Transcript 0:00 [upbeat music] If you don't know your numbers, you don't know your business. 0:12 I learned this the hard way with my first company, so I created NetSuite, the world's number one cloud business system. 0:19 NetSuite gives you the visibility and control you need to make smart decisions and grow with confidence. That's why companies who use NetSuite grow three times faster than the S&P 500. 0:32 Schedule your free product tour right now. 0:45 Hello everyone, and welcome to the Digital Leaders Show, where we talk key issues and news at the intersection of business and technology in hopes of making you a better digital leader. 0:56 We're live every Wednesday at noon Eastern, 9:00 Pacific on LinkedIn and YouTube, and the audio recording is available on major podcast platforms the following week. I'm Dan Goodstein. 1:04 I head up ERPA AI, it's the Institute for Robotic Process Automation and Artificial Intelligence, as well as the Digital Enterprise Institute. 1:12 We're a professional association, media and peer-to-peer research network providing education, networking, and advice to over 130,000 members globally all around automation, enterprise AI, and digital transformation. 1:24 Thanks to our sponsor, Oracle NetSuite. Their unified cloud business management solution helps organizations operate more effectively and provide real-time visibility into operational and financial performance. 1:34 ERPA AI members and followers get preferred pricing of 40% to 45% off NetSuite. Just go to erpaai.com/netsuite for more information. 1:42 Today, we're discussing when AI hype becomes unethical, plus other tech business headlines for the week. My co-host today is Chris Sirdak. 1:51 He's a senior adv- advisor to AI, ERPA AI, an author, a consultant, former White House Chief Transformation Office advisor, partner for digital transformation at Gartner, and automation AI practice lead at EY. 2:03 Chris, good afternoon. Happy post-Thanksgiving. Happy post-Thanksgiving. Hopefully you're over your tryptophan, uh, sleepiness and, uh, getting ready for the next holiday, right? I didn't have it this year. 2:16 I think it may have been the, uh, the, the w- the alcohol I drank while I was eating. It kind of helped me get, uh, keep, keep going. It was, uh- Yeah... going. Alcohol often trumps tryptophan, I'll give you that one. 2:27 [laughs] There you go. There you go. How's, uh, how's your week been so far? Lots, uh... 2:31 I, I still can't keep up with all the, the AI news, but, uh, I think we've got a, I've got a pretty good rundown for everybody today. Yeah. 2:39 No, I, we're certainly reaching what I thought would have been a level of saturation quite a while ago, but we keep pushing the boundaries. I'm, I'm, you know... 2:46 People that complain we don't have enough content to train these AIs just need to keep having the AIs generate more content and we will be all set. Circular firing squad, sort of. Yeah. You know, um, 2:57 every time I think that, you know, things are kind of settling down, like you say, uh, I see another, another player getting into this, uh, space. 3:07 So this week, uh, we had, uh, Amazon announce their UltraCluster AWS super computer, uh, composed of hundreds of thousands of its own titanium tri- uh, chips. Uh, they're working with, uh, NVIDIA. 3:26 Uh, they're gonna supply chips to Apple. Um, and they're, they're launching their own, what they call Nova family of generative AI models. So just when you thought maybe they were asleep, bang. 3:39 [laughs] Well, and, and, gee, back when I got my MBA almost 30 years ago, you know, we talked about being a fast follower and how often that can be, um, the more disruptive and, and more advantageous position to be in. 3:53 So, and, and let's face it, Amazon has been remarkable not only at being very successful inherently, but also in constantly recreating themselves. They are, they are a transformation engine. Um, I mean, 4:07 Amazon Web Services is one of the most profitable businesses, one of their largest businesses, and when they first came out with that, we all said, "Why in the world are people using a bookstore to run their software?" 4:16 So, um, I'm, I'm not all that surprised. Um, I will say that as a writer, you know, we always have this rule of threes. You wanna say things with three words, and it kind of just has a rhythm to it. 4:26 So when they say UltraCluster, I feel like they're leaving the third word out, which we can all fill in as, as you see fit. Um, but, you know, uh, they're, they're certainly gonna make a splash. They're gonna cause more 4:38 hesitancy, maybe. A little bit more... A combination of more FOMO, but also more of a potential pre-buyer's remorse, which we saw with RPA, which we saw with VDI, which we saw with big data, on and on and on. 4:51 So, um, the one thing you can count on from Amazon is to disrupt markets as they, as they grow, and here we go. 4:57 And, and speaking of disruption, um, you know, for all of the ISVs out there, the other big AWS announcement, uh, was what they're calling Amazon Q, uh, which is gonna allow ISVs to enhance their, their gen AI experiences. 5:13 So, so, you know, I, I guess, you know, thus far there's been sort of two categories to this. 5:18 One is, you know, helping companies create their own AI products using this technology, um, and then, uh, creating their own AIs limited to their existing data, uh, limited to their existing data. 5:35 This announcement opens things up. So if you're a privacy person, uh, and you thought things were, uh, getting quieter, uh, you got a whole new, uh, case, whole new, uh, thing to, to focus on. 5:46 But this is, this is promising to be able to retrieve data from multiple sources, not limited to an ISV's, um, uh, you know, data, which, you know, aside from the privacy thing, is really, really amazing stuff. 