Latest / The Adults In The AI Room / Human Cognitive Science - Daniel Faggella - Earley AI Podcast with Seth Earley & Chris Featherstone - Episode # 026

Transcript

Human Cognitive Science - Daniel Faggella - Earley AI Podcast with Seth Earley & Chris Featherstone - Episode # 026 · The Adults In The AI Room

• 0:03
  SPEAKER_00: Welcome to the Early AI Podcast.
• 0:07
  SPEAKER_00: Here we talk with thought leaders and practitioners of what's possible and practical with AI and the enterprise as we move from the current state to a future of cognitive assistance and embedded AI in all aspects of our personal and professional lives.
• 0:22
  SPEAKER_00: Thanks so much for listening.
• 0:23
  SPEAKER_00: If this is your first time, we invite you to subscribe wherever you listen to podcasts.
• 0:28
  SPEAKER_00: And now here are your hosts, Steph Early and Chris Featherstone.
• 0:35
  SPEAKER_03: Welcome to today's podcast.
• 0:37
  SPEAKER_03: I'm Steph Early and I'm Chris Featherstone.
• 0:39
  SPEAKER_03: It's good to be with you.
• 0:41
  SPEAKER_03: Nice to see you, Chris.
• 0:42
  SPEAKER_03: And uh the weather up here in uh New England is fantastic.
• 0:46
  SPEAKER_03: We're starting to see glimpses of spring.
• 0:49
  SPEAKER_03: So I'm very, very happy about that.
• 0:50
  SPEAKER_03: How's it looking in your your worlds?
• 0:53
  SPEAKER_02: Well, I appreciate you picking at the scab because you know uh in Montana it's still cold.
• 0:58
  SPEAKER_02: Uh we're getting we we we had some snow flurries that warmed up, but I'm supposed to get more snow on Sunday, on Saturday and Sunday.
• 1:06
  SPEAKER_02: So good lord, we needed I need this to end.
• 1:08
  SPEAKER_02: But I am only two hours from Canada, so the frozen tundra comes and affects us a lot.
• 1:14
  SPEAKER_03: What I did was I just reset the stakes to outline my driveway because they got knocked down in the last plowing.
• 1:20
  SPEAKER_03: So I'm sure that means we'll have no more snow as soon as I got that done.
• 1:25
  SPEAKER_03: Well, let's go ahead and jump in.
• 1:27
  SPEAKER_03: Today's guest is a black belt and national champion in Brazilian jiu-jitsu.
• 1:33
  SPEAKER_03: I've done a little bit of Brazilian jiu-jitsu to know that it is a very challenging sport and the workouts were brutal.
• 1:40
  SPEAKER_03: Uh, he's been called upon to speak about the implications and applications of emerging technologies uh around artificial intelligence by organizations like the World Bank, United Nations, Interpol.
• 1:52
  SPEAKER_03: I don't know if I'd want to go and talk to Interpol, they might find something on me.
• 1:56
  SPEAKER_03: In global, I have a story about going into Canada that I won't talk about now.
• 2:01
  SPEAKER_03: In global enterprises and banking, life sciences, and retail and other industries.
• 2:06
  SPEAKER_03: He currently lives in Western Massachusetts.
• 2:09
  SPEAKER_03: Please welcome CEO and head of research at Emerge, uh Dan Fagella.
• 2:14
  SPEAKER_03: Fagella.
• 2:15
  SPEAKER_03: Now, Dan, how do I pronounce your last name?
• 2:17
  SPEAKER_03: I should have known that.
• 2:17
  SPEAKER_01: I'm I'm one of those guys that doesn't really care.
• 2:19
  SPEAKER_01: It's it's Fagella technically, but even if you said flagellum, I wouldn't even correct you because I just I'm not I'm not all that uptight about it.
• 2:26
  SPEAKER_01: So don't worry about it.
• 2:28
  SPEAKER_02: So let me ask you something though, because uh there's there's kind of two um you know good rules of thumb that I was instilled, you know, by my parents at a young age.
• 2:38
  SPEAKER_02: First, you know, use the golden standard, right?
• 2:40
  SPEAKER_02: Treat others how they want to treat you.
• 2:42
  SPEAKER_02: Second, right?
• 2:43
  SPEAKER_02: Goes without saying, second, always beware of people with cauliflower ear.
• 2:50
  SPEAKER_02: Because they're just people that you want to have as friends, not as enemies.
• 2:57
  SPEAKER_03: Yeah.
• 2:57
  SPEAKER_03: When I when I when I met you, and it's funny, you they just mentioned we met at the AI um accelerator institute conference in Boston.
• 3:06
  SPEAKER_03: And I remember when I met you and I was talking to you, and I noticed that and I was like, oh, okay, man, this this guy, this guy's a fighter.
• 3:12
  SPEAKER_03: You know, this guy's a fighter.
• 3:13
  SPEAKER_03: Yeah.
• 3:14
  SPEAKER_03: Whether it's uh trained or or or not.
• 3:19
  SPEAKER_03: So tell us about your background.
• 3:21
  SPEAKER_03: Tell us about um about your training.
• 3:23
  SPEAKER_03: I told you I had to take time off because of my my shoulder uh yeah, you got an injury.
• 3:30
  SPEAKER_03: I haven't done uh uh jujitsu in a long time, but I would love to get back to it.
• 3:34
  SPEAKER_03: We'll do some Aikido and some Shotokan.
• 3:36
  SPEAKER_03: But tell us about your history there.
• 3:38
  SPEAKER_01: Sure, weirdly enough, very strange.
• 3:41
  SPEAKER_01: Uh uh jujitsu has actually ended up being my connection into artificial intelligence, so we can bridge that gap too.
• 3:48
  SPEAKER_01: But um, yeah, I started training when I was like 17 or so.
• 3:52
  SPEAKER_01: I did a little bit of wrestling in high school, but I didn't I didn't like it a lot.
• 3:55
  SPEAKER_01: I was pretty good, but I I didn't like it a lot.
• 3:57
  SPEAKER_01: I like jujitsu um because there's so many more ways to win and it's so much more strategic.
• 4:02
  SPEAKER_01: Uh and in jujitsu, I could compete against somebody twice my size, and even if they could stay on top of me the whole time, I could still win the match.
• 4:09
  SPEAKER_01: And so I did a lot of absolute divisions against really, really big people.
• 4:13
  SPEAKER_01: Uh, and I just I thought it was a a different kind of challenge with a lot more of a chess game to it, and I fell in love with it.
• 4:20
  SPEAKER_04: Yeah, totally, totally great.
• 4:22
  SPEAKER_02: Yeah, as I say, as they say, angle strangle, right?
• 4:26
  SPEAKER_01: So yeah, that's it.
• 4:28
  SPEAKER_01: For me, it was a lot of leg locks, uh, but certainly a lot of back takes and and chokes as well.
• 4:33
  SPEAKER_01: That was that was my life.
• 4:34
  SPEAKER_01: I I love it.
• 4:35
  SPEAKER_01: Um, but I did that for seven years.
• 4:38
  SPEAKER_01: Um, got my black belt, I think a little under seven years, and uh and I ran a gym to pay for graduate school.
• 4:46
  SPEAKER_01: So I went to UPenn for grad school, which is actually where I got introduced to AI, although I was there studying cognitive science um for my master's, and um paid paid my way through overly expensive uh graduate school with uh jujitsu gym.
• 4:58
  SPEAKER_01: And then I sold that when I turned 24 and got into the e-commerce business that eventually I sold to start what I do now.
• 5:06
  SPEAKER_01: Um and that that e-com business that was what had me in in EO.
• 5:10
  SPEAKER_01: So when I was in regular EO out there in the Bay Area.
• 5:12
  SPEAKER_03: So EO, for those of you who are not familiar, is the entrepreneur organization and uh it's an organization that that uh Dan and I both belonged to in the Boston chapter.
• 5:23
  SPEAKER_03: And then of course we met uh a few months ago at the uh AR Accelerator Institute, and then we ran into each other at the uh workshop the other day.
• 5:32
  SPEAKER_03: So that was great, very pleasant.
• 5:35
  SPEAKER_03: Small world.
• 5:36
  SPEAKER_03: You know, it's a very small world.
• 5:38
  SPEAKER_03: So tell me how your journey into AI started.
• 5:41
  SPEAKER_03: You said you started that with one of your other endeavors, but what was the connection point there?
• 5:45
  SPEAKER_01: Yeah, I'll I'll give you the idea here.
• 5:47
  SPEAKER_01: So um my uh I went to graduate school at UPenn studying, so positive psychology is sort of the program at UPenn that Martin Seligman started.
