Afraid AI Will Replace You? Here's the One Skill It Can't

WEBVTT
00:00:00.160 --> 00:00:02.319
I think there is something unique to AI.
00:00:02.479 --> 00:00:09.599
I think there's something unique to the ecosystem of AI in terms of like development and its like rapid increase in only a couple years.
00:00:09.759 --> 00:00:19.039
I think we as a community of professionals just working in the world, like haven't maybe had time to adjust to the rapid uh growth and investment uh in AI.
00:00:19.199 --> 00:00:19.920
It's everywhere now.
00:00:20.079 --> 00:00:23.199
But the fact that it's everywhere, I think speaks to sort of this universality.
00:00:23.359 --> 00:00:26.399
It's sort of a major technological innovation that people find ourselves.
00:00:26.719 --> 00:00:29.839
If AI has ever made you stop and think, wait, what is happening?
00:00:30.000 --> 00:00:31.120
You're not alone.
00:00:31.359 --> 00:00:34.719
I'm Mo, and I'm a security researcher asking the same questions.
00:00:34.960 --> 00:00:48.960
On Curiouser and Curiouser, we're having open conversations with experts, researchers, and leaders working at the edge of this space, talking through how AI is taking shape, what's shifting, and how people inside the work are thinking about it as it happens.
00:00:49.200 --> 00:00:52.479
So join us and listen in as the conversation takes shape.
00:01:06.400 --> 00:01:06.879
Hello, everyone.
00:01:06.959 --> 00:01:08.560
Uh I'm James Villubia.
00:01:08.799 --> 00:01:15.040
Uh I'm a former uh presidential innovation fellow for AI at NASA, uh, hence the rockets.
00:01:15.200 --> 00:01:17.439
Uh sadly, uh Moe is wrong.
00:01:17.519 --> 00:01:18.640
I did not build rockets.
00:01:18.799 --> 00:01:37.760
Um I built AI for the team that did planes and drones and all the things in the atmosphere that people more interact with, um, which I think is actually cooler because uh there's you know more things you touch uh that we worked on uh that humans actually experience because only a few people ever get to go to the moon, uh to the moon or fly in spaceships.
00:01:37.840 --> 00:01:42.239
Uh so I, you know, I I like the non-s uh the non-spaceship side of NASA.
00:01:42.400 --> 00:01:45.359
But I'm excited to be here, uh excited to talk about all sorts of things.
00:01:45.599 --> 00:01:46.000
That's cool.
00:01:46.079 --> 00:01:48.560
I didn't even know there was a non-spaceship side, to be honest.
00:01:48.719 --> 00:01:52.560
Like I just know that there's like freeze-dried ice cream, and that's pretty cool.
00:01:52.879 --> 00:01:59.680
So the group I worked with was sort of um like a like a weird research group within the aerospace side of things.
00:01:59.920 --> 00:02:02.400
So, like, you think about like what's the future of drones?
00:02:02.560 --> 00:02:12.000
What's the future of like un you know, unmanned aerial vehicles, and like can, you know, will New York ever have like air taxis uh like that fly me from building to building?
00:02:12.159 --> 00:02:14.240
Um, those sorts of like weird, crazy ideas.
00:02:14.400 --> 00:02:18.719
Well, someone's gotta be thinking about them and someone's gotta be thinking about like, what's the policy regulation there?
00:02:18.800 --> 00:02:20.319
What's the safety concerns?
00:02:20.479 --> 00:02:24.479
Uh, how do we deal with the wind tunnels of New York, you know, uh streets, et cetera?
00:02:24.639 --> 00:02:25.680
So lots of cool stuff.
00:02:25.919 --> 00:02:30.960
And then now you're actually like the CTO of Is It Your Own Thing that you're that you're doing now?
00:02:31.280 --> 00:02:35.120
It's a collection of uh private equity uh rolled up uh travel companies.
00:02:35.280 --> 00:03:01.280
So we do a lot of data uh analysis trying to basically uh provide the sales and data services for all those small businesses around the world that, you know, like, hey, I'm I'm renting a kayak or I'm renting uh, you know, I'm doing a sunset cruise on my honeymoon, uh, all those things that, you know, Airbnb or uh get your guide or uh you know, all those sorts of companies they sell, they're not the ones actually providing the uh that data service, uh collecting credit cards, et cetera, on behalf of those customers.
00:03:01.360 --> 00:03:03.759
Uh they're usually buying that data from someone else.
00:03:03.840 --> 00:03:16.479
Uh so we are uh we are sort of one of those big companies uh that helps the small businesses actually, you know, play in the big ecosystem of the internet and also help you know people on their honeymoons or their travel, et cetera, have you know cool local experiences.
00:03:16.879 --> 00:03:21.759
So basically, yeah, I was gonna say something about Airbnb, but no, I think it's a lot better than Airbnb.
00:03:21.840 --> 00:03:22.639
Or it sounds cooler.
00:03:22.879 --> 00:03:24.080
It's a partner to Airbnb, right?