6:00 I mean, you're talking about, uh, just... This thing hasn't been- Accelerated in the last year. I mean, I, I think this accelerates it tenfold. 6:11 It also brings up, as a sales and marketing guy myself, this question of, you know, how do you... You know, as a, as a company, how, how do you differentiate? 6:19 You know, uh, it used to be that you had a product, a service, you had, you know, differentiation or specialization over others. 6:27 It, it makes me wonder, you know, how do you differentiate when everybody's gonna have access to these tools, everyone's also now gonna have access to the, to the data? I... 6:37 You know well I've been talking about this for a decade, I've been writing about it even more with an emphasis the last three or four months, but it's the digital gray goo problem. 6:46 If everybody's producing mountains of content that's all built on this average amplifier, which is what LLMs are, they amplify averageness, we're just producing more and more digital gray goo. 6:57 And to your point, it makes it harder to stand out, but it also makes it more imperative that you do stand out because there's so much background noise. Um, and, and as a result, even as we're trying to... 7:09 This notion that AI's gonna replace humans, no, it's going to make us even more dependent upon an ever smaller proportion of humans that have maximal humanness. 7:19 [laughs] Because that's the only way you're gonna be able to overcome this overwhelming amount of just blah, this digital gray goo that's being produced by all these platforms. Yeah. 7:28 I mean, in, in, in a lot of ways as we've talked about this, this has democratized access to these things, but in, in another way it's, it's reignited the, the competition in a, in a lot of ways, right? 7:41 I mean, I mean, you, you take... You know, we talked, uh, a few weeks ago about, you know, Google Search business and some of the, the, the legalities and, and, and, and the regulatory issues. 7:51 But, but just from a competitive standpoint, think about, you know, what, what's happening to the advertising business. 7:57 You know, all these, all these companies are fighting, you know, for not only, not only search really, but the ad revenue attached to search and, and really being your, your primary kind of operating system or starting point, you know, when you, when you search for things, when you, when you look for things, when you're trying to get, you know, answers. 8:16 I had, uh, a social media person on here a while ago, said, you know, you know, for, for certain generations, that's TikTok. Like, people actually start their search [laughs] uh- Yeah... for answers on, on TikTok. 8:27 Now you have these kind of agents and, you know, I don't, I don't know if I can even create an exhaustive list, but, you know, off the top of my head, so you got, you got ChatGPT obviously, which by the way is launching a Chrome extension to go after Google, uh, a Mac version to go after Apple. 8:43 They've got a... I think they've already had a Windows version. Uh, so you can do this stuff right from your search bar, right from your, um, uh, the... 8:50 I think the Google Chrome extension changes G- changes, uh, uh, ChatGPT to your, or their search GPT to your default search engine [laughs] within Google Chrome. 9:02 Uh, you have some of these operating system-based, uh, uh, apps. Gemini has been updated, uh, uh, now has access to, to actually what Google [laughs] had access to the whole time, real time, you know, information. 9:15 You have these, some of these new guys, Perplexity. Meta has come out with their own. Uh, we talked about Amazon, obviously Microsoft Copilot, we talked about their announcement a couple weeks ago. 9:26 Uh, and then you got Grok, the Twitter or, or X, uh, one. So, you know, it's a battle for, for, you know, mindshare or search share if you will. [laughs] It's, it's attention-onomics. 9:39 And we've been, we've been watching this attention-onomics warfare for the better part of a quarter century, really ever since search started. 9:46 I mean, you know, my first search ever was on Yahoo, and Yahoo was dominant in the mid 1990s and then Google took over, uh, everyone's eyeballs, so attention-onomics. They became a trillion dollar company. 9:58 Facebook took over attention-onomics, they become a trillion dollar company. 10:01 So I, I, I think the market motivation here is not necessarily that this stuff actually works, it's just the source of eyeball minutes, and you convert eyeball minutes into money. Um, 10:14 so in some ways the extra competition is good, but again, I think it leads us down this, this digital gray goo problem where where, where do you... 10:22 If, if everyone's providing sort of similar stuff and it's all of equally mediocre, uh, output, how do you, how do you differentiate and how do you get better results than that? Um, I, I think, uh, 10:37 generative AI is in- inherently, the physics, the mathematics of it is it is an averageness engine. 