• 5:57
  SPEAKER_01: He's kind of the father of that field of uh psych.
• 6:01
  SPEAKER_01: My particular interest was more in um skill acquisition, skill development.
• 6:06
  SPEAKER_01: So they're really not there's a sports psych program in Springfield, which is actually quite good, uh, which I got into, but wanted to go to Penn.
• 6:14
  SPEAKER_01: Um uh, but there's not many programs where we can say that stuff.
• 6:17
  SPEAKER_01: So I went to UPenn to uh study skill acquisition, skill development as my thesis for my master's.
• 6:24
  SPEAKER_01: Um, got to interview all kinds of great folks, the founders of goal setting theory, the really the father of skill acquisition as a science, a guy by the name of of uh uh uh Andres Erickson, who unfortunately passed away recently, but um did a bunch of that.
• 6:40
  SPEAKER_01: And while I was there, I was hearing from people in the computer science side of things saying, hey, you're studying human learning, you should check out this machine learning.
• 6:49
  SPEAKER_01: And this is back in 2011, where it's like, well, what's machine learning?
• 6:52
  SPEAKER_01: Um, and that uh UPenn was partnered with Google for something called Google Zeitgeist at the time, I think that's what it was called, where they were combing social data and doing kind of basic sentiment analysis and things today that you know high schoolers would do with it with machine learning, right?
• 7:06
  SPEAKER_01: But back then it was like very special, very interesting.
• 7:09
  SPEAKER_01: And um, I remember seeing some of this and then thinking about the longer-term implications, reading Ray Kurtzweil, reading Nick Bostrom, and essentially becoming convinced that in the next 30 years, uh, there's really nothing more important than the trajectory of intelligence itself uh ascending to sort of grander level.
• 7:27
  SPEAKER_01: So the whole kind of you know, post-human transition artificial intelligence thing seemed to me as if there wasn't anything else that I should think about kind of from then until I die or get assimilated into the machine.
• 7:38
  SPEAKER_01: Um, and so uh yeah, that dawned on me at the end of grad school.
• 7:43
  SPEAKER_01: But here's the issue, Seth.
• 7:44
  SPEAKER_01: Um, by that time, I was running a martial arts gym.
• 7:47
  SPEAKER_01: I was selling some martial arts DVDs on the internet because I wanted another income stream, but I had no business in AI.
• 7:53
  SPEAKER_01: And in fact, I was trained as a cognitive science guy, so I had no background in AI.
• 7:57
  SPEAKER_01: So this is nine, 10 years ago, I just started interviewing startups with no idea of my direction at all.
• 8:04
  SPEAKER_01: I found scientists and neuroscience and AI.
• 8:06
  SPEAKER_01: I found people doing business applications, and I started interviewing them aimlessly.
• 8:10
  SPEAKER_01: And for three years I had no audience whatsoever.
• 8:13
  SPEAKER_01: Eventually, I sold that e-commerce company, um, moved back to the East Coast, and turned it into a market research firm.
• 8:20
  SPEAKER_01: Did work for the the World Bank was our first kind of formal, you know, uh market research slash kind of global speaking engagement.
• 8:29
  SPEAKER_01: And then from there it rolled into a lot of other other work.
• 8:31
  SPEAKER_01: So um yeah.
• 8:33
  SPEAKER_03: Nice.
• 8:33
  SPEAKER_03: And what were the big takeaways at the time from your uh studying of of human uh cognitive science and uh and how that impacted, you know, how machines learned?
• 8:44
  SPEAKER_03: What were some of the insights or some of the takeaways from that?
• 8:47
  SPEAKER_03: There's a lot of confusion, you know, when people say things like, oh, the technology works like a human does, and you know, neural memory, right?
• 8:56
  SPEAKER_03: And you know, in my counter to that is the human brain is so much more complex because you have hundreds of neurotransmitters, they're all analog.
• 9:05
  SPEAKER_03: One neuron can be connected to 10,000 others, you have three billion of them.
• 9:10
  SPEAKER_03: The combinatorial complexity and explosion is just beyond what we can comprehend.
• 9:15
  SPEAKER_03: So, but there are some analogies, but the analogies do stop, right?
• 9:20
  SPEAKER_01: I think well, Kurtzweil has a good book called How to Build a Mind.
• 9:24
  SPEAKER_01: Um and I actually uh I have a tremendous amount of respect for Kurtzweil.
• 9:30
  SPEAKER_01: I don't think he's right about everything, but um, I think he's on point about a lot.
• 9:34
  SPEAKER_01: And the rough, the rough analogies around some elements of of uh ML and kind of human learning are are are relatively strong.
• 9:44
  SPEAKER_01: The idea of kind of neurons and the idea of training on instances, there are actually direct correlations.
• 9:50
  SPEAKER_01: And curiously, I don't see this in the computer science literature, but in the skill acquisition literature, there are basically equivalents to underfitting and overfitting uh for problem spaces.
• 10:00
  SPEAKER_01: Like there's exact equivalents.
• 10:01
  SPEAKER_01: So, and the places where this is studied are things like musical performance, sport performance, memorization, things where uh skill from level A to level B can can be in some way measured.
• 10:12
  SPEAKER_01: Um, so there are some interesting correlations, but uh clearly we can't boil the mind down to linear algebra just yet.
• 10:21
  SPEAKER_01: But I would say the rough analogies, you know, I don't get offended by the rough analogies because some of them are pretty strong.
• 10:26
  SPEAKER_01: And the fact that there are direct neuroscience uh correlates, skill acquisition correlates to overfitting and underfitting is really interesting.
• 10:34
  SPEAKER_03: So um in and I and I I I do agree that there are analogies, but then they too go a little too far sometimes.
• 10:43
  SPEAKER_01: Sure, sure.
• 10:44
  SPEAKER_01: Yeah.
• 10:44
  SPEAKER_03: I mean, we can't if we treat the analogy it's not exactly the same, but you know, if if we treat them literally, right?
• 10:51
  SPEAKER_01: If we say, well, the brain is a computer, right?
• 10:53
  SPEAKER_01: I mean, of course, that's the most ridiculous thing you could possibly say.
• 10:57
  SPEAKER_01: Um but uh to say that you know, the way we train a machine today to detect breast cancer is more like a human brain than uh, you know, the old expert systems in 1982, would you know that you could say more, and I'm not offended.
• 11:13
  SPEAKER_01: If you say they're the the same as a human brain, I would say you're you're definitely wrong.
• 11:21
  SPEAKER_03: So um, so these days, you know, we start to see a lot of um hype uh around uh you know the new and emerging shiny bits, right?
• 11:33
  SPEAKER_03: Obviously, the chat uh GPT type of phenomenon in terms of generative AI, and generative AI is very powerful.
• 11:40
  SPEAKER_03: But the other day I was reading an article about from somebody who called himself the AI guy.
• 11:47
  SPEAKER_03: Like there's only one, right?
• 11:49
  SPEAKER_03: I am I am the AI guy, right?
• 11:51
  SPEAKER_03: Whatever.
• 11:51
  SPEAKER_03: Yeah, and and one of the things he said is it gives you precise, correct answers.
• 11:58
  SPEAKER_03: And we know that generative AI does not always give you precise answers.
• 12:02
  SPEAKER_03: They give answers that look like they could be right, but they're not right.
• 12:05
  SPEAKER_03: I just wrote an article about this, and it was giving me all sorts of uh, you know, all sorts of credentials that I didn't have.
• 12:13
  SPEAKER_03: You know, it was giving me, you know, say I wrote these books and I taught at this university, and I was like, oh wow, those are really impressive.
• 12:19
  SPEAKER_03: That could be that sounds reasonable, but it's wrong.
• 12:22
  SPEAKER_03: It was completely wrong, right?
• 12:23
  SPEAKER_03: So maybe we can talk a little bit about, you know, the the hot topic of the day, which is generative AI.
• 12:31
  SPEAKER_03: And then let's back into kind of, you know, we're organizations really should be in terms of their learning curve, in terms of their capability development.
• 12:38
  SPEAKER_03: Both what are your thoughts so far when you start seeing all the all the hoopla?
• 12:44
  SPEAKER_01: Well, I I mean, I've I've been excited about this uh for a long time.
• 12:50
  SPEAKER_01: So I have a seven-year-old article called Lotus Eaters and World Eaters um around the paradigm that I believe now we're crossing a chasm into.
• 13:00
  SPEAKER_01: It's a very old article.
• 13:01
  SPEAKER_01: Uh the this notion that um eight years, that's like you know, yeah, yeah, yeah, yeah, yeah.
• 13:09
  SPEAKER_01: It's a long time.
