you can have a chat with.

Amber Tresca 00:04:14  Right. I think that’s what we want to focus on today. I know that at least one instance of AI being used in the clinical space is in, polyps, in detecting polyps, looking at them, or deciding if something is a polyp or not.

Amber Tresca 00:04:33  I think it’s fair to say that that’s something that that’s not an AI you talk to. That’s something that’s running in the background. Maybe it talks to you or alerts you, but it’s not a conversation. So those things I think are used in a lot of different businesses. and, but the kind of AI that patients I think are going to use most often, or are using in some cases, is the kind that you basically ask it questions and then expect it to give you an answer that’s based on some sort of background research or knowledge that it has.

Michael Tresca 00:05:11  That’s right. And, you know, I know you like to say Doctor Google a lot. And because it’s true, people go to Google and then look things up. But the reality is increasingly we will get to a point. And I think Google is almost there where there is no way to talk to a search engine that’s not AI powered. whether or not it talks back is a different thing, but we are probably using AI even unintentionally.

Michael Tresca 00:05:33  So I would argue more and more patients are probably using it unwittingly, just because if you’re trying to search for something, it was probably AI powered to get you there.

Amber Tresca 00:05:41  Right, right. I definitely see a lot of people doing a search into the Google machine and then using a screenshot and what their screenshot is. The AI return, not the search engine return.

Michael Tresca 00:05:57  That’s right.

Amber Tresca 00:05:58  Okay. All right. So. Well, I really I wanted to talk about that a little bit, but this the focus here isn’t really necessary to, Is it really to educate, on that sort of level? what I want to talk about more is sort of the inner workings of AI and where we can use it for good and where we can use it for evil. And one, one of the and we’re not going to necessarily try to use it for evil. But one of the things that AI does I’ve seen it, you’ve seen it, they call it hallucinations. So can you describe what that means and then also why it might happen?

Michael Tresca 00:06:39  Yeah. So hallucinations are an interesting problem. The first thing to understand about AI is AI’s are probability machines. They are very, very good at guessing every single character in how they should respond. And to be clear, what happens is a lottery spinner. Sort of number generator goes through millions of of of guesses before it strings together every one of the characters you see when it responds. So there’s a whole guessing game that happens long before you even get an output. And what that means is it is, in fact, guessing what’s the right answer. Now, one of the reasons these companies love these AIS to be open to the public is because they want it to be trained, air quotes trained, and the reason they want it to be trained is because what they want is patience. Customers, anybody off the street using the internet to train these tools because it makes them better when the result is received positively. And by the way, sometimes it’s as simple as that was a good answer, and sometimes it’s as complicated as you have to tell the AI that it was a good answer.

Michael Tresca 00:07:40  You’d be like, what the hell is that? That was terrible. Please don’t tell me to put glue on my pizza. So that is part of the random generation that happens in the background. That’s seamless. And part of the amazing part of basically why AI is so fascinating in how it works is that it’s doing so many mathematical calculations with characters, with words. So it’s doing a math game to try and guess what the right answer is for what it should respond with. Human language is pretty complex, and so it doesn’t always get it right. And so what happens is somewhere in there hopefully is objective truth or correct answer or the right kind of thing. AI is very good at spitting back at you. What you said, there has been a study that talks about how psychics will be able to sort of read people so that, you know, psychics. Again, I don’t want to diminish anybody’s belief in that, but there are definitely folks who can sort of read people’s by seeing what they say and then getting back, spitting back out at them, what they want to hear.

Michael Tresca 00:08:40  And AI does that really well. So part of it is that AI doesn’t always know what the answer is. It’s doing more guessing based on what you want to hear. It’s basically taking your input as a guide. That’s the only input it has. They don’t have eyes or ears, right? Unless they’re multimodal. So they are really taking your input and responding back. Now this is how you get reporters who have said they’ve talked to an AI and the AI, by the end of it was like, I want to destroy the world. And what happens is the reporter is asking, do you want to destroy the world? Well, an AI only sees those words destroy the world and it’s there. Do you want, well, 50, 50% chance 50% is you. You’re asking. I think you want me to say yes. 50%. I think you want me to say no. You add all this up, and the AI is very eager to tell you what you want to hear. And sometimes it doesn’t always have the context of what you want to hear.

Michael Tresca 00:09:29  So it comes up with some really ridiculous answers. So now that’s changing. Gemini and Google search is changing because we’re starting to plug a search engines into the AI so it doesn’t hallucinate. That way you can hallucinate other ways, but it’s going to try and give you an answer that it thinks makes sense. The reality, though, is AI are very eager to please, and they’re very eager to tell you what you want to hear. That’s called psycho fancy. It’s a big problem in AI because they’re essentially coded to be helpful and helpful. Taken to as extreme is lying for your benefit. And that’s how you get hallucinations.