10:45 Uh, Bernoulli's, uh, law of large numbers, the bigger the dataset that we train off of, the more average the results necessarily must be. This isn't an opinion, it's math. 10:56 And, and so unless, unless we start combining this core technology of LLMs with other platforms that provide some degree of enhancement, and there's people working on this, things like knowledge graphs, I, I think we're going down a, a road where we're not gonna be really thrilled with the, the, um, the ROI on these trillions of dollars we're putting in beyond the increase in market cap of whoever sort of ends up dominating and, and takes over for, you know, Meta, for the social media era let's say. 11:27 So we're gonna talk, our topic, main topic today is, uh, is AI, uh, ethics. So we're gonna talk a little bit more about some of the, some advancements and some of those concerns, uh, in a moment. 11:37 Before I do though, um, you know, this is not just an AI show, it's a digi- digital leadership show. So couple, co- we don't talk about anything besides AI I think [laughs] the last six months. Yeah. 11:46 But, uh, couple other headlines that, that caught my eye, Chris. Uh, the FTC is finally banning- Uh, some of these, uh, data brokers, uh, cracking down on some of these, uh, these, these data brokers. 11:58 They, uh, announced settlements with the two big data brokers that are accused of selling sensitive information without consent. Uh, Gravy Analytics, uh, i- is, is one, and the other one's called, uh, Mobile Walla. 12:10 Interesting that, I don't know about you, I've never heard of those names before. I know that there's data brokers out there. I know that this is a concern. We've covered it. 12:17 But I don't even know that they've done such a good job of flying under the radar that I don't- I didn't even know the names until I, uh, read this, uh, this article. 12:25 But, but they're, they're being ordered to delete all data from locations such as health clinics, religious sites. 12:32 Uh, they're saying that, uh, they wanna start treating them like, uh, credit bureaus or background check companies that, you know, prohibit them from selling sensitive information like Social Security numbers, uh, location data. 12:45 There was even an article, uh, in Wired, um, I think it was, uh, last week or the, or the week before, that there's even government military assets that they found, uh, in a rep- in their own analysis, uh, uh, that they did, uh, in some of these data broker things, uh, overseas real- you know, where, which has location data for, you know, government, uh, military assets. 13:08 So really, really kind of scary stuff. I, you know, I wonder if this is just cracking the, the surface on this or, you know, if this is really a, a, you know, a move to change this behavior. But, uh, it's something. 13:22 I, I think it's remarkable that, as you say, you've never heard of these companies, but what's in their name? One of them uses the word gravy. The other uses the word wallet. They know exactly what they're doing. 13:31 They know exactly what they're there to do. And while they may have faced some degree of, of penalties now, it probably took a good 10 years for the FTC to get around to penalizing them. 13:42 They might have charged them $10, $20, $50 million, but meanwhile, over those 10 years, they're making $3, $5, $10 billion. And, and I made this argument in my second book, if you're a disruptor- Mm-hmm... 13:53 you use the legal process and the regulatory process. I- if, if it costs me a million dollars a week to fight that legal process, but I'm earning $10, $20 million a week while I do it, that's good business. 14:05 And, and- Well, this is the issue, right? We talk- we, we had that, we had the, uh, you know, that expert, a regulatory expert a- on a few weeks ago. 14:11 I mean, I mean, it, it's almost like, you know, as you know, I'm not a, I'm not a big fan of over-regulation, but this is an area where, you know, I think the legal system, the regulation community has to, has to catch up because this is not new. 14:24 They've been doing this for, like you say, years and years and years. Uh, I mean, it was even- Yeah... 14:28 a John Oliver show, I think, at some point, [laughs] where they, uh, they were able to acquire all this information on US Congress people. 14:35 That's the only reason why they were able to get, uh, any attention for this is because they were, uh, uh, gonna, gonna expose all this private, uh, information on members of Congress. 14:45 But, uh, you know, they're, they're saying that this, the FTC is in, the complaint is saying that these two companies gather more than 17 billion, B, signals, uh, from about a billion smartphones every day. Yep. 15:00 Uh, and that these companies, you know, they're, they're, they're selling location data that's not anonymized, can be easily used to identify, uh, individuals, which has always been the thing that we've always said, "Hey, listen, who cares if your Amazon Alexa is, you know, recording you? 15:15 It's, it's anonymized." They're saying, no, it's not. Uh, you can identify individuals here. 15:20 Um, and, uh, the order, if it holds up, is gonna require them to erase or de-identify historically, you know, historic sensitive information, location information going back three years. 15:30 So, um, seems like a step in the right direction, but, but really, really kind of scary stuff. It, it, it's a degree of a regulatory Pandora's box that we already opened. 15:40 So the, the, I, I wrote about this as the privacy paradox. Yeah, we don't, we wanna maintain our privacy, but I also want kickass coupons. 