• 13:10
  SPEAKER_01: And my first talk about this, you mentioned Interpol, was for Interpol, four years ago.
• 13:14
  SPEAKER_01: So four years ago, there was some really they they wanted to know about surveillance and computer vision.
• 13:20
  SPEAKER_01: And what I was trying to convey is I I back then there was no name for this stuff, right?
• 13:25
  SPEAKER_01: Generative AI was not a term.
• 13:26
  SPEAKER_01: So it was surveillance technology, creation technology.
• 13:30
  SPEAKER_01: I I didn't I didn't have a keyword for it at the time.
• 13:32
  SPEAKER_01: But back then, the MIT, some C Sale lab had, you know, they take a screenshot of a beach or a wave crashing on a beach and then have the machine generate the next 10 slides.
• 13:42
  SPEAKER_01: So what are the next 10 frames?
• 13:44
  SPEAKER_01: Um, and then do that with a duck walking on grass and do that with a bunch.
• 13:48
  SPEAKER_01: And it was goofy, but I had seen the progression of, you know, generating faces, which four years ago faces were still pretty goofy, but they were much better along than they were two years prior.
• 13:58
  SPEAKER_01: And so um, looking at that and looking at a bunch of correlates, uh, NVIDIA was doing a bunch of things half a decade back that were also, you know, child's play compared to today, but very interesting.
• 14:09
  SPEAKER_01: Um, so I've been on this tip for a while.
• 14:11
  SPEAKER_01: Uh, and I would say um there's two directions we could go here.
• 14:17
  SPEAKER_01: Um, there is the big game of where this takes us, uh, which ultimately is you know what I'm probably most interested in.
• 14:24
  SPEAKER_01: But but there certainly from the business perspective, there's also the the near-term use cases within banking, lifesi, whatever.
• 14:31
  SPEAKER_01: Where do you want to land first?
• 14:32
  SPEAKER_01: And I'll just kind of go along with you because there's a lot of directions.
• 14:35
  SPEAKER_02: Let's double click on that last point because before the big picture pieces, because this is you know, it's always interesting.
• 14:41
  SPEAKER_02: I always get super, you know, you hear that you see the tea lease, but then you start to hear people talking about it that have no business in technology or whatsoever.
• 14:49
  SPEAKER_02: And they're like, Well, you know, I'm just gonna have uh Chat GPT write that for me.
• 14:55
  SPEAKER_02: I'm like, oh hell, here we go.
• 14:56
  SPEAKER_02: You wanna be a subset of, you know, you go through the the rig and row of trying to explain why it could be successful and it could very probable probabilistically be, you know, unsuccessful, right?
• 15:09
  SPEAKER_02: So you know, so I want to double-click on that, on what, in your opinion, what companies should be thinking about today in regards to this, how to not necessarily embrace it, but how to corral it so they can get to the right perspective to know how to get to the next generation piece of that, which is the bigger picture.
• 15:31
  SPEAKER_02: Does that make sense?
• 15:32
  SPEAKER_01: Sure.
• 15:32
  SPEAKER_01: Yeah, yeah, totally.
• 15:33
  SPEAKER_01: So let's let's kick off there.
• 15:35
  SPEAKER_01: Um, and just for context, I'm not a futurist by trade.
• 15:39
  SPEAKER_01: I never called myself one.
• 15:40
  SPEAKER_01: So the only reason I've ever been in UN headquarters, the only reason I've ever been at an interpol event, the only reason I'm going to the OECD in Paris uh in four weeks is because um my whole life, uh, 80 hours a week, is talking to people in the in the Fortune 500 who are doing this stuff and then talking to basically the unicorns.
• 15:58
  SPEAKER_01: So the companies that are mature enough to have boots on the ground use cases, who's spending money and who's impacting workflows.
• 16:05
  SPEAKER_01: At the end of the day, when the when the night comes, that's all I do.
• 16:09
  SPEAKER_01: So that's the only that's the only thing that I ultimately have.
• 16:12
  SPEAKER_01: So my biggest credibility is in the near term, but I I do have an interest in extrapolating.
• 16:16
  SPEAKER_01: So let's let's hang out where I hang out most of the time.
• 16:19
  SPEAKER_01: So um generative AI right now, today, if you were to say, where is it landing in the enterprise?
• 16:25
  SPEAKER_01: Um, the answer is currently uh nowhere really unified.
• 16:32
  SPEAKER_01: It's super scattered, it's so nascent, it's almost ridiculous.
• 16:35
  SPEAKER_01: So I don't know if it was five months ago, four months ago, we interviewed the the um uh CTO of Microsoft Azure, for example.
• 16:42
  SPEAKER_01: So um uh and and uh Risinovich.
• 16:46
  SPEAKER_01: And he's got he's got some examples of companies like CarMax that are sort of generating descriptions of vehicles and things like that.
• 16:54
  SPEAKER_01: Um, you know, and then he had some hypotheticals, and then that's it.
• 17:00
  SPEAKER_01: Now, do I think we're gonna stay that nascent forever?
• 17:02
  SPEAKER_01: No.
• 17:03
  SPEAKER_01: But if you're an enterprise, right?
• 17:04
  SPEAKER_01: So six years ago, every enterprise felt like they were the only people not using AI.
• 17:08
  SPEAKER_01: Newsflash, that wasn't true.
• 17:10
  SPEAKER_01: Today, everybody feels like the old they're the only ones not using generative AI.
• 17:13
  SPEAKER_01: Newsflash, almost nobody is.
• 17:15
  SPEAKER_01: However, what's happened, what's happened with Chat GPT, which is a wild phenomena, by the way, um, is we're starting to see these things bubble up, particularly in text.
• 17:25
  SPEAKER_01: I think a Dale equivalent, so I think the visual will have a similar pop-off in in the coming, um, you know, maybe, maybe it'll be 18 months, um, is we're starting to see some people experiment with social, experiment with proposals, experiment with certain kinds of content writing support.
• 17:41
  SPEAKER_01: What's not happening is anybody takes very rarely, at least from what I can tell, is anybody in an enterprise is looking at a whole process and saying, oh my God, I can push one button now.
• 17:51
  SPEAKER_01: That's not happening.
• 17:52
  SPEAKER_01: What is happening, which is the story of AI in general, is people are looking at junctures within workflows.
• 17:59
  SPEAKER_01: Junctures within workflows.
• 18:01
  SPEAKER_01: And they're saying, in this juncture, I might be able to push this button and make this portion of what I do slightly better, slightly faster.
• 18:10
  SPEAKER_01: ChatGPT is different in that now, kind of Sally, who writes our tweets, is able to futz around with it.
• 18:17
  SPEAKER_01: Back in the day, the only people tinkering with with junctures were a vendor who had to do all the hard work of selling, land somewhere, and then say, God, where can I make some money around here?
• 18:26
  SPEAKER_01: And by the way, most of them didn't.
• 18:28
  SPEAKER_01: So fun fact for you, most of them didn't.
• 18:30
  SPEAKER_01: So um, so they had to land and then roll the dice.
• 18:33
  SPEAKER_01: And that's that's what it looked like to find a juncture back in the day.
• 18:36
  SPEAKER_01: Generative is gonna find its pockets.
• 18:38
  SPEAKER_01: Exactly where those settle in, nobody knows.
• 18:42
  SPEAKER_01: But what I can promise is that there's gonna be few areas where an entire like underwriting for insurance in general, it's done, boys, it's over.
• 18:50
  SPEAKER_01: We'll never do it again.
• 18:51
  SPEAKER_01: That's not gonna happen.
• 18:52
  SPEAKER_01: What will happen is certain paperwork-oriented processes, certain search or document-oriented processes may be streamlinable and bake inable with this tech in ways that kind of refine them.
• 19:03
  SPEAKER_01: So we're still at a spot where the dust has yet to settle on the early cluster of those use cases in insurance, okay?
• 19:11
  SPEAKER_01: The early cluster of those use cases in banking, okay?
• 19:14
  SPEAKER_01: The early cluster of those use cases in life side.
• 19:17
  SPEAKER_01: The the the first level of dust is not even settled.
• 19:20
  SPEAKER_01: So that's where we are with generative today.
• 19:22
  SPEAKER_02: I think it's it's also gonna be sorry, Seth.
• 19:24
  SPEAKER_02: I think it's also gonna be put on steroids the notion of like human augmentation, right?
• 19:30
  SPEAKER_02: For instance, and make it super surgical.
• 19:33
  SPEAKER_02: Subtitles for ADA compliance have to be a hundred percent accurate.
• 19:36
  SPEAKER_02: And if they're not, right, you know, so that's also obviously taking speech rec and putting it into subtitles or you know, providing other sound models so you get some of those other pieces.