00:03:24.159 --> 00:03:34.080
Because they great, you're renting your house, you know, they're there, but like, but like, oh, I want to go, you know, do uh like a, you know, it could be some, you know, some guy who sells uh who's doing local tours, right?
00:03:34.159 --> 00:03:42.319
He has a tour business, like, hey, you know, I was in Lisbon, for example, uh, with my wife a while back, and we found a small company, uh, didn't know we were actually, we it was right before it took this job.
00:03:42.400 --> 00:03:48.159
We were using uh the uh one of the companies that was our customer of ours, and we were trying to book like a small local tour.
00:03:48.240 --> 00:03:57.919
It's like it's a guy, it's his like electric vehicle, he picks us up, he knows our names, he gives us like a personal tour of like two or three of the castles around Lisbon, uh, like in the various periods in history.
00:03:58.000 --> 00:04:00.000
I was like, this is this is awesome, this is perfect.
00:04:00.159 --> 00:04:01.840
I would have never found this.
00:04:01.919 --> 00:04:04.319
Uh, I would never booked it if it hadn't been so easy.
00:04:04.479 --> 00:04:06.719
Like, great, put in my credit card, book the time.
00:04:06.879 --> 00:04:08.960
You know, here's where he picks me up at my hotel.
00:04:09.039 --> 00:04:12.319
It was just all seamless uh to have like a really good experience.
00:04:12.400 --> 00:04:14.719
Um, and I didn't have to know anything about Lisbon beforehand.
00:04:14.800 --> 00:04:16.800
It was just great, we'll we'll take care of that for you.
00:04:16.959 --> 00:04:22.480
Um, and that's the stuff, that's the experience that we want to provide uh to the end user, the you know, the the tourist, et cetera.
00:04:22.560 --> 00:04:25.680
But also we want to make it easy for the businesses to like manage all that.
00:04:25.839 --> 00:04:35.920
So uh yeah, it's not, it's not uh we are trying to get more AI uh involved, and I think that's why they brought me on, because obviously the travel world, everything's getting you know consumed by AI.
00:04:36.079 --> 00:04:46.240
Uh but you think about sort of like AI helping you plan your trips, try to find those really weird niche experiences that are like that are really good fit for you or for your kids or your wife or your you know partner, whoever it may be.
00:04:46.399 --> 00:04:51.040
Like that's like navigating and finding those things and setting them all up is is a complex problem.
00:04:51.120 --> 00:04:52.879
And I think you know, AI is the feature of that.
00:04:52.959 --> 00:04:55.199
And hopefully we'll we'll bring this this company there.
00:04:55.360 --> 00:05:00.560
It's been fun to see how AI has kind of made its way into the consumer space in this way, right?
00:05:00.720 --> 00:05:04.000
So like it started off as like, oh, well, where's the best place I can go?
00:05:04.160 --> 00:05:10.720
And it was just doing like basic searching, and then we got agents, and it's like, okay, well, book this very specific trip for me, right?
00:05:10.959 --> 00:05:15.360
Where you go and now it like books the trip, but it might be one part of the trip, right?
00:05:15.680 --> 00:05:18.160
Now the next step is like more complex.
00:05:18.240 --> 00:05:19.600
You get a couple more steps in.
00:05:19.839 --> 00:05:23.439
How do you actually go and build like a travel experience?
00:05:23.600 --> 00:05:25.759
Oh, well, let's do all these couple of things.
00:05:25.839 --> 00:05:26.399
What do you like?
00:05:26.560 --> 00:05:28.160
Do you want to center it more around food?
00:05:28.240 --> 00:05:34.000
Or do you want to go visit more things, or you know, depending on what you want, and then go find flights for it.
00:05:34.079 --> 00:05:57.759
And then, you know, it's so cool how like everything's evolved and how we've seen kind of AI go from chatbot to um this planner, then go to actually executing on these things, and eventually travel almost a travel agent, um, for like a full stack travel agent from like, oh, you're going to all these places, let me find all the hotels for you.
00:05:57.920 --> 00:06:01.839
This is actually in a chain that is connected to your like Amex card.
00:06:02.000 --> 00:06:04.240
So why don't we use, you know, like it's crazy.
00:06:04.319 --> 00:06:05.360
Like the amount of complexity.
00:06:05.519 --> 00:06:11.120
I see you've got the Chase Blue Platinum card, and you get points if you stay here, and like we've negotiated and the point value is this.
00:06:11.199 --> 00:06:17.759
So, like, this is actually a savings, even though the dollar value is like it can do all that for you way faster than you could have otherwise.
00:06:17.920 --> 00:06:21.839
Um, yeah, no, it's it's the world is changing very fast in this regard.
00:06:22.000 --> 00:06:36.160
And I think part of the thing that got me excited because my background before, even before NASA was social impact startups in education, uh, healthcare, et cetera, and a prior stint uh at uh in the Obama White House uh and you know around around the you know Obama administration.
00:06:36.399 --> 00:06:38.079
So there's been this like you know public policy ben.