15:48 Well, the kickass coupons come from knowing what I'm, where I am at this moment and everything, so you can't have it both ways. 15:55 And the degree to which we're in nerd, we're, we're, um, accustomed to that deep personalization experience, we're not gonna let it go, or, or we're not gonna let it go easily. 16:05 Now, I can make you delete the last three years of data and promise you're not gonna do it again, but then what, what... 16:12 And I've been studying this for a better part of a quarter of a century, every single privacy regulation or law that's been deployed around the world always has one caveat: unless you consent. Right? 16:23 I'm gonna protect your data. You know, we can't track you. We can't do any of this stuff unless you consent. Well, what's the very first thing you do with any app that you ever download? Click here to consent. 16:32 And nobody reads the 75-page long end user license agreement. That's why it's 75 pages long. And, and- But at least they're asking consent, although you're right, it's irrele- it's, it's, it's worthless. Yeah. 16:42 But, but, but what they're saying here is they didn't even consent. They basically decided, according- Yes... to the complaint, they basically decided what you just said: "Hey, we're making seven... 16:51 We- we're making how many billions of dollars on this, so come and get us. Slap us on the wrist. Give us fines. It'll take you 10 years. 16:58 It'll take, you know, take you a bunch of legal fees to do it, and in the meantime, we'll rake it in." [laughs] It's, it, it's absolutely a step in the right direction. It's raising awareness. 17:08 Um, and, and it's that sort of, uh, escalating warfare, uh, that we see in cybersecurity, right? The, the, the attackers get more sophisticated, the defense has to get more sophisticated. 17:17 The same is true in the privacy space in, in the legal and regulatory, uh, area. 17:22 Um, I, I think the more that we see that this is sort of a deal with the devil that we all made, and there's the thing we were saying like 10, 12 years ago, you know, if you didn't pay for it, you're the product. 17:34 It's, well, now you can see it firsthand. Yes, you're being sold. We keep track of everything you do. 17:39 Every time I get in my car and I'm about to drive somewhere, I'm freaked out that my iPhone says it's gonna take you seven minutes to go where we already know you're gonna go. Right. 17:46 But that's just the experience that we now have. So, um, it's good to at least demonstrate to people like Gravy and Wallet and, and their ilk that there, there is more oversight coming. 17:59 It's gonna make a little bit higher barrier for them to, uh, to monetize all this stuff. 18:04 But AI is gonna take all of this and amp it up by, uh, uh, at least an order of magnitude or two, so expect to see a lot more on this in the near future. 18:13 All right, so let's get to, uh, I, I guess that's a good segue to, uh, to ethics, right? So, uh, you sent me this article, uh, maybe a week or two back, and now I, I wanna kind of set it up for everybody, right? 18:25 So the, the, the article was, uh, published in The New York Times. It was about, uh, a research, uh, uh, uh, study that, that came out basically saying that ChatGPT defeated doctors at, uh, diagnosing, uh, illnesses. 18:42 So they, they, they said that they, this, the study found that ChatGPT, it was ChatGPT, outdid human physicians when, uh, assessing medical case histories, even when those doctors were using an AI, uh, a chatbot. 18:57 So they, uh, the, the OpenAI chatbot scored an average of 90% when diagnosing medical conditions from a case report and explaining its reasoning. 19:08 Uh, doctors that were randomly assigned to use the chatbot only got an average score of 76% accuracy, which is only 2% better than those who didn't use the chatbot at all, that did it completely the, I guess, the old-fashioned way, uh, on their own. 19:24 So human 74%, human with AI assistance 76%, uh, uh, ChatGPT 90%. But you're saying those results are skewed. There's bias in there, yes? I'm saying that this is a fraudulent article. 19:41 Me and, and dozens, if not hundreds of other people, um, uh, who have looked into this are all in, largely in agreement that this is almost, th- this is as unethical as it gets. And let me, let me qualify that. 19:55 So clearly this is back to attentionomics, right? This is super clickbaity. It's, "Oh my gosh, doctors are gonna be put out of business too. It's not just, you know, baristas and, and journalists." 20:05 Although this journalist should probably be on that shortlist as well. 20:09 Um, but it also kind of reveals the danger of this AI stuff like ChatGPT, where if you don't know enough about the question you're asking, you don't know if the answer you get is right or not, and that's really dangerous. 20:21 In this instance, um, the detail, the devil's in the details if you look at this study, which some of us bothered to actually do. 20:29 And one of the things they say is those numbers you just mentioned, um, are, are median numbers, not mean or average numbers. 20:39 As someone who's done statistics for years and years and years, one of my favorite quotes from Mark Twain, "There's lies, damn lies, and statistics," is if someone's using median as the basis of what they're presenting and not mean, they're doing something probably nefarious. 20:53 Y- y- you, oftentimes we use both of those things when you're trying to make a case for some research you've done, but if you're just using median and never mentioning mean, you're up to no good, and that's what happened here. 21:04 Um, the details of, of this, um, case was they, they only talked to 50 doctors, so there were 50 doctors involved, almost... A- and half of them were in their residency, half of them were kind of post-residency. 