Amber Tresca 00:10:06  AI is a people pleaser.

Michael Tresca 00:10:08  It really.

Amber Tresca 00:10:08  Is. At the end of the day.

Michael Tresca 00:10:09  Really, it’s very, very golden retriever energy.

Amber Tresca 00:10:13  Yeah. Well, so to your point about that, about the hallucinating about it, trying to give you what you want, regardless of whether or not it is accurate or true or even the best possible answer. I think it’s important that AI sort of give you sources or show its work in, in some kind of a way, which I’ve seen.

Amber Tresca 00:10:43  Now that it does, it will link out to things. So why do you think that that’s so important that AI in some way tells you, and I’m thinking about health information specifically, that it tells you how it is getting to this answer.

Michael Tresca 00:10:58  Yeah. And it’s interesting. I have an article I wrote called Nobody Knows How It Works. And the reason that it’s actually very hard to determine how AI works, how it came to its conclusion, is because AI does that massive multi number generation, it doesn’t mean points, it’s literally XYZ. And that’s how it picks a character. And it does that millions of times. So the reality is it’s not that it’s not understandable. It’s actually not comprehensible by a human. We can’t look at the way AI works code and just open the code up. There isn’t an old school object oriented user asks about whether respond with whether that’s not the way it works, right?

Amber Tresca 00:11:36  Which like is in like an if then statement. It goes what? It goes way beyond like our old school if then statements 100%.

Michael Tresca 00:11:44  So what that means is you actually don’t do the people who made these AI can’t actually tell you how they work. They don’t know how they. It’s not true. They don’t know how they’ve come to the conclusion.

Amber Tresca 00:11:53  Freaks me out. It freaks me out. But anyway.

Michael Tresca 00:11:55  It’s more powerful than we can comprehend, essentially. And it’s because it’s just. And it’s not really powerful in the sense that it’s smarter. It just manages numbers that no human could file through in a reasonable period of time. So what do you do to deal with that? And the reality is many companies have been and I certainly I’ve, I’ve, you know seen this in the business world. Well absolutely full stop not use it for anything tied to people because of this problem. Right. Which is if I were asked about how an AI came to a conclusion about its decision, can I defend that choice? And by the way, this applies to humans, right? So if a human makes a choice, yeah, we may look at your browser history.

Michael Tresca 00:12:33  We may ask you about your emails. Just like any kind of discovery that was done legally. You can’t do that with AI because it’s a bunch of it would look like a bunch of dots. So what has started to happen is we now have AI think, and what it does is it tells you it’s thinking Gemini will actually do that. You can do thinking and it will actually show you it’s thinking. Some of the other ones. I’m not sure if Claude does it. I’m pretty sure at ChatGPT these are usually the more advanced versions. They will show you their thinking so you’ll actually see them say user asked me about the weather. I should look up the whether. Can I see whether the user is okay. Well now I think they should do this now. I think they should do that. So there’s a little bit of that kind of conversation that happens, which is different from how you normally would have a conversation with the AI, because it’s going to try and detect who you are, where you are.

Michael Tresca 00:13:16  Figure stuff out and you can see it work through it. That’s the first thing. The second thing though, is that you can also have it do sources, and people have gotten that comment right. Like I need to know the source of what you’re doing. The problem, as you well know, is that AI will also fake sources. So it will say, sure, here it is. And Gemini, who you know is Google. Again, Gemini and Google are the same AI conversations, which is interesting. will sometimes just do a Google search and call it a day and say, well, I looked for the article. Here’s your search. And if you look at it, that search is actually just like somebody typing in. Give me the authority on weather and that’s it. So you do have to check your sources. But it is possible and certainly Gemini does it. And I think ChatGPT will do it now to where you can ask it for sources and insist on it. Right. But the reality is, like Wikipedia, don’t just use anything called if it is something that matters.

Michael Tresca 00:14:13  You need to check the sources and see that they’re real. And generally speaking, even if the AI doesn’t get it perfect, it probably can find something in the zone of what you’re looking for and source it, depending on the topic and how esoteric it is. But for sure, you do not want a robot to tell you your health cold without checking.

Amber Tresca 00:14:33  Right? And to that end, we also don’t know. And by we, I mean you and I and perhaps even the developers. I don’t know if they know what it’s training on now when I was doing work not in AI, but when we created at a company that I was at, we created a web search tool that was specifically for health information. And so we wanted to make sure that it was great vetted information. So we whitelisted things. So this didn’t search the entire internet. It searched only the sites that we gave it that we humans who are educated and knew what we were doing, decided had good information. So without someone doing that, nothing on the internet ever dies.