00:06:38.240 --> 00:06:41.439
So it's like, okay, this travel company is this really you know exciting for me.
00:06:41.680 --> 00:06:52.319
But I also saw that like AI is coming for all these businesses because you know, you used to be, oh, hey, I can Google, I can go, I can search all these things and find a tour or something that might be of interest to me.
00:06:52.480 --> 00:06:54.879
We had to put in the work, but it was still there.
00:06:55.120 --> 00:06:56.560
AI has made that work a lot easier.
00:06:56.639 --> 00:07:01.360
It's like, oh, just ask and it gives me a recommendation, but just one, maybe two.
00:07:01.600 --> 00:07:05.439
So like AI is sort of making a lot of choices on behalf of us.
00:07:05.680 --> 00:07:14.160
And I don't necessarily know if all those models are treating all of the sort of vendors, all the tour guys out there with the same, you know, degree of uh, you know, balance.
00:07:14.319 --> 00:07:23.759
Like, hey, like, are they pushing everyone to the same vendors just because that's the where the LLM sort of like that was what is what got remembered by the LLM when you know in the training process?
00:07:23.920 --> 00:07:31.120
So I think there's a there's a concern that a lot of small businesses will just get sort of left behind uh in the sort of coming AI replacing search.
00:07:31.199 --> 00:07:41.920
Uh and I think it's like, well, if we can help them tackle that and stay out of that and give people better experiences, give them more optionality, um, it's a it's a win-win for a lot of small communities that depend on tourism around the world.
00:07:42.240 --> 00:07:48.240
You built also um an AI, an educational AI for X Prize.
00:07:48.560 --> 00:07:55.120
I'm wondering how you feel about like AI in terms of reducing the or how people are using AI.
00:07:55.199 --> 00:08:00.079
So you you equivoc uh equivocated it to it like search, right?
00:08:00.560 --> 00:08:18.959
But I think we're seeing a lot of use cases where people are relying on AI for answers and like not just like going to and finding the answer from search, but then going, copying, pasting that same answer and just uh kind of downloading it as knowledge into their own like you know, memory.
00:08:19.279 --> 00:08:20.240
It is changing.
00:08:20.399 --> 00:08:22.879
I mean, I think there's some great psych papers out there right now.
00:08:23.040 --> 00:08:28.959
Like the use of AI is actually changing how brains are forming, how like neuropathways are forming, how we retain information.
00:08:29.199 --> 00:08:35.360
And I don't know if we've quite figured out, like we can see early on, we can see what's being diminished.
00:08:35.600 --> 00:08:37.759
I don't know if we quite know what is being grown.
00:08:37.840 --> 00:08:48.960
Like, what are the new neural pathways, uh, the types of analytical thinking that will like will have to be respond, will have to grow and respond uh in this sort of world of AI.
00:08:49.120 --> 00:08:52.000
But I think it's something we should we should be conscious of for sure.
00:08:52.240 --> 00:08:54.480
We are relying on AI so much.
00:08:54.720 --> 00:09:01.919
And like you said, um there was actually this really good podcast I was listening to a while ago, um, like the the cognitive something.
00:09:02.159 --> 00:09:02.720
It was cool.
00:09:02.879 --> 00:09:21.440
Um, but uh in one of the episodes, it was this educator who's or this researcher who was talking about um how there was a study and they gave people AI and they limited other another group to like not using AI, and they wanted to find out like who was learning better or whatnot.
00:09:21.840 --> 00:09:38.559
And it just showed that like people who were using AI in a way to like just get answers were actually performing worse than people who were like using AI in a sense to uh more give them um leading principles to find their own answers, right?
00:09:39.039 --> 00:09:46.960
So it's more like using AI as a guide versus using AI as the the you know the source of truth.
00:09:47.279 --> 00:09:54.080
There's a sort of a skill that develops in like I think all career paths, um, where eventually sort of you can sort of smell smell bull.
00:09:54.320 --> 00:09:59.039
Uh or you know, like you've been around like I don't know the answer, but that feels wrong.
00:09:59.279 --> 00:10:04.320
Um, and you put in the sort of the 10,000 hours mindset, you know, uh was it Malcolm Gladwell?
00:10:04.399 --> 00:10:22.559
Like you, whatever the framing is, the idea that like eventually you are exposed to enough things that like your brain actually does start to pick up on a lot of like subtle notions, subtleties that you maybe can't put to words, but you're like, this something's about this is off, and I want to like, you know, click in there a little bit more, see, answer, ask a few more questions.
00:10:22.960 --> 00:10:34.879
That is something that really comes with time, but that really is the skill that is best applied to AI, because AI is wrong a lot of times, mostly because it just doesn't know what you really want.
00:10:34.960 --> 00:10:41.120
It doesn't have the full context of the situation, and trying to give it all of that information is tedious and time consuming.
00:10:41.360 --> 00:10:51.519
So you sort of like summarize what you need, ask it, and it gets like, well, I think based off of this, here's you know, here's regress of the mean, here's sort of like the the optimal point.
00:10:51.759 --> 00:10:52.799
But it can't know everything.