• 19:46
  SPEAKER_02: So now we're merging models together where we're getting birds chirping, dogs barking, you know, taking out background sound, all that kind of stuff.
• 19:54
  SPEAKER_02: Where human augmentation still plays an absolute critical role in making sure that that's you know, fits compliance and it can go out, right?
• 20:03
  SPEAKER_02: I mean, I see I still think there's scenarios where it's going to be need to have super surgical human augmentation pieces there, but to your point for those specific junctures to actually make them to make them able and then learn from them and then go out.
• 20:19
  SPEAKER_03: So Chris, those are great points, and I and I completely agree.
• 20:23
  SPEAKER_03: You know, this is a tool in the toolkit.
• 20:25
  SPEAKER_03: I'm doing a workshop with uh with a uh charter school um you know, service underrepresented populations.
• 20:32
  SPEAKER_03: I've been kind of working with it for a number of years, and I'm helping them show, I'm showing them how they can use it day to day and augmenting what they're doing.
• 20:40
  SPEAKER_03: So it's a tool to improve people's uh uh efficiency and effectiveness, and at the same time helping them understand that it's not always correct, right?
• 20:49
  SPEAKER_03: And that it's not going to necessarily be the answer, but it can get you started.
• 20:53
  SPEAKER_03: And the critical thing is asking the right questions, even teaching students to ask the right questions, right?
• 20:59
  SPEAKER_03: And you know, how do you frame up your question to begin your research?
• 21:02
  SPEAKER_03: And it's very powerful in that way.
• 21:04
  SPEAKER_03: Now, what we're also starting to see, and this is my big area of concern, is you know, just like we used to hear people talk about knowledge and search, and oh, we want to make it like Google, we want a Google experience, right?
• 21:18
  SPEAKER_03: And you'd have to say, well, if you put as much time and energy and resources into optimizing your content as people do who want to get good rankings in Google, your search experience will be like Google, right?
• 21:31
  SPEAKER_03: So, you know, so so it's kind of like it's not just adding on a magic bullet here, it's not just you know, plugging this in.
• 21:38
  SPEAKER_03: You know, when Google search appliance famously failed because of that very issue.
• 21:43
  SPEAKER_03: Oh, yeah.
• 21:44
  SPEAKER_03: Right now, the thing about now, what we're starting to see is people say, I want Chat GTT for my uh internal experience.
• 21:54
  SPEAKER_03: And it's like, yeah, you know, we have to help people understand.
• 21:57
  SPEAKER_03: And I know there's going to be a lot of snake oil sales.
• 22:00
  SPEAKER_03: And a lot of vendors that are going to say, yeah, we can bring you Jack GPT for your enterprise.
• 22:07
  SPEAKER_03: And it's and it's not going to happen because it's not the same paradigm and it's not generative.
• 22:12
  SPEAKER_03: We're not using it in that way.
• 22:14
  SPEAKER_03: It's not going to replace knowledge management, knowledge architecture, knowledge engineering, ontologies.
• 22:19
  SPEAKER_03: It's not going to fix your data.
• 22:21
  SPEAKER_03: It's like we need a lot of industry education about what this can and cannot do because there's going to be a lot of wasted money.
• 22:28
  SPEAKER_03: There's going to be a lot of happened to us before is we'd have these great projects that would be teed up.
• 22:34
  SPEAKER_03: And all of a sudden there'd just be this giant whooshing sound for all the money getting sucked into the want the five million dollar Watson project, which ended up failing, you know, at one particular uh uh life sciences company three years later.
• 22:49
  SPEAKER_03: And so the point here is that executives need to understand that look, yes, this is exciting technology.
• 22:55
  SPEAKER_03: Yes, this is going to be very helpful, but do not buy the snake oil, you know, sales stuff of we're gonna do this for your enterprise.
• 23:03
  SPEAKER_03: I'm gonna plug in this technology.
• 23:05
  SPEAKER_03: It's a lot of complexity to the internal ecosystem of information.
• 23:10
  SPEAKER_03: And we still need to be able to have reference data.
• 23:13
  SPEAKER_03: We still need to be able to have, you know, core data architecture, core data quality.
• 23:19
  SPEAKER_03: This, these types of tools will not fix those problems.
• 23:22
  SPEAKER_03: And I think a lot of executives know that.
• 23:24
  SPEAKER_03: If you try to walk into an ex C-suite and say, I'm gonna use machine learning, fix all your product data and all your customer data and so on, they're like, get out of here, right?
• 23:34
  SPEAKER_03: But there's at this inflection point, we're gonna see more of that misunderstanding.
• 23:40
  SPEAKER_03: So why don't you speak a little bit to that?
• 23:42
  SPEAKER_03: What's realistic, what's not realistic, and how do we kind of bridge that gap from large language models to what will have to be more highly curated language models for a specific domain for an organization?
• 23:56
  SPEAKER_01: Yeah, I mean, you know, we could drill down in a hundred individual areas within business, but I'll just throw some arbitrary stuff out there.
• 24:05
  SPEAKER_01: So maybe we're talking about internal FAQs, right?
• 24:09
  SPEAKER_01: So people are talking about, well, I work at a corporation, and there's companies, even companies in Boston, Tamar, for example, uh, that were doing some of this.
• 24:18
  SPEAKER_01: I don't know if they still are.
• 24:19
  SPEAKER_01: I I was at their headquarters like half a decade ago.
• 24:21
  SPEAKER_01: I have no idea where they are today.
• 24:23
  SPEAKER_01: Um, but uh, you know, hey, I I want to know what our parking policy is at the Chicago office.
• 24:28
  SPEAKER_01: I want to know uh what happens under this circumstance with this HR policy.
• 24:33
  SPEAKER_01: I want to know um where we keep the logo files for these things and what the standards are for logo use on print material, whatever, right?
• 24:41
  SPEAKER_01: Um the in order to make that click, so I'm gonna try to take your general uh snake oil and and and kind of you know validate this with what I might imagine as an example.
• 24:56
  SPEAKER_01: So we might imagine that we would just plug in ChatGPT, tell it to read all of our FAQs, and then just answer everybody's questions.
• 25:05
  SPEAKER_01: Um, but of course, that would be ridiculous and have almost a zero percent chance of of succeeding right now.
• 25:12
  SPEAKER_01: It'd probably confuse more people that it would help.
• 25:14
  SPEAKER_01: What's going to have to happen is um there will have to be some kind of evolution on in some way the way that we store and structure different FAQ type stuff.
• 25:28
  SPEAKER_01: So that there will be a way that that stuff is found and that stuff is tagged and that stuff is organized.
• 25:35
  SPEAKER_01: There will be, you know, there will there will be some degree of knowledge architecture, some kind of maybe, maybe even human ontology.
• 25:42
  SPEAKER_01: Um, I'm not saying it needs to be ontologized all the way down to the final atom, but probably some skeleton for that stuff to sit on, some some way for it to be accessed, and also some kind of feedback loop for humans to say what's wrong, what's right, and be able to train that system.
• 25:56
  SPEAKER_01: And that's gonna be that's work.
• 25:58
  SPEAKER_01: We're talking data infrastructure here.
• 26:00
  SPEAKER_01: We're talking about ontologies, we're talking about feedback loops.
• 26:04
  SPEAKER_01: This is a project, it's not a push button.
• 26:06
  SPEAKER_01: So, yeah, that you know, I'm with you on that.
• 26:07
  SPEAKER_01: I'm with you on that.
• 26:08
  SPEAKER_03: Yeah.
• 26:09
  SPEAKER_03: So, and I think that you know, what's promising about the tools, and you know, when we look at large language models, a language model is really saying, here's how I understand this world, right?
• 26:23
  SPEAKER_03: This domain, this these common tasks, right?
• 26:26
  SPEAKER_03: Here's how I understand.
• 26:27
  SPEAKER_03: And again, because uh Chat GPT is so trained uh on so many different uh corpuses, corp I, corpus, I don't know, corpus, corp eye, uh, you know, there are lots of different data sources, and there are a lot of powerful algorithms in there that are really making predictions about text and predictions about text patterns.
• 26:51
  SPEAKER_03: Again, that is not going to necessarily solve the knowledge and the data problem of the organization, but you know, it will be able to uh uh I think that the tree, the trick and the key for organizations like mine or other consultancies is to bridge that gap, to say, okay, here's a generalized language model.
• 27:11
  SPEAKER_03: It's not going to have all of the scientific data points for life sciences, although you do start seeing uh language models for life sciences.
• 27:21
  SPEAKER_03: What I have found with those is that they are too big, they're too general.
• 27:25
  SPEAKER_03: They're even though they are specific for life sciences, they're gigantic and they're not usable.