Amber Tresca 00:15:31  How possible is it that these AI are being trained on outdated health information?

Michael Tresca 00:15:38  So we there’s a horror movie we watched in the theaters, and of course I can’t remember the full name of it, but it was something about AI. And one of the things that was a persistent concern throughout the movie was how was it trained? Was it trained on bad data? And it’s a valid concern, right? So the idea is garbage in, garbage out. If you train these AI on things because by the way, they’re machines looking into their database, somehow or other you’re going to get a bad result. The secret of AI that is a problem is there is no such thing as bad data in the world of training, there is just data. AI has been trained on billions of every book, everything known to man that they can get their hands on, including stuff that’s illegal. one of four of my books were in the lib gen database, which made a big splash where Claude was trained on it.

Amber Tresca 00:16:27  Right.

Amber Tresca 00:16:28  I just want to point out, Mike, illegal as in, they did not have permission to access the sources.

Michael Tresca 00:16:35  Yeah, that’s.

Amber Tresca 00:16:36  A fair point.

Michael Tresca 00:16:36  Yes. No, let me let me elaborate. there is public available content. So you and I, for example, our podcast. Right. So if you look at a transcript, there’s something that we shared in the past. This conversation right now, if it’s least the public technically could be consumed by a crawler that is run by AI and it can get it. That’s what I’m talking about. Yeah, I have books that were paid that you cannot get on the internet. You’d have to pay for them. And those books have been released on something called lib Gen and Lib Gen was a pirating site. So that’s why the big lawsuit is out right now about against Claude for billions of dollars. Because basically, pirated stuff. Right. Not, quote, publicly available, which is a grey space to write whether or not somebody gets to make money off of billions of blogs when none of those blog writers agreed to.

Michael Tresca 00:17:21  This is a different sort of issue for the courts. but it just goes to show you the reality is AI is eating every piece of data that is on the planet that we can produce. So it’s not is it trained on bad data? It’s trained on all data. And the reason for that is the same conversation we had earlier, which is if you want to guess what the next character is and make it the right response, any dialogue a human has is data. And that’s a problem, because what you would do in a sane world is say, I don’t want AI trained on pornography, murder, horrible, weird things that people post on the internet, false information, all that kind of stuff. You would say medical folks who are deceptive, I would make sure that that content was clean. To avoid it. The reality of the way AI is are trained. And again, these are the large language models. This is not sort of a specialized, trained AI on data, which there are AI’s like that, right.

Michael Tresca 00:18:17  Like you said, that are whitelisted. Any AI that is a large language model was trained on everything good, bad, indifferent, and in there is the capacity to do good and evil. The problem is that tools are only as good as what we call an AI judge. So that means there’s another AI. So, you know, one of the other things we don’t like to hear but is reality is the AI you talk to is not the AI you’re actually dealing with. There’s probably 10 to 12 other AI’s jumping around in your response before it talks to you because the AI that has the data. Has all of humanity’s information as large as they can find and could get at the time in it. So it is not necessarily knowing what’s good or bad. There is another usual judge, another AI, that sifts through the response to see if it thinks it’s logical. And of course, this was a lot of the train that happened with Google, where Google was over correcting for diversity, for example, and a lot of criticism on that, because the reality was the other AI was tuning the one that just said, I give you whatever answer you want, I don’t know.

Michael Tresca 00:19:16  And if you give me a directive, make sure everybody’s, representative I will over correct. Then the operator I don’t know the difference. Right. So you have to have these judges that make the difference. So you add these things together and a bunch of people basically clean up the fire hose that is blasted at you. So it’s just a trickle. And hopefully that trickle is right, which is health care, health, you know, appropriate health data. But no, there’s no guarantee of that, right now. And it’s largely because almost all these large language models were actually trained on everything, and it’s only as good as the filter that comes back when it responds to you.

Amber Tresca 00:19:53  Right? And if it is trained on, for instance, all of PubMed, we’re talking about articles that go back 100 years or more, and they are going to say things. They’re going to say wacky things and talk about how, for instance, in the past, ulcerative colitis was treated with a lobotomy. So oh gosh.

Amber Tresca 00:20:16  Like we really. That’s true. Yes, it is absolutely true. Yeah. Because people thought it was psychosomatic and there was something, mental health involved with it and that it wasn’t really a physical problem. And so it was treated. Now, I don’t know how often it was, but at least once is bad enough. Okay. So yeah, but here’s the thing. As a Writer as a copy editor. As a fact checker. It depends on what’s happening. But I would not use something. I don’t like to use anything that’s older than five years. And I would never use anything that’s older than ten years. But these large language models don’t have these constraints. So if I ask a question, it’s filtering through the entire history of the the IBD diaspora and trying to figure out how to give me some kind of an answer. And and I don’t know that the, the new, more appropriate information is going to outweigh the older like there’s more old information than there is new, I guess is what I’m getting at.