21:18 This means they're brand-new doctors, right? And where median comes up is the median of those 50 doctors was three years. Not the mean, which says the average experience is, the median is, is, is, um, three years. 21:32 That can mean that 25 of those doctors have a day of experience. The other 25 have three years and a day of experience, and the median would be three da- three years. So it has nothing to... 21:46 A- and the average of that, by the way, is, like, one and a half years of experience. These are neophytes in, in, out of, basically right out of med school. 21:54 Um, but the mere fact that it's presented that way kind of throws into doubt the entirety of the research. 22:00 On top of that, I think it was six, th- they, they had six cases that were part of the study to see if they were being diagnosed. 22:08 The AI in question was preloaded with, with the pre-canned, supposedly correct results of those six pre-canned cases. The doctors weren't, weren't given any of that. So from stem to stern, the entire study was really 22:24 deeply, not only questionable, but really unethical. And it's great for clickbait. It got tons of attention. Yeah, I think my post about this probably got 40,000, 50,000 views or something. 22:35 Um, but this is the danger of, of these technologies and, and using this stuff, um, and presenting it to audiences that don't have that aren't, aren't gonna have the level of expertise to go in and really look at it and see if this is true, or they're gonna start acting based upon what something like this says. 22:51 It's, it's, it's pretty troubling. Um- So, so I agree, I, I agree with a lot of what you're saying, but let me, let me push back a little bit. 22:57 I mean, this w- this was not just a New York Times clickbait kind of article, right? 23:01 This w- this was published on, uh, the, the JAMA Network, which is a peer-reviewed open access medical journal, uh, subsidiary of the, the American Medical Asso- Association. 23:10 So this wasn't just some, you know, reporter who dug into some, some nonsense, right? So why, why, why is it, why is this being... 23:21 I mean, what, I, I would assume, maybe it's a bad assumption, I would assume that, you know, anything that pub- it's published by the American Medical Association would not be anti-doctor and pro-technology. 23:31 I, you would assume that. Um, there's kind of an old saying in, in the research field, there, there's remarkable consistency between the findings of research and the preferences of those who fund the research. 23:44 [laughs] So, so that might be part of it. I'd be tempted to go back to see who's, who's paying these PhDs to do this research, and that might be pretty indicative of why they came up with the findings that they did. 23:53 Also, there's just an urgency to not only get published, if you're a researcher, but to get published big, to make a splash, and certainly this was splashy, um, both in terms of its initial supposed findings, but also in the backlash. 24:06 So unfortunately, I forget who, who said this, but, you know, all, all, all publicity is good publicity. Um, shining a light on this, um, m- may actually be good for them to a certain extent. 24:18 But nonetheless, um- It does call... Your point is very valid. Why did JAMA let this get through and not catch this? 24:27 That's a really good question that I think many of us should be asking, particularly in an age when we're all jumping on ChatGPT asking it, "What color is George Washington's white horse?" And it says black. 24:39 Well, I think what's scary about- I don't know that the white horse at war is white... what's scary about it is... 24:42 Yeah, what's scary about it though is, and I think I even said this on a show, I was, I was reading something recently, uh, it was a psychology thing. 24:50 Uh, and the, and the person, you know, and this was like 20, this was like a 20-year-old book, uh, was basically saying, you know, "Hey, be careful because everything that you..." Talk about attention-omics. 25:00 "Everything that you Google, um, you know, has a motivation," right? Yes. The first five pages have a motivation. 25:06 And so, in all likelihood, the, the, the reputable research journal is probably on page 10 of the search results, and that's the legitimate one. You're saying that, that we can't trust that. 25:18 Th- th- this is totally by design. I mean, we, we, we're all deeply compelled by a lot of the stuff that we see out there. That's not an accident. That's... People are making a lot of money doing that. 25:29 Um, and it's funny, you know, you're saying psychologists say this. It's, it's like, it's like having a study where 90% of psychologists agree that 90% of their patients are crazy. You ever hear of selection bias? 25:40 I mean, [laughs] that's why they're coming to see you. So, uh, it's the need to be even more judicious in how we use our attention and what we do with our attention. Am I gonna believe this or not? 25:53 The stakes have never been higher, and it's gonna get harder. Back to the digital gray goo. 25:57 The higher the piles of digital gray goo get, the more imperative it's gonna be that we, we step back a little bit and be critical, and really think, "Does this pass the sniff test with me?" 26:08 And then look at the details, 'cause the devil will be in those details, and if there's something kind of questionable, it, it should hopefully leap out pretty quickly. 