• 27:30
  SPEAKER_03: And you can't have a thousand terms for disease as a drop-down, right?
• 27:35
  SPEAKER_03: They can't build that user experience.
• 27:37
  SPEAKER_03: So we still have to curate those language models.
• 27:41
  SPEAKER_03: We still have to curate those knowledge models because those have relationships between terms, those understand term frequencies and can make predictions around terms.
• 27:51
  SPEAKER_03: But again, they're not specific to the knowledge of the organization.
• 27:54
  SPEAKER_03: And the point you made about building that enterprise ontology is critical, right?
• 27:59
  SPEAKER_03: Because it's your knowledge scaffolding, right?
• 28:02
  SPEAKER_03: It's the stuff that is going to help you get a handle on your information and your knowledge and your landscape.
• 28:08
  SPEAKER_03: It's in if and if you do what everybody else is doing, you're not going to have competitive advantage, right?
• 28:13
  SPEAKER_03: The you know, the standards for efficiency, differentiation gives you competitive advantage, right?
• 28:20
  SPEAKER_03: So that's where I think the trick is.
• 28:22
  SPEAKER_03: How do we leverage these large language models in a way that bridges the gap between, okay, I don't want to do all the heavy lifting, but I want to do very focused uh work around our uh information in our enterprise.
• 28:38
  SPEAKER_04: Can you speak a little again?
• 28:40
  SPEAKER_03: Yeah, go ahead.
• 28:41
  SPEAKER_02: Just comment generally think about that.
• 28:43
  SPEAKER_02: Well, we what we've been pivoting on is is text.
• 28:46
  SPEAKER_02: We've got large corpuses of text that are out there.
• 28:49
  SPEAKER_02: You start generally stepping into synthetic media, and good lord, it's it's you know, it's jumped back like five years, you know, it feels like.
• 29:00
  SPEAKER_02: So for instance, we've had some you know, engineers trying to do some things like, hey, build this image, build this grouping of image, build uh, you know, I was watching a video on Twitch the other day, and they said, build an astronaut sitting on top of a horse on the moon, right?
• 29:14
  SPEAKER_02: Pretty straightforward.
• 29:15
  SPEAKER_02: As you can imagine, I mean, it's super, you know, arbitrary.
• 29:19
  SPEAKER_02: But the types of imagery that were being built, it was super grotesque.
• 29:24
  SPEAKER_02: So they did another test and they were like, okay, draw a picture of a woman.
• 29:29
  SPEAKER_02: Well, they're they're creating, you know, all these images that are, you know, that are out there on the web or the corpus of that.
• 29:35
  SPEAKER_02: And it came back as a very, you know, beautiful picture of a woman.
• 29:39
  SPEAKER_02: But guess what it was?
• 29:40
  SPEAKER_02: It was idealistic to the marketing pictures of women.
• 29:45
  SPEAKER_02: And it was very, we'll say, sexual in terms of, you know, being a scantily clad drawn woman.
• 29:51
  SPEAKER_02: Well, you know, it's because of all of the content and information that's out there on the web, this is what we're pulling from.
• 29:58
  SPEAKER_02: And that's all we have.
• 29:59
  SPEAKER_02: So the the thing that's that's awesome and that also sucks about both sides is you know, those the you know, the huge corpus of language models and data and all the stuff that we have in text is one thing.
• 30:10
  SPEAKER_02: Then we get over to media and it exponentially, you know, drops in terms of its of its abilities, and there's automatic natural bias in there that we've put in there for how many you know decades or or what, right?
• 30:27
  SPEAKER_02: So, you know, I I I think we, you know, part of this too is when we talked to you know, Henrik about synthetic media, that was only built off of deep fakes that were trained by somebody doing the work first and then you know it learning off of that single model.
• 30:41
  SPEAKER_02: So, you know, I think the the you know, we're way ahead, in my opinion, on text, but we're so so in the dark about media.
• 30:49
  SPEAKER_03: Yeah, and I guess the question comes down to what's the practical organization enterprise application of that.
• 30:55
  SPEAKER_03: You know, one of the things that you were, you know, so when you talk about that longer game of generative AI and synthetic media, you know, I mean, what should organizations be concerned with there, you know, to Chris's point?
• 31:10
  SPEAKER_03: Do you have some sort some guidelines or some thoughts around how to consider that?
• 31:15
  SPEAKER_01: There's a bunch of categories here.
• 31:17
  SPEAKER_01: Like nobody's dumb enough to ask Open AI something and then decide where they're gonna stick the scalpel, you know, and they'll have somebody on the operating table.
• 31:24
  SPEAKER_01: And very few people are gonna ask OpenAI something and then throw their clients' uh net worth on that one stock pick or something, right?
• 31:32
  SPEAKER_01: So there's some instances like that where I think quite naturally there's gonna have to be all kinds of decision-making processes where AI is informed and and you know, accountability can be baked in.
• 31:43
  SPEAKER_01: I'm not worried that we're all of a sudden gonna make stupid decisions automatically per se, per se.
• 31:48
  SPEAKER_01: I don't, I'm not saying I think humans are all brilliant, uh anything but, but I I, you know, there's some physical risk, sure.
• 31:55
  SPEAKER_01: Then there's also stuff that is kind of the other extreme of not as concerning just for me personally.
• 32:03
  SPEAKER_01: Um, like, you know, if you type in basketball player, you're probably not gonna uh find a guy that looks like me, for example, right?
• 32:11
  SPEAKER_01: What am I gonna do?
• 32:12
  SPEAKER_01: What who am I gonna sue over that?
• 32:14
  SPEAKER_01: I I kind of personally don't care.
• 32:15
  SPEAKER_01: I'm just gonna be frank with you.
• 32:17
  SPEAKER_01: I just personally don't care.
• 32:18
  SPEAKER_01: Um, however, if you work at Big Corp X, right, and you need to have a certain percentage of every hue of skin color, or you all get your heads chopped off in the whole department.
• 32:27
  SPEAKER_01: Well, then you're nervous about every single goddamn image and every single picture you ever put up about anything.
• 32:32
  SPEAKER_01: And so good on you.
• 32:33
  SPEAKER_01: I don't work in corporate.
• 32:34
  SPEAKER_01: So thank God I don't have to worry about that.
• 32:37
  SPEAKER_01: Um, but I think people who work in corporate, you know, those are the eggshells they walk on, and that's the difference between them, you know, never working again in corporate or uh, or or getting a pat on the back from their boss because they had the right representation of palette.
• 32:49
  SPEAKER_01: Is there stuff in the middle?
• 32:50
  SPEAKER_01: Sure.
• 32:51
  SPEAKER_01: My guess is where this stuff is going to add value is in a workflow and a junk shore where there is automatic bacon accountability.
• 33:00
  SPEAKER_01: In other words, you know, paperwork lands on X part of the workflow, XYZ generative process happens, XYZ fact-checking, human accountable, somebody's nads are on the line for this process happens, moves forward in the pipeline.
• 33:12
  SPEAKER_01: I don't think anybody's gonna let go of the reins, and all of a sudden, every insurance policy written in for auto for the next two years is gonna be totally screwed up because some evil AGI is taking it over.
• 33:23
  SPEAKER_01: So for me, I think that the dust settling will imply finding those accountability junctures.
• 33:29
  SPEAKER_01: And I'm not too worried about those being those of y'all in corporate, be nervous about all the stuff you got to.
• 33:35
  SPEAKER_01: Be nervous.
• 33:36
  SPEAKER_01: I'm not saying I don't have sympathy for you.
• 33:38
  SPEAKER_01: I just don't have those problems.
• 33:39
  SPEAKER_01: And then on the other side, I don't think anybody's gonna drive the scalpel into somebody's chest based on AI either.
• 33:44
  SPEAKER_03: So I just wanted to remind people uh we're we're uh over halfway through, and we are speaking with uh Dan Fajella, who is the CEO and head of research at Emerge, uh, which is a consulting firm in uh the area of artificial intelligence and emerging technologies.
• 34:03
  SPEAKER_03: Um, so so I know we kind of went down the rabbit hole of generative AI and ChatGPT because it's so popular and it is getting so much noise.
• 34:11
  SPEAKER_03: And I think we set some good parameters and guidelines around that.
• 34:14
  SPEAKER_03: Tell me what you're seeing in your client uh engagements and your current initiatives.
• 34:19
  SPEAKER_03: What's what is what is what is first, what is top of mind for executives around AI?
• 34:26
  SPEAKER_03: There's still a lot that needs to be done in in uh machine learning uh DevOps, there's still a lot that needs to be done in operationalizing stuff, there's still a lot that needs to be done in measuring the value of this stock around governance.