00:10:52.879 --> 00:11:15.919
And if you can't push back, and that's where I think that sort of the 21st century skills in a lot of sort of the educational circles where they're focused, like we need to build AI to help kids have 21st century skills, which is mostly critical thinking about how to use AI, uh, how to push back on AI, how to have in you know, interrogatory, inquisitive question and answer sessions with AI saying, oh, it gave me this answer.
00:11:16.159 --> 00:11:17.519
Where do you think it might be wrong?
00:11:17.679 --> 00:11:18.559
Okay, well, ask it.
00:11:18.639 --> 00:11:22.080
Okay, well, like how do you know that the second answer was correct?
00:11:22.240 --> 00:11:35.200
Um, there was a great study that came out, I don't know if it was a great study, but it was a study that came out um a few weeks ago where it said, I think uh what is it, 50 to 60 percent of AIs will change their answer with just one pushback.
00:11:35.440 --> 00:11:38.399
So, like, hey, you know, plea, you know, like we recommend A.
00:11:38.480 --> 00:11:39.679
Or like, I don't know about A.
00:11:39.759 --> 00:11:41.519
It's like, oh, okay, you're right, we recommend B.
00:11:41.600 --> 00:11:53.440
And it's just like without hesitation, because they overalign on like trying to make people feel good, uh, on like their answers, like, oh yeah, you like my answer as opposed to the answer was correct.
00:11:53.679 --> 00:12:06.639
Um, and until we, I think, as a culture learn to think like how we converse with AI in a way that is sensitive to that like high variance, uh, high willingness to change answers, like we will struggle.
00:12:06.799 --> 00:12:07.360
People will struggle.
00:12:07.519 --> 00:12:11.440
But the people who are naturally maybe a little bit distrusting, I think will succeed.
00:12:11.519 --> 00:12:14.320
They are, they are sort of coming to the game with that sort of critical lens.
00:12:14.559 --> 00:12:18.799
And also, I think people later in their careers who've developed that critical lens can also do it.
00:12:18.879 --> 00:12:28.159
And I think that's why you're seeing a lot of more sort of the entry-level light-collar jobs are struggling, because it's it's hard to it's hard to hire people without that sort of growth yet.
00:12:28.399 --> 00:12:36.000
But you know, I think we as a culture will have to find a way to upskill those people into that critical thinking without having them, you know, waste 20 years doing the grunt work.
00:12:36.320 --> 00:12:42.240
I think one of the big parts about that too is ensuring that there's material that people can actually use and access.
00:12:42.399 --> 00:12:55.360
And um, again, like there, there's still like these tech deserts, right, in in the United States, even right, where people just don't have access to the technology, or they don't even have access to learn about the technology.
00:12:55.600 --> 00:13:06.000
And they kind of get it through this trickle-down effect where they hear rumors about it, or they see like the second or third lay um costs of the of the technology.
00:13:06.480 --> 00:13:12.480
So I am bringing this back to actually your project, which I want to I want you to also describe a little bit more.
00:13:12.639 --> 00:13:17.840
But it's been described as like the Wikipedia um like a new a new version of Wikipedia.
00:13:18.080 --> 00:13:20.240
At least that's how I kind of looked at it.
00:13:20.480 --> 00:13:27.039
Um again, it and I could be wrong about this, but it seems like it was just a self-improving textbook.
00:13:27.200 --> 00:13:34.320
Yeah, so this was a project uh called Mount Cleverist, uh, because puns were really, really cool at the time, and we we just loved that.
00:13:34.559 --> 00:13:37.679
But it was, I think, decidedly ahead of its time.
00:13:37.840 --> 00:13:48.639
Um, because we were hacking together without the billion dollar budget, we were hacking together NLP and AI models, trying to build um what could, you know, like you could knock out overnight right now with the current models.
00:13:48.720 --> 00:13:55.840
But the idea was um that schools, like there's so many there's technical tech, as you said, tech tech deserts, right?
00:13:55.919 --> 00:14:06.240
There's a lot of schools um whose you know, procurement process to get tech technology in the classroom is very slow, the tech that they're buying is bad, uh, or um it's they're disproven.
00:14:06.320 --> 00:14:09.120
It's like not, you know, it doesn't have like a lot of statistical validity.
00:14:09.440 --> 00:14:14.240
Most software used in schools does not have any strong efficacy analysis.
00:14:14.399 --> 00:14:20.000
Basically, like you can't prove that the students were smarter and better and retained more at the end of using your tool.
00:14:20.159 --> 00:14:21.759
Like almost none of them have that.
00:14:22.000 --> 00:14:25.759
It is very hard to do from a like science statistics perspective.
00:14:25.919 --> 00:14:30.080
The data is really hard to collect because you got student privacy rights and a lot of all these things.
00:14:30.240 --> 00:14:39.120
And the companies that can sort of both afford to operate at like educational tech margins and hire the data sci data, you know, data team to do that, there's a few and far between.
00:14:39.279 --> 00:14:43.919
And like uh, you know, the sort, you know, like Duolingo was able is able to get you up to a certain level.