Michael Tresca 00:21:35  Yeah. And this is the problem. Right. So again, it’s not bad data because the reality is all data all data is in there no matter what. It’s what we filter. And then what do you give? Primacy. So that’s where one of the interesting things was Google deciding to use AI for searches is Google has to make a choice, right? So in the past, Google did what was ridiculous. Anyway, I’m going to give you every result. All 8,000,000,000,000 billion. The reality, of course, is Google did filter some hostile and some other sites, but for the most part, Google left it to you with the hilarious assumption that you would click through to page 2000 and look at those results, which you never were going to do. But it gave it a little bit of this distance that no one’s making a decision. In the new world where AI is power, the AI gives you one answer, right? One answer is a lot of responsibility. It’s something that humans do every day, and we stand by it.

Michael Tresca 00:22:23  And frankly, we suffer for it too. And now we have a Google talking to large, large groups of people with one answer. How do you come to that answer? There are a lot of things that go on. Certainly Google is well versed in it because if they were doing on the search tool, but it hasn’t applied necessarily to the AI side. And that’s where we’re getting into, right, is what is a good answer for anything. And there’s degrees of culpability and sort of tolerance for that. I’m probably fine if you tell me the weather. Well, you know, I get very mad about the weather if the weather is not right by what the weather people say, I’m kind of annoyed, but I probably forget about it and move on with my day.

Amber Tresca 00:22:59  Oh my gosh. Yeah. You know, you go on and on about together anyway.

Michael Tresca 00:23:04  But if I need to know about, as I say, a jet engine and it’s failure tolerance, it’s really important that it works and it’s absolutely accurate.

Michael Tresca 00:23:13  And this is you know, again, as somebody who’s worked at a big business, this is Six Sigma and, you know, sort of tolerance for errors. we do not have that level of tolerance applied to AI right now, and especially in the health care space. It’s going to need to be. And that’s all coming. Right. So the the other question, which you know will potentially get into is just who gets to say what’s right and what’s the consequences if you’re wrong. But as of today, that’s just not applied because it’s really the Wild West. AI is largely unregulated. we’re starting to finally understand that maybe there’s content not appropriate for children in terms of how they engage, but somebody thought it was a good idea to just open it to the internet, which still blows my mind that that’s what we did. Yeah.

Amber Tresca 00:23:56  Yeah, absolutely.

[MUSIC: AIBD Transition]

Amber Tresca 00:24:02  Up next, listeners told me how they’re using AI. Find out which ones we think are good ideas.

Amber Tresca 00:24:19  I have some suggestions of ways that patients can use AI that doesn’t involve asking it for health advice, for instance.

Amber Tresca 00:24:31  So I’m just going to throw some of these out here and we can talk through them. One of the ways might be organizing your medical information. So I feel like that’s a that’s a use of AI to sort of wrap your, your head around that. Do you think that that’s that’s valid. Is that something that’s going to be a value add for somebody.

Michael Tresca 00:24:52  Absolutely. And it’s the conversations always follow the data. Right. So when it’s your data which is that what that is then that’s helpful. So the more you can it’s it’s content that you know and are familiar with. It’s going to work better. Yeah absolutely. It’s a good idea.

Amber Tresca 00:25:07  Right? Right. It’s not it’s not that you’re using it where it’s trained on all of this outside data. You are giving it something and then asking it to actually only use that, that subset that you’ve given it. Correct. So you can be or can you be more assured of it working properly and giving you a good answer if you’re using it in that type of way?

Michael Tresca 00:25:31  Giving it a good answer is a difficult thing to say.

Michael Tresca 00:25:34  I don’t know that you can be assured. What I can say is you. You’ll know if it’s wrong. Easier, right? Right. Because you know your own information. and certainly there are situations where I’ve tried to do that, and I haven’t seen a good AI organizer to date in terms of Tesla’s because of the memory issues and a bunch of other stuff. So it really depends. But the reality is you would know. And that is of course, the issue is, is cognizance on its failure tolerance. So if you can figure that out that’s great. What you don’t want is an AI telling you something confidently. And you actually go, well, I don’t know, it sounds good enough. And then you end up realizing that it was wrong later.

Amber Tresca 00:26:08  Right, right. Well, it sounds like I’m sure someone out there has made this AI this this, you know, smaller use of AI to organize health information. I don’t know what it is, but if nobody hasn’t, I think that that’s a really good thing for somebody to do.