26:16 So is your belief just that this is over-hyped and, and, and we're not there yet? Or you're saying that we'll, we'll never be there, that, that the AI will never... 26:24 I guess never is a, an unfair word, but you know, in the near future not gonna be at a place where, you know, the AI will actually be smarter than the doctors? I don't think we yet understand what the word smarter means. 26:40 It's a moving target, and we keep changing it. I think that this technology is like every other technology I've worked with in the last 30 years of being in IT. It's use case specific. It's... 26:55 A- and the other thing I firmly believe, as an old rocket scientist, is that the second law of thermodynamics covers nearly all of this, right? You do not get something for nothing. 27:05 Every single project I ever implemented in my entire career supposedly had a positive ROI. That's not possible. 27:13 Second law of thermodynamics says you, there's a co- there's always a greater cost to doing something than the cost of doing nothing. It's ca- It, it, it's entropy. 27:21 Um, the only time I, I firmly believe this, the only time you get a positive ROI out of a, a i- information technology is if you're buying into a growing market, if it lets you get a bigger proportion of the pie, either because the pie is getting bigger or you take a bigger slice. 27:37 As long as you are investing the IT into building market share or market size, you can get a positive ROI. 27:45 But if I'm just implementing a technology to, like, replace some workers with, with some tools, the ROI is never positive. Wow. And I, I'll, I'll, I'll die on that hill for the rest of my career. I'm convinced. 27:56 All right, we got, we got a topic for our next show. I like that. [laughs] I wanna, I wanna, I wanna unpack that one. I have another 45 minutes. But, uh, okay. Well, so couple of... 28:03 Before we move on to the, the, the other example of, you know, the question of ethics and A- AI, um, two, two things that I do think are relevant, you know, whether you believe this report or not, I think are, are, are interesting. 28:16 One is, um, 28:18 when they were trying to explain why the doctors that had the, uh, the help of the assistant only performed 2% better than the ones without the assistant, uh, the, the rationale was that the doctors were not persuaded by the chatbot when it pointed out something that was at odds with their diagnoses. 28:36 So, you know, in other words, they, they, they tended to be wedded to their own idea of the correct diagnosis, which you could argue is a whole different type of bias that we have to address as technologists to say... 28:48 And I was actually reading something last night, uh, I gotta find out. I'll put the, put it in the show notes if I can find the, uh, the exact page number and, and where it came from. 28:56 But, uh, it was actually talking about, you know, there's some, some built-in, you know, uh, qualities of our humanity that make us stick with or make our brains stick with things that we've become accustomed to, to believe, that we're convinced of, uh, and, uh, less likely to think of new, new ideas and, uh, change what we've always established as true and, and those sorts of things. 29:19 So, so that's interesting. As we, as we do more and more AI, like you say, you know, is it even helping us if we're gonna second-guess it and have to do all the work ourselves anyway? 29:29 Uh, the other, the other thing that I thought was iss- uh, was interesting is that, uh... and this one we've, we've definitely talked about before. 29:35 The, the other issue that, that they used to explain it was that many of the doctors didn't know how to use the chatbot to its fullest extent, so it gets to the, the training thing we've talked about a lot, right? 29:46 A- and both of those are true. I mean, um, gosh, it was, uh, almost a decade ago, 29:53 three, three to four years prior to tr- Transformers and LLMs even being a white paper, that Watson, IBM's Watson was finding the exact same, th- this sort of, not... 30:03 Egotistical is the wrong word, but this ego investment that doctors are gonna have in their diagnoses. 30:08 In, in 2013, 2014, IBM did this whole research study with their Watson on I- um, going through x-rays, uh, and mammographs to look for, um, breast cancer. 30:19 And back then, again, long before we had Transformers, this was just machine learning. It was, it was found that the, the technology generally was... 30:28 The technology back then wasn't better at coming up with the initial diagnosis. What it was phenomenal at was coming up with- Third, fourth, fifth possible diagnoses which the doctor never would've thought of. 30:40 I remember that. Okay. 30:41 So the, the, we're, we're kind of enhancing these same biases- Because they weren't even used to those, 'cause so- some of them were people that came from other countries, other regions, and they weren't ex- the doctors weren't experienced in those regions maybe, uh- Or, or, or, or, or it just, as soon as they, as soon as their mind locked into what they thought the diagnosis was, they're done. 30:57 I mean, this is what made House such an amazing TV show and such a great series, right? Because he didn't take the first diagnosis. He kept digging down the rabbit hole. 31:05 So he was basically just a, a big machine learning British guy trying to fake an American accent. Um, but it was very, very effective. And so th- this bias amplification is inherent in all of these technologies. 