00:14:44.000 --> 00:14:48.399
And like they, you know, have have, I think, you know, uh struggled to get to that next level of fluency.
00:14:48.480 --> 00:14:53.840
And a lot of companies say, hey, look, they, you know, even they, the top, this huge company, still is having trouble.
00:14:54.080 --> 00:14:59.279
So I think our goal with Mount Cleverist was to say, okay, well, how do I get something in the hands?
00:14:59.440 --> 00:15:01.039
There are there is all this information out there.
00:15:01.200 --> 00:15:13.120
It's Wikipedia, there's all these textbooks, like something more engaging than like the crappy textbook from the 1980s that, you know, like says our current president, Reagan, uh, that's being you know, hand-me-down uh in these sort of you know, rural school districts.
00:15:13.279 --> 00:15:20.799
How do we get something in them that is better, but not assume that our choice of what better is is always gonna be right.
00:15:21.039 --> 00:15:32.879
So the idea was well, we're gonna scrape internet content that the teacher selects, we'll build questions, and we're basically gonna assume that any content we create or pull and any questions we we you know pose or generate are just bad.
00:15:33.200 --> 00:15:35.200
That helped because our AI wasn't very good.
00:15:35.279 --> 00:15:37.600
So the questions were actually sometimes really bad.
00:15:37.759 --> 00:15:44.399
But the idea was that it would, as people, as students took it, and that maybe not every question they got exposed to was like a really good question.
00:15:44.480 --> 00:15:45.919
Sometimes it was just like a bad question.
00:15:46.000 --> 00:15:48.720
It was poorly formed, uh, hard to understand.
00:15:48.960 --> 00:15:54.559
But over time, we would be able to pull out like, oh, which questions actually led to long-term retention?
00:15:54.720 --> 00:15:59.440
And over time, you would start to see this like this slow growth of like, okay, well, what do those questions look like?
00:15:59.600 --> 00:16:00.960
How do we repeat those questions?
00:16:01.120 --> 00:16:06.320
Okay, so every time a student took a test, it improved the quiz for the next student and then the next student, the next student.
00:16:06.399 --> 00:16:08.080
And that it wasn't limited to one classroom.
00:16:08.159 --> 00:16:11.919
It's like anyone who ever, you know, our classic example is like War of 1812.
00:16:12.080 --> 00:16:21.360
Anyone who ever took a quiz on the War of 1812 would benefit from every teacher and every student who'd ever had been asked any question at that like grade level about the War of 1812.
00:16:21.519 --> 00:16:32.879
And like being able to sort of have this like this slow building roll-up of true efficacy-backed, uh statistically backed uh education and games was like was the dream.
00:16:33.120 --> 00:16:39.519
The problem was that the AI, like our AI, which is like we couldn't afford to get like to the top top-tier models.
00:16:39.600 --> 00:16:42.720
Um, and then COVID hit, uh, and like we had shut it down.
00:16:42.799 --> 00:16:44.240
I thought God, God bless.
00:16:44.399 --> 00:16:50.960
Um, I was working for two startups at the time, one in remote work uh analytics of how to make remote work profitable.
00:16:51.120 --> 00:16:56.399
Um, and then the other was in online education uh for like asynchronous online education at home.
00:16:56.639 --> 00:17:03.840
Both of them shut down in November 2019 because the investors are like, we don't think these things will all take off.
00:17:04.319 --> 00:17:07.839
Um that one, those, those sting.
00:17:08.000 --> 00:17:18.319
So yeah, so the the remote work one obviously, I think uh the the stuff that we wrote, the the stuff we published became like the default policies for I think like Deloitte and Facebook and all these big companies like just took it wholesale.
00:17:18.400 --> 00:17:20.160
It's like this is all policy now.
00:17:20.240 --> 00:17:21.519
Um, and they had some success.
00:17:21.599 --> 00:17:26.799
And the education, you know, investors call me up six months later, like, hey, is that still thing still thing still alive?
00:17:26.960 --> 00:17:27.920
We we use it.
00:17:28.000 --> 00:17:29.200
It's it's it's perfect.
00:17:29.359 --> 00:17:30.799
This is your this is your moment.
00:17:30.960 --> 00:17:32.400
It's like, well, so we shut it down.
00:17:32.480 --> 00:17:33.759
Like it's it's it's gone.
00:17:34.079 --> 00:17:36.960
Well, do you think that is the future of education and AI?
00:17:37.119 --> 00:17:56.960
Is that thinking that like this comes back to this conversation where you were talking about sort of AI and its growth, like there's a if you start with the assumption that what the AI is producing is distrusted, um, and it's like it's it's not great, but if we can sort of collectively start to, as a as a culture, as an organization, as a community, start to say, hey, but like this was good.
00:17:57.039 --> 00:17:58.400
This is what good looks like.
00:17:58.799 --> 00:18:00.720
Hey, everyone, let's learn what good looks like.
00:18:00.799 --> 00:18:02.319
Let's all share this learning.