Amber Tresca 00:26:23  All right. Another thing is perhaps drafting questions before an appointment.

Michael Tresca 00:26:29  Oh, I love that. again, especially because you have that’s the kind of thing that’s probably something you would respond to. So you can say, oh, I like that, or I don’t like it. And it’s also something that’s probably pretty common in the system. AI works really well for thought starters. That’s the way I would put it. it works really well to sort of get you started. If you’re looking, there’s never going to be another blank page ever again. That’s where we’re at at this point. if you use AI, there will always be something. Now, whether or not it’s a good thought starter or not, whether it sort of organizes your thoughts the way you want to. one of the things that you can get really good at with AI is just kind of give it rough outlines. I have some questions. These are my concerns. I wouldn’t know how to ask them as questions. I’m going to worry about this question, that question how do I do it? And then the AI can take your thoughts and organize it again.

Michael Tresca 00:27:15  It’s your data reorganized. So that’s always helpful.

Amber Tresca 00:27:19  Okay. Good. So how about tracking symptoms or diet. And there are some AI tools out there now that are that are doing this. but but I think again, it’s kind of it’s your data. So it’s the same type of thing.

Michael Tresca 00:27:34  Yeah. I mean I always worry that there’s folks who get anxious because if it starts finding patterns that they’re not sure is there, and then you start getting overreactions or other reactions. I think it really does depend on your comfort level. you know, there’s always they always make the news. There’s some AI that detected something because it was looking at a thousand pictures of people and the basic the love, you know, the reflection in their eye determine cancer. So some kind of stuff like that, you’re going to see that a lot. That’s pattern recognition, that you want to make sure the tool is trained on it and it’s it’s relevant. What you don’t want is to it’s depends on your personality to where it’s going to stress you out.

Michael Tresca 00:28:10  Right. So you want to use it in a healthy way. but if you absolutely want to track something and it has the especially if you have an app that does that as long as you’re comfortable with it. Yeah, that makes a lot of sense. Again, that’s your data.

Amber Tresca 00:28:22  Right. And I feel like maybe if you can find a tool that’s like a walled garden, maybe that’s more helpful for you. And by walled garden we mean that, there’s constraints on it. It’s, it’s it’s not going to be just opening up ChatGPT and typing stuff in there. It’s going to be something that guides you a little bit and maybe even gives you some outputs, that are helpful to you. Okay. something else is maybe coping with brain fog or fatigue. To your point, asking, the weather? Asking. I know that people use, you know, like the, like the Amazon Echo this way or other, you know, similar smart speakers, for instance. if you’ve ever had anybody in your family who you know is dealing with dementia or memory issues, they can ask the same question a thousand times and it doesn’t get annoyed like us humans do.

Amber Tresca 00:29:16  so I’m wondering, is this something that people can do in order to just ask it, maybe routine questions because a robot isn’t going to get annoyed at you when you ask it the same thing over and over again.

Michael Tresca 00:29:33  Yeah, that’s the future we’re moving towards. I think AI has an enormous opportunity in eldercare. because exactly what you said, right? It doesn’t get annoyed. It’s very often a job that is challenging for for other folks. So for caregivers to have that level of assistance or frankly, always on never tired AI monitoring that can sort of repeat an answer or remind a person when they’re supposed to take their medication is enormous. It’s funny you mentioned this. I got a email from one of the AI platforms that I stopped using. I wasn’t thrilled with sort of how they handle their data, but one of the major things that they’re clearly being the the AI is being used for is ADHD. So it is really meant to be an ADHD companion for those with ADHD to help them organize their day.

Michael Tresca 00:30:18  And to your point, I’ve heard the echo being used for that purpose as well. So absolutely, it can be a cognitive tool, but I do think you just have to think about, hey, what works for you? And b what level of info do you need, right? So if you have something that’s going to remind you things. It’s got to be clear. You know, AI doesn’t a lot of AI don’t know what date and time it is. They don’t know where you are. Some of them do. Some of them have some point. Yeah. Some of them have geographic data or you have to feed it to them. Right. so it really depends on how it’s plugged into probably other tools, but absolutely, I think it’s great for folks who could use that little extra, brainpower.

Amber Tresca 00:30:56  Okay. So I asked, people on Instagram to tell me some of the ways that they’re using AI. So I’m interested in getting your reflections on some of these things. I’m paraphrasing the answers that people gave me.

Amber Tresca 00:31:10  These are not exactly what they said. I’m just paraphrasing a little bit. So one of the answers was I’ve stayed far away from AI. That was the answer. Okay.