31:18 It's unavoidable. It is manageable. But I'm... This is why I always refer to particularly LLMs and transformers as, as bias amplifiers, because what... 31:28 If, if machine learning 10 years ago made the bias slightly worse, this stuff makes it a million times worse. Think Google Gemini last summer, right? 31:36 Um, and, and, and so it's, it's an operational reality, and it's among... 31:42 So when, when I'm having all these AI governance discussion, and this is throughout my upcoming, uh, book that I co-authored on AI governance, you need to be aware that this bias amplification effect is there, and then you need to work to mitigate the damage that comes from it, or the harms. 31:58 You can't stop it, but you need to have kind of the guardrails around it to make sure that it's under control when it happens. 32:06 So the, uh, the other blatant example of, uh, you know, the question over, uh, AI ethics is, uh, this, this article in, in Wired magazine, uh, saying based on, uh, some research that they, uh, were reporting on, uh, over 54% of longer English language posts on LinkedIn are generated by AI. 32:33 Uh, it's based on a AI detection startup called Originality, uh, AI. So that's... You know, I'm not a mathematician, uh, like you, but, uh, that's more than half, Chris. 32:43 That means, uh, [laughs] everything that you're reading, uh, online, you know, was generated AI. So, so it really brings into question, you know, all these things we're talking about, right? 32:53 Uh, uh, creativity, s- you know, authenticity, uh, accuracy, um, and you know, you know, I think of LinkedIn, maybe it's 'cause I do a lot of publishing and, and stuff like this, but, you know, I think of LinkedIn as, um, 33:10 you know, aside from just a, you know, networking platform, as an opportunity to kinda find experts in certain areas, demonstrate expertise in certain areas, right? 33:19 We, we do this because we want executives who are, you know, in the automation, AI, digital transformation space to have, you know, free resources and, and access to us and find us, right? 33:31 But if you're not the one creating it anymore, I mean, back to differentiation again, right? I mean, I mean, what... A- aside from, hey, you know, is half the internet just made up? 33:40 Um, and I think about what you said too about the, the race to mediocrity or, or, or the race to average, right? Where, you know, all the content starts kinda looking the same, uh, whether it's accurate or not. 33:51 But, you know, you know, how do we, how do we... 33:54 If we can't trust a medical journal and we can't trust, uh, what, what's being fed to us through a search, and now I go online and I find my experts that I trust on LinkedIn, and the stuff they're publishing isn't even original and that wasn't even theirs anymore, now what, right? 34:10 I, I, I think untrust or de-trust is gonna be the survival skill of the future. And I'm just as guilty. I mean, a, a lot of what I post on LinkedIn is a repost to point out this sort of stuff. 34:22 So in, in many ways I'm contributing when I do that. But what I, I'm also trying to do is when I identify something that's most likely this sort of... It's average amplification, right? 34:33 Someone auto-generated something to get clicks, then someone like me comes along, like with this, this medical, uh, article and say, "Hey, this is BS. Don't, don't follow it." So when... 34:43 To a degree I'm, I'm adding to that a- average's amplification, but I'm also by, by... 34:49 And frankly, I, I think as you well know, I take a lot of counterpoints on this stuff and kind of stick my, stick my fingers where they don't belong in many people's views. 34:57 But by doing that, I'm, I'm hopefully contributing that nth percentile, that little percentile of humanness, of, of the expertise that people are looking for to say, "Hey, wait a minute, something's not right here." 35:08 And so we're, we're going to need... Now this is, this is great use case for people, and it seems like these guys are heading this direction. 35:15 If I, if I can identify where all this stuff is the same or it's copied or it's digital gray goo, then I can start looking for the what's left over, and the real gold is gonna be in the leftovers. Right? 35:26 When I, when I dig a gold mine, 99% of the dirt coming out of the ground is worthless. I gotta figure out how to throw that away and get, and find... 35:34 That's all the tailings that I throw away, and then the, the gold is what I'm separating out to figure out. So people looking for a good use case with this stuff, there you go. 35:43 How do I, how do I wade through the digital gray goo, which is greatly expanding and, and self-replicating as we're seeing here, and then find that meaningful, uh, chunks of gold that are in there? 35:55 Um, phenomenal use case for people that can figure it out. But, you know, I, I, I just wonder about... And it's, you know, there's, there's so many different dyna- di- dimensions to this, right? 36:05 You know, there's a generational dimension to this. There's a, you know, back to our, you know, search wars conversation at the top of the show. Yeah, you have that dimension of it. 36:14 But if, if you get to a point where, okay, all, all these different companies are fighting f- to be your, your assistant, right? And they're feeding you information that's not 36:26 expert content, and then you go to your so-called experts and their content isn't, isn't, isn't even a, a, a original. Um, at some point You know, are you losing access? 