00:18:02.480 --> 00:18:04.319
Um, suddenly we can sort of move forward.
00:18:04.400 --> 00:18:12.160
And that that applies to like all uses of AI, including this, you know, one uh in education that was just very sort of specific example.
00:18:12.240 --> 00:18:22.319
Uh, but is I think tied to this, like the next step that we need for all of these, you know, all of these industries that are trying to tap into AI is is saying, great, yeah, it's not great now.
00:18:22.400 --> 00:18:23.519
It will eventually get there.
00:18:23.599 --> 00:18:28.640
And the the fastest way we can sort of get over that hump is to share our learnings and to sort of cooperate.
00:18:28.799 --> 00:18:32.480
It doesn't need to be sort of like, you know, eat or be, uh, you know, kill or be killed out there.
00:18:32.880 --> 00:18:44.720
I think having something that's more focused around continuous learning, but is also able to iterate, super important to just keep people like the neuroplasticity, kind of keep the connections alive, right?
00:18:44.799 --> 00:18:52.079
Because we don't want to lose that ability to like have to actually critically think and then actually go and read the material.
00:18:52.240 --> 00:18:55.519
Um, I think there's a lot of interesting stuff with generating new questions.
00:18:55.839 --> 00:18:57.359
I think you hit the nail on the head.
00:18:57.440 --> 00:19:03.200
So I've been working with a uh a great uh nonprofit as part of the American Education Research and Development Fund.
00:19:03.279 --> 00:19:05.920
Uh it's a big, big, big fun sort of DARPA for education.
00:19:06.160 --> 00:19:10.880
It's a group called Augment Ed, and they are like focused on this research problem and they're they're you know kicking off some great work.
00:19:10.960 --> 00:19:20.079
But I think one of the conversations that we had uh sort of the early days of design was sort of thinking, well, like what is the modern experience with AI look like for homework for students?
00:19:20.240 --> 00:19:28.000
And it's like, well, if we've always said, hey, we can ask a question and the student will give us the answer, and we're basically measuring, did they memorize the answer?
00:19:28.160 --> 00:19:30.240
It's regurgitation to some degree.
00:19:30.480 --> 00:19:33.920
And having those sort of like deeper essay-based questions is harder.
00:19:34.000 --> 00:19:37.519
It's way harder in earlier grades where like writing isn't as strong.
00:19:37.680 --> 00:19:41.839
They don't need to have the skills yet to communicate those sorts of uh ideas.
00:19:42.079 --> 00:19:46.559
But it's always been a gap that we have this like just memorize all the facts, move forward.
00:19:46.720 --> 00:19:49.119
Um, that has been a problem in our education system.
00:19:49.279 --> 00:19:59.440
I think AI has like said, well, if that is, if cheating is effectively with AI so easy that everyone has sort of a copy of the test from the prior year, then the test is useless.
00:19:59.839 --> 00:20:02.960
Just stop pretending, stop trying to make the test more useful.
00:20:03.119 --> 00:20:04.880
It will never be useful again.
00:20:05.200 --> 00:20:07.200
But what then becomes useful?
00:20:07.279 --> 00:20:08.400
It's like, well, great.
00:20:08.720 --> 00:20:15.119
Have it, you have it, you know, and ask the question and say, okay, well, what questions do you have of the story for your to kill a mockingbird?
00:20:15.279 --> 00:20:15.759
It's like, great.
00:20:15.839 --> 00:20:21.119
Well, here's here's something like here's a statement that's like someone made about to kill a mocking bird.
00:20:21.279 --> 00:20:22.000
Do you agree?
00:20:22.240 --> 00:20:23.359
Why or why not?
00:20:23.519 --> 00:20:27.359
And if you can have an AI actually not give the answers, but just like probe.
00:20:27.519 --> 00:20:30.559
Uh it's like, well, I don't like what in the book makes you think that?
00:20:30.720 --> 00:20:32.160
Like, hey, like what was a part of the book?
00:20:32.240 --> 00:20:47.039
Like, you know, cite your source effectively, but like they don't have to like memorize the exact lines, but having that sort of interrogative conversation in a way that feels fluid and easy for you know a fifth grader or sixth grader or eighth grader, whatever it may be, to have, uh, not overly burdensome, not super technical.
00:20:47.200 --> 00:20:54.640
And the AI is not like, you know, it's not some strict rubric because you're like, I just want to make sure that like, did the student engage enough to ask some critical questions?
00:20:54.799 --> 00:20:57.200
They had a novel idea and maybe they were rewarded that.
00:20:57.279 --> 00:20:58.559
Oh, no one's asked that before.
00:20:58.720 --> 00:20:59.359
That's really cool.
00:20:59.440 --> 00:21:00.880
No one's posed that idea.
00:21:01.119 --> 00:21:05.599
Like that sort of reward structure is again only possible with AI.
00:21:05.920 --> 00:21:17.279
So I think, you know, the way out of the, oh, students are all cheating on tests with AI these days, is like, yeah, well, the only way out at scale is really to get AI more into the classroom.