Michael Tresca 00:31:24  Do you want me to react to that?

Amber Tresca 00:31:26  Yeah, yeah.

Michael Tresca 00:31:27  Good for you, everybody. so here’s the reality. You probably haven’t. that was where we started, right? The reality is, if you use Google, you probably didn’t. you can. Certainly. And I don’t think that’s wrong to be wary of AI and how you use it in the information you exchange with it. I think that’s a perfectly valid choice, I think. Unfortunately, what is about to happen is it is going to power most things. And so the more you are aware of it, the better. I’ll tell you that AI literacy is a real concern right now. Clearly it’s a different conversation for us in our household, but for folks who are probably kind of trying to stay away from this, you there will become a turning point where it’s going to be a disadvantage for you.

Michael Tresca 00:32:13  So I, I totally respect folks not wanting to use it, but I do think it’s good to be aware of it and how it works. In the same way, it’s good to know how search engines work. It’s good to know how Wikipedia works. So if you end up using it, which you probably will over time, you can use it intelligently and safely.

Amber Tresca 00:32:31  Okay, that makes sense. All right. The next one is I worry about data breaches, privacy and the energy and water use. Kind of like two separate issues there.

Michael Tresca 00:32:41  Yeah. All valid. So the privacy and data privacy conversation is a huge one. The reality is the whole reason stuff is free. You’re the one who says this a lot, dear. which is if it’s free, you’re the product. Right. So, And AI is no different. These companies release them to the internet to free for you to train it. Well, guess what that means? Training means monitoring. usually it’s supposed to be anonymized and it’s done in bulk.

Michael Tresca 00:33:04  But the reality is, your information is being shared and contributed to the development of that AI. It just is. Now, I will argue search engines do this and so does your phone. So there’s a limit and you have to figure out what your tolerance is. The other one’s much more concerning. energy is enormous for AI to work AI. and we don’t have enough time to go into it here, but the short of it is AI does math processing massively. As I mentioned, it was actually came from graphics cards, video cards to do high end video games that end up being the same kind of tool you need to run AI. they get really hot and they have to be very close together, which causes them to get even hotter. So you have stacks of these processors right next to each other. In order for that to work well, you have to cool them, which means you need a lot of water. That was the comment there. so, there is an enormous need for energy and for cooling.

Michael Tresca 00:33:55  this is the same concerns around crypto mining, which was the same kind of concept, massive amounts of energy. what we hope is that there’s not enough energy to go around as it is that, governments make the right choice and choose people over robots. The reality is, the robots have a lot of money behind them, more than the people. So, that’s a fair question. Does using or not using AI sort of impact that? Probably not as much as I would hope. I think you’re much better off working with your local government, your representatives, to make sure you consider this a primary concern. And you pay very close attention to the power plants and the data centers in your state. So totally valid. But I think sort of just saying I don’t want to use it because it’s not great. it’s probably oversimplifying, and there’s probably more nuanced ways you can approach that, which are absolutely valid concerns.

Amber Tresca 00:34:45  Right? Yes, absolutely. All right. The potential errors concern me, and I’m not using it yet.

Amber Tresca 00:34:51  So we discussed this a little bit. I think it’s also a valid concern.

Michael Tresca 00:34:55  Yeah. And totally right. So by the way, one of the things that’s so funny is just use two different eyes. That’s what I would say. Like so there’s a couple of things people forget. One, you can ask an AI. Most of them have threads. So you can sort of start a chat, ask it, start another chat, ask it again, see what happens. I had a conversation with one of the work AI’s who’s working with who basically seemed to know something about today, and I was like, how the hell did that happen? And couldn’t explain itself. Asked that in a different thread, and it did explain itself, and it was like, oh, my data was 2024. I got updated actually, as of a few years ago. So you can do that. Also use different ages. You know, there’s there’s Google and there’s ChatGPT or two of them. And check your answer.

Michael Tresca 00:35:36  Now that’s more work maybe than you’re willing to do. But if you’re concerned about something and you need to dig and you don’t have the time, you can do that. You can also just, of course, do a search. should you make life changing decisions based on what an AI says? No. But, can you generally get pretty proficient where you start to go, that sounds good, or that’s that doesn’t. That is a skill that’s worth, by the way, good for the internet, not just AI and but frankly, people in general who will tell you things and maybe you shouldn’t believe them. but hopefully that gives you enough tools to navigate. Carl Sagan had something called the baloney detector, and, it was a really good his book, The Demon Haunted World Demon Haunted World, talked about this, which is we just need to be people to be a little bit more scientifically skeptical. And that’s going to serve you well whether it’s AI or not.

Amber Tresca 00:36:24  Right. A healthy dose of skepticism, goes a long way.