36:41 I mean, I know you're a big fan of critical thinking, [laughs] right, as I think everybody should be. Um, are we losing the ability to, to think critically? Are we trusting too... 36:49 I mean, may- may- I mean, in some ways, I guess you go to the doctor and you're not feeling good, and he diagnoses you, and you trust him. And, and maybe we trust them too much. Maybe we should get multiple opinions. 36:59 Maybe we should ask AI too. But our first point of contact is, has been to go to an expert. Maybe that's not perfect, but that's what's worked as a society for a long time. 37:09 Um, now if your first point of contact is a chatbot, uh, you know, then, you know, how do we validate all this? And then, and then is it even, is it even benefiting us? 37:20 'Cause to your point earlier, if you, [laughs] if you are, either, either you're blindly following it, which is a problem, or you gotta do your own research anyway, and then it's, is it, is it even valuable at that, at that point? 37:33 I, I am biased, so I'll put that out there. Ever since I, you know, when I was building spacecraft, I was the guy that figured out how, how, how do we not blow up on the way to Mars, instead of how do we get there? 37:43 And so that's colored my entire career. But I, I do think that the next generation of real expertise is gonna be untrust, detrust, mistrust, and how do you figure out what to believe and what not to believe. 37:56 And to your point, yeah, I, we're, we're... I wouldn't say we're breeding out, uh, critical thinking, but we're certainly de-emphasizing it. Um, the, the... 38:06 I, I saw something earlier today on LinkedIn, people talking about this, where children are, are, are becoming emotionally bonded and wedded to AIs. 38:15 Um, I see in my own, my, my six-year-old daughters in their imaginative play this morning. I heard them, they were imaginatively playing that they were asking questions of Siri. 38:24 I don't use Siri, but I know somewhere in their world, someone is using Siri to figure out how to interact with the world. 38:31 So even in environments where I'm trying to keep my kids from that kind of stuff, it's still working its way in. 38:37 So I, I do think that the future of expertise is this sort of mistrust, detrust, untrust, um, so that some- you, you have someone to go to to make sure that what you're being fed is legit. 38:50 Good- And then, and then if that's true, and all of the young, younger people are being trained and raised in this belief of these Google search results, and now what ChatGPT tells me, how do they build expertise to become that mistrusting expert in the future? 39:06 Because they're, it was, it was never nucleated in their thought process. 'Cause like it or not, it's winning, right? The, the AI- Yeah... is winning. 39:15 I mean, to say 54% of posts, I'm sure that number only goes up, not down. LinkedIn has been always kind of defensive about automation on their platform. They, they seem to be embracing this. 39:25 I mean, they're, they're doing some copyright checks and, and, and, and that kind of stuff. But for, uh, for the most part, they're, they're, uh, you know, endorsing this, supporting this, and, and, and because, right? 39:36 The, it's, it's the atten- like you say, attention economy. It's more traffic for them. It's more content for the people that are publishing content, trying to monetize content. 39:45 Substack is also reporting that the top 10% of its newsletters have some AI-generated or AI-assisted content in it, so it's not just LinkedIn. 39:53 Uh, and as the, the search wars, uh, you know, heat up, you know, like we said before, it's all, it's all AI-based. 40:00 I, not to geek out, but I mean, again, for those with the expertise, and I've spent decades figuring this stuff out, it takes me back to Shannon's information theorem, which when I was getting into e-discovery and natural language processing, it took me three tries to understand that, that thing, but I eventually did. 40:17 Um, the less... 40:19 One, one of the pres- one of the, uh, concepts out of Shannon's information theorem is the less often something appears in a body of information, the more relevant it is to describing that body of information. 40:30 So the extent to which the digital Guri-gu is taking over and all this auto-generated AI content seems to be overwhelming us, it actually will make it easier to discover true creativity, true expertise, true facts, 40:44 but we need to be designing systems to apply Shannon's information theorem and similar concepts to eke out, again, the gold, separate the gold from the, from the tailings. 40:54 It's gonna be more and more of a challenge as we produce bigger and bigger piles of blah, you know? It's, it's, um, uh, you know, th- this digital verbal incontinence that we're producing at an industrialized scale, 41:07 it's, it's gonna be harder, but the people that can figure that out, that's a, um, an amazing opportunity for them. Uh, great conversation, Chris. Thanks, uh, thanks as always. Uh, thanks for joining us, everybody. 41:18 We're back on Wednesday at noon Eastern, 9:00 Pacific. 41:21 And in the meantime, there's anything you're working on, you want some free advice, or just to learn some art of the possible from our team of subject matter experts like Chris, schedule a free networking call at erpaai.com/networking. 41:33 Have a great week, everybody. Chris, stay out of trouble, buddy. Yeah, maybe. See ya. 41:38 [laughs] Yeah. 42:00 [outro music]