00:21:17.359 --> 00:21:29.519
Um, and I think my concern for the next you know, 20 some odd years is that education will take, takes that long for like those sorts of innovations to really take seat in the in the ecosystem.
00:21:29.680 --> 00:21:41.200
Uh so the kids going, my kids, uh, you know, Gen Alpha, who will be going through school without, well, not, you know, with a lot, like call, as you point out, a lot of diversity in the accessibility of various tools and technologies.
00:21:41.440 --> 00:21:45.920
Like, there's we're gonna see a real chasm between sort of the haves and has nots in education.
00:21:46.079 --> 00:21:47.119
COVID pointed this out.
00:21:47.200 --> 00:21:52.400
I mean, my dad, like, even as of top top tier schools, I was, you know, talking to schools, like, well, how do we, how can we help?
00:21:52.480 --> 00:21:53.119
How can I help?
00:21:53.279 --> 00:21:59.200
Even if we deployed some of my my AI tooling, it's like, yeah, well, not every household has a laptop.
00:21:59.359 --> 00:21:59.599
Certainly.
00:22:00.160 --> 00:22:02.880
Not three or four laptops, one for each kid.
00:22:03.039 --> 00:22:03.200
No.
00:22:03.359 --> 00:22:04.799
So okay, now we have to take turns.
00:22:04.960 --> 00:22:06.240
Okay, well, you know, 9 a.m.
00:22:06.400 --> 00:22:07.039
to 11 a.m.
00:22:07.200 --> 00:22:08.880
is the sophomore's time to get on class.
00:22:08.960 --> 00:22:12.880
And okay, then you know, then near my fifth grader hops on because there's only one shared laptop.
00:22:13.039 --> 00:22:17.119
Well, they all have phones, but oh, the phones can't do this because none of the software was designed for phones.
00:22:17.200 --> 00:22:20.400
Because the for years, like, oh no, no, we don't want foo phones in schools.
00:22:20.559 --> 00:22:27.119
And then COVID comes along, and phones are the only at scale accessible device in the hands of every child to actually do that.
00:22:27.200 --> 00:22:28.960
But they weren't, none of it was designed for that.
00:22:29.119 --> 00:22:38.640
So these are sort of like all the weird things that like the gotches that are gonna come for this ecosystem that AI is, you know, gonna introduce more of.
00:22:38.720 --> 00:22:51.519
Um, and this is strangely, this is the stuff, these are the questions that that group at NASA, like this is what we asked, like this trying to play out these games and come up with solutions, usually to like drone and and and aerospace problems.
00:22:51.680 --> 00:23:02.799
But we were also looking at like, you know, at education, like what you know, what does AI's impact on education do to the crop of aerospace engineers that we coming up in 20 years?
00:23:03.119 --> 00:23:07.279
And will we have enough to support you know the space economy for the US?
00:23:07.519 --> 00:23:09.599
They're all all this is you know interrelated.
00:23:09.839 --> 00:23:15.359
At the beginning of our conversation, you also mentioned there's another neuropathway that we really need to get ready for.
00:23:15.519 --> 00:23:17.599
And we don't really know what it is.
00:23:18.000 --> 00:23:24.960
Um just this morning I was having a conversation with our CTO, and we're working on a couple of really cool projects.
00:23:25.200 --> 00:23:32.559
And one of them was like trying to figure out well, how do we like um we're trying to build like this agentic solution?
00:23:32.799 --> 00:23:37.599
How do you think about agents in how do you solve an agentic problem, right?
00:23:37.920 --> 00:23:43.920
Um, you have to solve it agentically, but the only way you do it is by understanding agents enough to be able to implement them.
00:23:44.079 --> 00:23:47.839
So, how do you learn about agents if you're not iterating and playing with agents?
00:23:48.000 --> 00:23:52.799
And I think this is like where we get to this new skill set that people need to have.
00:23:52.960 --> 00:24:02.240
And we've seen it pop up on job boards where it's now like AI coordinators and just AI engineers, which are basically prompt engineers, right, with extra steps.
00:24:02.480 --> 00:24:10.000
And we're seeing AI implementation engineers, which are engineers that know how to plug AI into workplace processes.
00:24:10.319 --> 00:24:15.839
So, you know, you've advised people to play with AI enough until you see it fail, right?
00:24:16.160 --> 00:24:20.000
And understand why it's making those mistakes, which is awesome.
00:24:20.160 --> 00:24:28.400
I think this is actually like a perfect example of like a new pathway that we need to start building, which I don't really know what it's called, right?
00:24:28.640 --> 00:24:33.279
Because this iterative motion isn't something that I think we've had to deal with, it's usually question-answer.
00:24:33.519 --> 00:24:37.440
But now it's like like, what if something is doing all the work for you?
00:24:37.759 --> 00:24:43.920
What is what does this new observability uh kind of pathway look like now that you need to become more observant?
00:24:44.160 --> 00:24:46.160
As humans, we've always been observant, right?
00:24:46.240 --> 00:24:49.920
It's uh it's biological, it's what we needed for survival at some point.