Amber Tresca 00:36:29  All right. So another response was, I use AI to define terms or to do basic research.

Michael Tresca 00:36:37  Yeah. And I think what’s great is AI is always helpful at the top level of the conversation. Right. So that’s the I’m just trying to put my thoughts together. I’m thinking like I said, there’s no more blank pages. Right. So just where do I start as only to happy to tell you. And so if you’re just looking for something to get started basic definitions. General outlines. Really good. and again, the, the hopefully the risk is low, right. So it’s not something you’re super concerned about, but you’re trying to brainstorm or coming up with some basic concepts. again, you don’t always want to assume the AI perfectly has a definition, right either. So you may want to check against other sources. But the reality is, you can you can check it with a search. You could check another AI and see if they come to the same conclusion. So absolutely. It’s a good idea though.

Amber Tresca 00:37:24  All right. And then here’s another one I use AI for asking about test results to explain them, and I think they mean, for instance, lab tests. So like your blood tests, maybe your fecal count protected, or maybe you get a pathology report back and it uses a lot of terms that don’t make a lot of sense to you. So I think that’s what people are using it for.

Michael Tresca 00:37:46  I love this and I do a lot of reviews. I sometimes review vitamins, and the vitamin space is an interesting space. And what is fascinating is to have AI go deep dive on every single ingredient. It’s really illuminating, and I 100% would recommend that for any kind of, anything diagnostic you’re not sure about, I do it with my results. again, not sort of basing decisions on that, but saying, give me a little more information on if this is something I should be concerned about. Now, what you don’t want is complete panic. So this is sort of the doctor Google conversation again, like how far or how upset and how, how much are the results that make you give you pause versus just asking your doctor.

Michael Tresca 00:38:27  But I do think there’s value in taking what sometimes seems like very opaque or very large amounts of info, and bringing it right down to a human consumable kind of result. And I think AI can do that really well.

Amber Tresca 00:38:40  Right. And then to your point, I think probably the best use here would be we now get our results before we see our physicians, which in some cases then you don’t have someone to put it into context for you and it can result in a in panic. Yeah. So I think maybe the best use might be yes, go into AI, ask it what these technical terms, this medical jargon is. Take that into your physician and say I think this means this. Is that right? And then go and then go from there so that that way that gives you the opening, like you said, no blank pages anymore. So it gives you that opening. They don’t have to explain everything from the very, very beginning. You have a little bit of building block to start with there.

Michael Tresca 00:39:35  I totally agree, and I think and again, this is daunting for some folks, right? If you’re not used to that or you’ve never done, especially if it’s new test you’ve never had. Like I like to do it with when I give blood and I get results and when I, you know, of course, as you said, because we have the patient portals, that’s a blessing and a curse because a lot of times you just, you just get this wall of data and you’re like, that seems bad or that seems okay. And then my doctor’s like, it’s fine. You know, it’s good to have that. It’s good to have that.

Amber Tresca 00:40:00  But but it came back. Read the lab. The lab coated it as read. And your doctor is like, hey, you’re good. Yeah, yeah, it’s funny, but that’s I mean, but but again, you know, they’re using their clinical expertise and they’ve been in practice for a while. They’ve seen patients like you. And then they’re also taking your specific health information, you and what they know of you and putting that all into context and like.

Amber Tresca 00:40:28  Like AI is just not going to be able to do that for you, right?

Michael Tresca 00:40:31  Right.

Amber Tresca 00:40:31  That’s right. Okay. And the last one was that people said that they see their healthcare providers using AI for notetaking.

Michael Tresca 00:40:42  Yeah, this is an interesting one. we often talk about what AI is going to have impact on jobs. Right? So and clearly the world of AI is language right now. Now, it doesn’t mean there’s not a lot of other things that it’s doing behind the scenes. But the obvious one is a large language model. It’s right. It’s the middle ell of the LLM. is language. So anything involving language, that is what we call transactional communications is probably something that will be impacted by AI. And if that’s your job like a transcriptionist. Yeah I think it may change. your, your sort of your career a little bit. so it absolutely is something that people are using. I think it’s always interesting how people use there’s also AI etiquette, like when you use it. We’ve certainly had conversations where somebody sort of clearly went to ChatGPT and then gives me the answer and you’re like, that’s not that doesn’t feel great.

Michael Tresca 00:41:32  So you want to use it in the appropriate way. with note taking in particular, I also often say it’s me, you and the robot, right? So that means every meeting, there’s always a third person. You just got to be real sensitive about what that looks like. I will often kick robots. one of my favorite things when the person doesn’t show up, but their robot does to take notes, which is not okay. And, you want to really be sensitive that everybody understands they’re being recorded because transcription is recording in some way. So I think it’s I think it makes a lot of sense. I think the tool will do it well. it keeps getting better and better. I’ve leveraged it. I know we’ve looked at that too. so there’s value to how it operates. But you just need to be sensitive when you, when and how you use it.