• 34:38
  SPEAKER_03: What are you seeing as top top of mind with well?
• 34:42
  SPEAKER_01: I'll give you a sense, there's so much.
• 34:44
  SPEAKER_01: And if you if you said e-commerce, if you said life sci, and if you said life sci commercial versus life sci drug dev, I'm just gonna drop so many different priorities on you.
• 34:53
  SPEAKER_01: But what what I will what I will say to answer your question broadly, and let me know if this will suffice, is I'll talk about what we've seen for like two, three year trends, like how is C-suite shifting around conceptions of AI, priorities of AI, kind of high level.
• 35:06
  SPEAKER_01: So a couple of things I'll bring up here.
• 35:08
  SPEAKER_01: One of them is um there was a time where buzzwords would get ink on paper for a pilot project.
• 35:15
  SPEAKER_01: Those times are over, mostly in North America.
• 35:19
  SPEAKER_01: Mostly in North America, those times are over.
• 35:21
  SPEAKER_01: And thank God, what a horrible waste of time and money that was.
• 35:24
  SPEAKER_01: Um, what we're seeing is that the mid-market sort of um, I think understands that there is no silver bullet.
• 35:32
  SPEAKER_01: So some people in the big companies who are willing to roll the dice or wanted to make a big career move um went with just buzzwords and and they all ate it.
• 35:40
  SPEAKER_01: And I think the mid-market has kind of learned from that.
• 35:42
  SPEAKER_01: There's been a general level up.
• 35:44
  SPEAKER_01: I'm not saying executive AI fluency is high, but I'm saying brute, gruesome ignorance is mostly gone in terms of expectations of silver bullets, even mid-market.
• 35:53
  SPEAKER_01: So there's still pockets of gruesome ignorance.
• 35:58
  SPEAKER_01: Sure, sure, sure, sure.
• 35:59
  SPEAKER_01: I'm just saying, I'm just saying if we if I look across, if I look across every bit of ink on paper across the vendors that we work with, just as an example, there's a lot less of like, I don't know why these guys sign this thing.
• 36:10
  SPEAKER_01: I I guess we have a pilot for this.
• 36:12
  SPEAKER_01: Who knows?
• 36:12
  SPEAKER_01: Right.
• 36:12
  SPEAKER_01: There's less of that.
• 36:13
  SPEAKER_01: There's less like, what is this?
• 36:16
  SPEAKER_01: There's less of that.
• 36:17
  SPEAKER_01: So um that's that's one thing.
• 36:20
  SPEAKER_01: A second thing that kind of pairs with this is that initially, four or five years ago, I think everybody wanted to be a platform.
• 36:28
  SPEAKER_01: Everybody still does kind of, but they all wanted to be a platform.
• 36:30
  SPEAKER_01: They all wanted to tie into all your most important data, be the central place that could process that.
• 36:35
  SPEAKER_01: And then you would depend on us for all your marketing decisions, and then eventually all your customer experience decisions, and then eventually all of this.
• 36:42
  SPEAKER_01: There's a couple players that have done well in the platform game.
• 36:44
  SPEAKER_01: I could talk for six or seven hours on what it looks like to do that right or not right.
• 36:48
  SPEAKER_01: So, like a data IQ did that pretty well with kind of lighthouse use cases.
• 36:53
  SPEAKER_01: Their chief customer officer was on the podcast, actually talked a bit about their philosophy.
• 36:58
  SPEAKER_01: We've also just seen them grow and we did some work with them back in the day.
• 37:01
  SPEAKER_01: So I've been a little bit under the hood enough to know that, you know, uh better than most.
• 37:07
  SPEAKER_01: Am I saying I, you know, put my, you know, uh life savings on their stock alone or they go public?
• 37:12
  SPEAKER_01: I'm not necessarily saying that, but I'm saying they've done it better than most.
• 37:15
  SPEAKER_01: And then folks like uh cognitive scale, who just had to sold sell for like scrap metal, um, you know, did not do that very well, for example.
• 37:21
  SPEAKER_01: And we could talk about all the differences.
• 37:23
  SPEAKER_01: Um, but everybody wanted to be a platform.
• 37:26
  SPEAKER_01: And what people learned is that mostly what they became, and a Yazdi sold for scrap as well.
• 37:32
  SPEAKER_01: Uh a Yazdi was kind of in the same boat, um, is that you're basically doing really high touch, high context, unique consulting projects for every person you work with.
• 37:40
  SPEAKER_01: And the the cost models, you raise venture money, there's just no way you're gonna make that work out.
• 37:44
  SPEAKER_01: Um uh Andreasen Horowitz did a good blog post on this uh a while ago as well, but people learned that.
• 37:51
  SPEAKER_01: So, last two, three years since COVID, we've seen a lot of pressure closer and closer and closer to what we call kind of like the AI SaaS business model.
• 38:00
  SPEAKER_01: So there's the full-blown AI services model.
• 38:03
  SPEAKER_01: So, Seth, this might be what you guys do.
• 38:04
  SPEAKER_01: You step in, you drink in very high context, you do totally unique projects and solve best built problems.
• 38:10
  SPEAKER_01: There's always going to be a market for that.
• 38:11
  SPEAKER_01: You're not gonna get automated anytime soon, and you're perfectly aware of that.
• 38:15
  SPEAKER_01: The opposite extreme, the opposite extreme is you plug in, we bill you like SaaS.
• 38:20
  SPEAKER_01: Hopefully, we don't even talk to you.
• 38:22
  SPEAKER_01: Hopefully, you just pay us in 12 months and we don't talk to you.
• 38:24
  SPEAKER_01: That's the other extreme.
• 38:25
  SPEAKER_01: Okay.
• 38:25
  SPEAKER_01: There's a there's only a few examples that are really all the way there in the enterprise.
• 38:30
  SPEAKER_01: Basically, it scrapes the web and looks at pricing and kind of uh uh availability of certain product skews across different areas of retail.
• 38:37
  SPEAKER_01: So if I'm Walmart, I can have a dashboard of how everybody else is pricing this pair of jeans.
• 38:42
  SPEAKER_01: And now instead of having to hook into Walmart's own data and then hook into Target's data and pull that stuff together, I'm scraping the goddamn internet.
• 38:49
  SPEAKER_01: And then I'm giving you a dashboard and that's all you're looking at.
• 38:52
  SPEAKER_01: I don't, I don't need to integrate with you.
• 38:54
  SPEAKER_01: I don't need to talk to 15 of your IT people, none of whom ever want to work with anybody, ever.
• 38:59
  SPEAKER_01: And so um, so I don't, so I don't have to do that if I'm edited.
• 39:03
  SPEAKER_01: There's other folks like AI Doc, for example, Seth.
• 39:05
  SPEAKER_01: Um, AI Doc has done pretty well in terms of adoption, certainly in terms of funds raised.
• 39:10
  SPEAKER_01: Um, they're Israeli based, but they figured out all the core machinery that does, I forget what kinds of scans they do.
• 39:18
  SPEAKER_01: Let's just call it breast cancer.
• 39:19
  SPEAKER_01: Okay.
• 39:19
  SPEAKER_01: They figure out all the machines that do the breast cancer scanning.
• 39:22
  SPEAKER_01: They figure out all the machines where the doctors look with their eyeballs to analyze for breast cancer on the images, and they just say, let us do, let us pipe out of all seven of the major companies that take the pictures, let's pipe them up, let's process them, let's learn how to pipe them down into all seven or or you know, eight of the companies like Phillips, whoever else, G E that of that make these visualization technologies.
• 39:44
  SPEAKER_01: And let's literally tell people your workflow is not gonna change.
• 39:48
  SPEAKER_01: Work with us, we hook you up, you're gonna look in your same screen and you're gonna have a red circle around the spots that have cancer.
• 39:54
  SPEAKER_01: That's the only thing that's gonna change for you.
• 39:55
  SPEAKER_01: We've seen a lot of companies be forced to do that in order to get a lick of adoption because deep, robust data infrastructure change, workflow change, that's 95% of what's hard about this.
• 40:06
  SPEAKER_01: The people running companies who came out of Stanford with six PhDs, they want to tell you it's the six PhDs that's the difference.
• 40:12
  SPEAKER_01: I will say there's a tremendous credibility to intelligence and technical chops.
• 40:16
  SPEAKER_01: Much more of the friction is in the high context workflow change stuff.
• 40:20
  SPEAKER_01: Huge pressure moving in the direction of AI SaaS.
• 40:23
  SPEAKER_01: Huge pressure over the last three to four years.
• 40:26
  SPEAKER_01: So that's another general trend.
• 40:28
  SPEAKER_01: And I think enterprises are more receptive to that because they're not ready to put a jackhammer to their data infra for the most part.
• 40:33
  SPEAKER_01: Right.
• 40:34
  SPEAKER_03: Right.
• 40:35
  SPEAKER_03: And then that's going to certainly give them some value, but eventually they do need to do the jackhammer.
• 40:41
  SPEAKER_03: I mean, they do jack fix, absolutely.
• 40:43
  SPEAKER_03: That's a temporary fix, right?
• 40:45
  SPEAKER_03: And that's all increasing fixed.
• 40:47
  SPEAKER_01: Yep, 100%.
• 40:48
  SPEAKER_01: And I I've said it a thousand times on LinkedIn, I'll say it another thousand times.
• 40:52
  SPEAKER_01: Um, this is a phase, and this is a phase where if it's if it's a bad phase, we're gonna start jangling ourselves with a thousand new plugins and technical debt in a thousand directions.
• 41:04
  SPEAKER_01: That's if it's a bad phase.
• 41:05
  SPEAKER_01: If it's a good phase, and this is the companies that'll survive, it will be a familiarity with how AI works.
• 41:11
  SPEAKER_01: It will be a leveling up of AI fluency, and it will be done on a trajectory of building capabilities towards overhauling the way we fulfill operations and handle our customers.
• 41:22
  SPEAKER_01: And so smart companies will take all of that as a learning about how to inform core strat and and kind of core tech.
• 41:28
  SPEAKER_01: The companies that are kind of silly are gonna just throw more toppings on the pizza um in until the pizza just busts in the middle.
• 41:35
  SPEAKER_02: Um so would you would you say then, because you know, definitely taking like 10, 20 steps back, like you said, um, they're gonna reach a limit to where their data can and cannot provide abilities, right?
• 41:49
  SPEAKER_02: That's just where it's gonna go.
• 41:50
  SPEAKER_02: Yeah.
• 41:50
  SPEAKER_02: Mostly in my mind, and this is where I'd love to get your thoughts on it, where you know, we see a ton of data producers, and those data producers produce so much information from all different aspects of the organization, and yet they they don't allow for subscribers to get all of that data to actually do stuff with it.
• 42:12
  SPEAKER_02: Right.
• 42:12
  SPEAKER_02: So it seems to me like the the jackhammering, you know, in terms of looking at that, can be a little bit more um, let's say prescriptive around, okay, let's look at this from the perspective of how are we doing generating data?
• 42:26
  SPEAKER_02: What are we doing for data governance?
• 42:28
  SPEAKER_02: And what does that look like for data discovery and consume consumption?
• 42:31
  SPEAKER_02: And can we put some guardrails around it for good for hell's sake's, right?
• 42:35
  SPEAKER_01: Yeah.
• 42:36
  SPEAKER_01: I wish that conversation was happening in every enterprise.
• 42:39
  SPEAKER_01: Of course, companies are very big, um, and it is tough.
• 42:42
  SPEAKER_01: But I I think the companies that are smart, there will be a good number of kind of catalysts who are pulling for, okay, guys, I'm glad we got some popcorn projects running.
• 42:53
  SPEAKER_01: Um, let's really think about uh what needs to be collected, what needs to be harmonized, what needs to be accessible, how, and let's let's think of sort of what the new normal would be for overall capability and strategy support moving forward.
• 43:07
  SPEAKER_01: Unfortunately, a lot of that is still going to happen in pockets.
• 43:10
  SPEAKER_01: So we had a woman on from Stryker, a big medical device company, uh, who works on kind of metadata architecture at the company.
• 43:19
  SPEAKER_01: Really good episode, by the way.
• 43:20
  SPEAKER_01: Fantastic.
• 43:21
  SPEAKER_01: She talked about some of these dynamics, and this is what we've seen as well.
• 43:24
  SPEAKER_01: We might have ideas about this big picture, but either the CEO or the department's heads, it's only a certain number of people that are willing to do it.
• 43:30
  SPEAKER_01: So it might be great to say, oh, but if we harmonize this in accounting, it's gonna be stupid if we don't harmonize it in marketing.
• 43:36
  SPEAKER_01: Well, you know what?
• 43:37
  SPEAKER_01: You're not gonna get buy-in.
• 43:38
  unknown: Yeah.
• 43:39
  SPEAKER_01: So you're gonna have to Frankenstein your way up.
• 43:41
  SPEAKER_01: I'm sorry.
• 43:41
  SPEAKER_01: This is legacy enterprise.
• 43:42
  SPEAKER_01: What do you want to do?
• 43:43
  SPEAKER_01: You want to reinvent the business?
• 43:45
  SPEAKER_01: So it's still gonna be chunky.
• 43:47
  SPEAKER_01: It's gonna be real chunky.
• 43:48
  SPEAKER_01: But if there's somebody over there pulling the strings and hoping that eventually these pieces are gonna fit together, that's much better than if nobody's there.
• 43:55
  SPEAKER_01: So you're right, the conversation should happen.
• 43:57
  SPEAKER_01: We should expect a lot of Frankenstein to get there.
• 43:59
  SPEAKER_01: But so long.
• 44:00
  SPEAKER_01: we move in the right direction it could be a good Frankenstein.
• 44:02
  SPEAKER_02: Yeah.
• 44:02
  SPEAKER_02: I think too a lot of people think when you start talking about data, you know, data um data producers and and just the notion of what governance even looks like.
• 44:11
  SPEAKER_02: It's everything from system governance to definitely people and human governance and who gets access to what because in in in your you know point Dan, most of the time those um that chunkiness comes from oh hell no, you're not touching my data.
• 44:26
  SPEAKER_02: I'm not given any access to this.
• 44:28
  SPEAKER_02: No way.
• 44:29
  SPEAKER_02: Hold his arm in the corner saying, no way in hell are you coming anywhere near this.
• 44:33
  SPEAKER_02: Especially if it's financer sensitive data like that.
• 44:35
  SPEAKER_02: And it's like, good lord, we did this it's you know it's like the most secure thing you know provides zero access.
• 44:42
  SPEAKER_02: Right.
• 44:43
  SPEAKER_02: So we're gonna have to figure out what that looks like in terms of we grant access to every one of our of our employees through you know our InfoSec ops team and they provide you know multifactor authentication all these all the standard terms for for accessibility within the organization.
• 45:00
  SPEAKER_02: We're gonna have to do the same approaches with data and who can get access to what so it's audible and transparent and accountable and things like that.
• 45:09
  SPEAKER_02: But at the same time, you know, things change all the time in Morph.
• 45:13
  SPEAKER_02: So that governance like you said has to have some type of catalog on there of who, what, and you know when is happening to all this.
• 45:21
  SPEAKER_02: And I think there's organizations that like you said that are still um yeah it will be chunky, but we're trying like crazy to make sure that you know those pieces are in place.
• 45:29
  SPEAKER_02: Because otherwise good luck you're not going to get a damn thing done or a project going that actually makes any sense until you actually have the most holistic perspective of the data and the guardrails around it.
• 45:39
  SPEAKER_02: So you know talk about your junctures.
• 45:42
  SPEAKER_02: Those are you know human junctures that have to happen within the data layer.
• 45:47
  SPEAKER_01: Great episode.
• 45:47
  SPEAKER_01: I think it's going live in like uh 20 days or something with Palantir.
• 45:52
  SPEAKER_01: So uh kind of VP of federal at Palantir talking about how uh the the tie of the like what it well she was very more frank than I would expect a vendor to be about navigating the emotional element to data silos.
• 46:09
  SPEAKER_01: The data silos are kind of built on emotion as kind of like the first component there.
• 46:14
  SPEAKER_01: And and I I people dance around that very few people will say it because it almost sounds insulting uh to to the people that are managing these silos.
• 46:21
  SPEAKER_01: But you're right.
• 46:22
  SPEAKER_01: I mean there's so many of these reactive responses it's it it's going to be a tough climb.
• 46:27
  SPEAKER_01: But I think if there's leadership that has a consistent vision over years, we can at least move in a good direction.
• 46:33
  SPEAKER_03: So um we've been we've been asked to actually I mean we'll do like day long like you know Seth will and stuff like immersion days and we've actually you know put training in place to do a mindset immersion day to talk exactly what you're talking about about the psychological psychological behaviors of barriers to entry that it's your humans good lord if we just had some generative you know AI that could just do this and stuff we would be I'm totally kidding so so Dan when you think about you know uh what a good what executives should be really concerned about and you know you talked a bit about you know executive fluency organization struggling you know executives want to feel smart and maybe not necessarily share the gaps in their knowledge and that's a little bit of a kind of a psychological you know uh maneuvering that has to be done.