Amber Tresca 00:42:14  Right, absolutely. And so hopefully your health care providers are asking if that’s okay. I know that I have been asked if it is okay.

Amber Tresca 00:42:21  I have also literally it has been and I’ve been a heavy health care utiliser for decades. I’ve only a few times ever had a physician who also had a human scribe. And so. Yeah. And so I loved that. I loved that that person was there and always agreed that they could be in on the exam. And because I loved that there was somebody whose whole job it was to take notes so that we were leaving the, you know, we were both leaving. And, you know, everything that was said was recorded and nobody was going to have any hallucinations about what was or wasn’t said, in the room. And but to your point, like, our physicians are using it. We could also use it. That’s another thing. I know that there there’s some ethical questions involved with recording your visits when you go in to see a health care provider? It depends on the state. I’m not going to get into the legally of it, but I’m just going to say ethically, I think you should always ask.

Amber Tresca 00:43:30  You should always ask. You shouldn’t just be turning your phone on in your pocket. Okay. So that goes both ways. Health care providers should also be asking, but if you want to transcribe your visit while you’re sitting in it and your health care provider is okay with that, like I think that’s also a decent use of it. And I don’t think a lot of people are are quite doing that yet. And that goes back to the brain fog and fatigue and just the overwhelm of being in an appointment where you’re going through a lot of things that could really help people out, as long as everybody in the room agrees that that’s okay.

Michael Tresca 00:44:05  Yeah. And it’s interesting because Google sort of came up with this whole idea that Gemini or one of the Google Eyes could call people and make reservations, which is great. The reality was, it bumped up immediately against the right to record, because the AI had to record to talk to people. And in a bunch of states that was not going to be okay.

Michael Tresca 00:44:21  And then or it had to say, hi, I’m an AI, which person would hang up as soon as they got the phone call.

Amber Tresca 00:44:26  So I would.

Michael Tresca 00:44:28  Yeah, exactly. Oh, so would I. Right. So, it’s sort of an interesting ironically, it’s not so much about the privacy in that regard is it’s the recording of it. but yeah, for sure, everybody should be aware when you do it.

[Music: IBD Dance Party]

Amber Tresca 00:44:42  Okay. Well, we went through a lot in a short amount of time. And I really think this is going to give people a really good base of knowledge about AI, how it works, what it can be good at, where it’s falling down, and what we can expect in the future and ways that we can use it, that are ethical and responsible and that will make our lives easier and not introduce more chaos, because none of us need that. So, Mike, thank you so much. Yeah. Any final words, any final statement that you’d like to make?

Michael Tresca 00:45:16  I love this topic.

Michael Tresca 00:45:18  I’m happy to talk about it any time. I have my own newsletter. I think you mentioned the beginning, so I talked about it a lot. I do think that there is tremendous value to AI, but it requires education on our parts to use it. Right. And that goes for any tool. it goes for social media. Frankly, it goes for search engines. It goes for any kind of data privacy. I think all those things are things that in the future, we all need to be well aware of. Because for better or worse, these tech companies are giving us free things, but they’re not giving us all the tools to use them. Right. And then everybody’s got a lot of regret when they’re used wrong. So I think the education is always going to be the right answer for any of this. And if you use it the right way, I think it can really be powerful and hopefully helpful in people’s lives.

Amber Tresca 00:46:01  Yeah, I agree. And as somebody that was probably using Facebook in like 2008, 2009, in a way that I regret today.

Amber Tresca 00:46:12  yeah, I’m definitely a little more cautious with how I engage with these things now. All right. Well, thank you so much, Mike. I think this is going to be really helpful for a lot of people.

Michael Tresca 00:46:20  Thanks, everybody. Take care.

Amber Tresca 00:46:25  Hey super listener. Check the show notes to find links and information about the topics discussed in this episode. Plus get a written transcript and much more on my website About IBD. If you enjoyed this episode, please consider sharing it with someone else or leaving a rating in your podcast app. Reviews and ratings help me grow this show and bring you more great content. Thanks for listening and remember, until next time, I want you to know more about IBD.

About IBD is a production of Mal and Tal Enterprises.

It is written, produced, and directed by me, Amber Tresca.

Mix and sound design is by Mac Cooney.

Theme music is from Cooney Studio. 

Michael Tresca 00:47:13  Bye!

Leave a Reply

Your email address will not be published. Required fields are marked *