• 47:31
  SPEAKER_01: What are your thoughts about kind of the the the where where there where the fluency still needs to be built and how enterprises need to think about that because you know there's executives trying to build that fluency but there's a lot of um fear that people don't have the answers or should have the answers or feel like they need to have the answers but they don't and then and then that whole trust and vulnerability maybe doesn't exist in in many of those uh senior uh uh organization you know senior levels of the organization right because again there's kind of it is a lot of success theater that goes on you know people kind of pretending that they're on top of this yeah we got it um what are your thoughts about that whole dynamic and and that that level of fluency and knowledge in how can you how can that be moved effectively what are the barriers to that how can that be moved effectively well look I'm gonna speak from a really hardcore realist perspective here super hardcore there's a friendly as hell answer but I I just I'm just I'm not even like capable of I don't know I'm not I can't even lie about it.
• 48:43
  SPEAKER_01: So look here's as far as I can tell there's nothing other than what I'm gonna say I don't lie because I don't have a good enough memory to remember the laws so you know yeah same I I'm I'm with you.
• 48:55
  SPEAKER_01: You got to juggle those in your head it's so hard.
• 48:57
  SPEAKER_01: So I I think and I think this is important for both vendors and and buyers to understand generally the way we've seen this happen in terms of how does fluency live so we have a model for this executive AI fluency if people Google that term they'll end up bumping into an infographic from us.
• 49:13
  SPEAKER_01: Sort of comprising for the most part an understanding of what AI does conceptually an understanding of the representative use cases that at least we'd be we should be pseudo familiar with at a conceptual level and then a pinch of how that feeds into overall strategy.
• 49:27
  SPEAKER_01: If you got a little bit of those things you're lifting your fluency a bit um how is that going to lift?
• 49:31
  SPEAKER_01: Well I mean I've said I've also said a thousand times that the lack of fluency is really the core driver of dollars into dead ends because people don't know what this stuff can do.
• 49:41
  SPEAKER_01: They don't know how it's important strategy that it's it's being tacked on in ways that are useless.
• 49:46
  SPEAKER_01: So this is a huge issue Seth it's absolutely monstrous and you see this up close and personal in ways that I don't even see because you're working right up with these folks at these big companies.
• 49:55
  SPEAKER_01: So I just have to interview them you actually have to go in and uh fix their stuff.
• 49:59
  SPEAKER_01: So the the way that that fluency lifts in enterprises the high level here is the way we've seen fluency lift within companies there's kind of two inroads.
• 50:08
  SPEAKER_01: Normally there's a a subset of people within the company there's going to be some people in pure success theater mode there's gonna be a subset of people who are genuinely self-educating when events were happening every two days for AI, you know, four or five years ago they were going to them and not just to get drunk.
• 50:24
  SPEAKER_01: They were going to actually learn things and really and they were actually thinking about what they could do there.
• 50:29
  SPEAKER_01: They're listening to podcasts uh diving in on newsletters following what their competitors are up to and in so much as they can um and and are aiming to consciously build their own fluency.
• 50:39
  SPEAKER_01: They lack some of the absolutely absurd misnomers that that maybe other folks might hold in their mind.
• 50:45
  SPEAKER_01: So there's some people who are don't know it but they're sort of building themselves to be potential AI catalysts.
• 50:50
  SPEAKER_03: Those people one eyed people in the kingdom of the block.
• 50:54
  SPEAKER_01: Yes, exactly so so where some cyclops is uh are are awakening right and this has been happening for half a decade.
• 51:00
  SPEAKER_01: So there's some of them there's some of them and they're learning a little bit where the rubber meets the road is the following is when people spend money and they work with a vendor of some kind.
• 51:10
  SPEAKER_01: I wish I could tell you any learning happened without this.
• 51:13
  SPEAKER_01: And then this isn't because I'm trying to side with the vendors here.
• 51:15
  SPEAKER_01: I'm on nobody's side of a market research firm but uh learning happens when people spend dough and then two things happen when spending happens two critical things one if you're working with a company that um kind of gets it uh a company uh if you're working with people on on the platform side or on the services side that really get it they're going to have to do the skill nobody talks about but I talk about constantly which is um making you feel like this is a co-creation process and it is but in reality educating you making sure you I get rid of all the dumb ideas in your mind but this feels like we're both putting ideas on the table.
• 51:57
  SPEAKER_01: Right, right.
• 51:58
  SPEAKER_01: It feels it feels like co-creation and it is okay I'm not saying it's deceptive but it's me avoiding your ego and making sure I get rid of your the stupidest ideas in your mind which aren't your fault I'm not blaming you but they're super dumb.
• 52:10
  SPEAKER_01: And also informing you around the strategic path but it's got to feel like co-creation.
• 52:16
  SPEAKER_01: It's got to feel that way.
• 52:17
  SPEAKER_01: And so a smart vendor does that and by making the sale people are actually smarter because they can't actually get the deal closed unless they make them smarter.
• 52:25
  SPEAKER_01: So the rubber meets the road when a sale happens now here's a second thing that happens when a sale happens I know we're short on time.
• 52:32
  SPEAKER_01: After that project happens whether it fails or it doesn't fail a lot of them are going to fail.
• 52:36
  SPEAKER_01: Whether it fails or doesn't fail if if we've done our job well as an if our those internal champions I talked about have done their job well, they've kept leadership and the vendors in constant cahoots and if nothing else when the project ends we know more about our our data info and how it's set up.
• 52:55
  SPEAKER_01: We know more about the signals in the noise of our marketing info that's actually valuable.
• 52:59
  SPEAKER_01: We know more about how to pull in subject matter experts not to screw with our process but actually help our process.
• 53:05
  SPEAKER_01: And so we've lifted our capability just a bit just a bit and so we've got to spend cash and if if we spend cash we can actually get one kind of ROI almost guaranteed and that's learning ROI.
• 53:17
  SPEAKER_01: Um how does how does fluency lift Seth I wish the internal experts could build themselves up so high that the problems are solved you get a little bit of that and then you spend money and terrible things happen.
• 53:27
  SPEAKER_01: And if you're smart you learn and that's the only thing I've seen work in legacy enterprise.
• 53:34
  SPEAKER_03: Well now you're getting engagements and you're getting people's attention and they're they're they have to pay attention they have to try to wrestle with those problems their the name their name is on the project they have to have and then and again uh we like to say you you win or you learn right you win that is it that's it yes exactly great lessons learned really what needs to happen.
• 53:58
  SPEAKER_01: You know this has been fantastic Dan I really enjoyed it um we'll always love catching up with you and chatting no matter what and uh looking forward to the next we're gonna have a fire soon uh have one of our bonfires so uh next couple weeks so hopefully you can make it over for that up to Carlisle and uh we yeah give me give me give me an email I'm gonna pop something over yeah absolutely come over it would be wonderful we still need to plan we need to still need plan out you know doing those uh Tomcat boat racing in the ocean though oh yeah the the Thundercat yeah Thundercat yeah uh just do a search on Thundercat Tom uh Davenport mentioned it in the in the intro to my book he said uh crazy fast powerful racing catamaran in which I I risk my safety and uh that was a lot of fun so we'll have to get out there as well but um this has been great thank you so much for your time um it's been a pleasure um what else do you want to share with folks before we drop you have your podcast you have some documents we're gonna put these in the show notes but what do you want to remind people yeah I mean and anybody with an interest in sort of enterprise AI applications our show is just the AI and business podcast so um relatively substantial show you know we talked about generative AI you know the people who coded Dolly to uh the chat GPT team like these are the folks that we kind of have on and other folks in big enterprise um yeah anybody interested in that audio media check it out otherwise emerge if they're interested in market research just emerj.com uh and feel free to find me on social and tell me you heard me on Seth's show and it'll put a smile on my face.
• 55:33
  SPEAKER_01: So that's about it for me.
• 55:34
  SPEAKER_01: I don't have anything else.
• 55:36
  SPEAKER_01: Awesome.
• 55:36
  SPEAKER_01: That's wonderful.
• 55:37
  SPEAKER_03: Thank you Dan thank you Chris Dan it's been a pleasure we'll talk very soon.
• 55:42
  SPEAKER_00: Yep have a good weekend guys see ya take care thanks for listening to the Early AI podcast with your host Seth Early at Chris Featherstone.
• 55:55
  SPEAKER_00: We hope you enjoyed the program and took away some ideas that you can start to implement in your organization right now.
• 56:01
  SPEAKER_00: If you're stuck and need help check out www.early.com for information about early information science consulting services and of course check the show notes for links to any resources we mentioned today.
• 56:14
  SPEAKER_00: That's all for now see you next time