00:24:50.160 --> 00:25:00.480
But at some point it feels like um the technology was very much like okay, you do one thing, you give me a response, all right, we'll just keep going back and forth with this until something works.
00:25:00.640 --> 00:25:04.960
But now it's like sit back and watch the screen is what a lot of people are doing.
00:25:05.440 --> 00:25:09.200
So it doesn't mean that this is the worst thing that can happen.
00:25:09.359 --> 00:25:18.240
Um, it might mean that this is the next step in how we think about where engineering is today, where it needs to go.
00:25:18.640 --> 00:25:20.480
Like, how do we evolve this?
00:25:20.880 --> 00:25:33.039
Um, but I guess, in a sense, how do you kind of get people comfortable with playing enough with these tools and failing in an environment where they feel like they can experiment, right?
00:25:33.200 --> 00:25:37.839
And I can imagine that this is something that you had a challenge with at NASA for sure.
00:25:38.079 --> 00:25:44.720
I mean, I think one of the things that excited me about uh when when you all reached out about sort of coming on the show was like the title itself, like Curious and Curiouser.
00:25:44.799 --> 00:25:52.079
And I was like, great, like I I talk at length, uh like when I give keynotes, like it's almost always like the title is Curiosity in the age of AI.
00:25:52.160 --> 00:25:53.279
It is, it's the focus.
00:25:53.359 --> 00:26:01.440
And I think that has been sort of the linchpin deciding factor of like I think people's ability to adapt in the current ecosystem.
00:26:01.599 --> 00:26:08.240
People who are like, I saw AI, it's like, nope, I I it's it's too complicated or it's too scary or I don't trust it, and I'm just gonna put it to the side.
00:26:08.400 --> 00:26:11.920
And then people are like, oh, I don't trust it, and I don't know why.
00:26:12.240 --> 00:26:14.160
Let me figure out why I don't trust it.
00:26:14.319 --> 00:26:16.319
Or uh, oh, that worked really cool.
00:26:16.559 --> 00:26:17.519
How did it work?
00:26:17.680 --> 00:26:19.359
That's like that's that saved me 10 minutes.
00:26:19.440 --> 00:26:25.359
But I wanna, I before I make built my whole life or process around this, like I want to understand the limits of it.
00:26:25.599 --> 00:26:33.200
And that that sort of willingness to ask that sort of second order question, not the oh, did it do what I asked, but like why or how?
00:26:33.359 --> 00:26:34.720
How far can I push it?
00:26:34.960 --> 00:26:42.720
Um, the second order level, which to me is just like just sort of core level curiosity, not the thing in front of you, but what comes next?
00:26:42.880 --> 00:26:45.279
That is gonna be the I think the deciding skill.
00:26:45.440 --> 00:27:00.000
Um and it really is, I think the the way that we sort of get to a better feature is trying to figure out uh from like a leadership perspective and orgs out there trying to do this, like, well, how do I foster curiosity in my team?
00:27:00.240 --> 00:27:02.079
Um, and that's a that's a hard question.
00:27:02.160 --> 00:27:08.880
Because like, hey, we've we've you know, we've been we've for so long focused on I want you know, work ethic, work ethic, work ethic.
00:27:09.119 --> 00:27:20.319
And now it's like, well, I, you know, if I have a AI coding all day, every day, and it's just stopping to ask me questions, the volume of code that it's producing might be the ex almost the exact same as everybody else.
00:27:20.559 --> 00:27:21.119
Fine.
00:27:21.359 --> 00:27:26.720
What distinguishes me from everyone else, a good developer from a bad developer?
00:27:26.880 --> 00:27:38.640
And like all the tools that for the last 20 years the ecosystem has been building, like to track software development as an ecosystem, or you know, like, hey, like how many, you know, uh how many changes did you make to the code?
00:27:38.720 --> 00:27:41.039
And like, you know, how many features did you deliver?
00:27:41.119 --> 00:27:46.160
You know, very straightforward metrics uh when like that was all that we could think about.
00:27:46.400 --> 00:27:48.000
Now that doesn't that doesn't really matter.
00:27:48.160 --> 00:27:53.039
And I think there's so many other places in ecosystems and industries where like it's gonna be hit.
00:27:53.200 --> 00:28:00.319
I think software is getting hit real hard right now with this problem, but it does go back to sort of your point of like the the skill that I see where people are thriving.
00:28:00.400 --> 00:28:02.640
And this was at NASA, this is at my current company.
00:28:02.720 --> 00:28:04.480
Um, it's sort of that growth mindset.
00:28:04.640 --> 00:28:10.160
Uh it's the thing that it's that 21st century critical thinking, um, but with sort of an eye towards what's next.
00:28:10.319 --> 00:28:13.680
Don't just ask, you know, don't be, don't criticize what's in front of you.
00:28:13.839 --> 00:28:17.119
Imagine what could could come and be critical of that before it gets here.
00:28:17.279 --> 00:28:19.920
Help your, you know, think of that planning loop in your mind.
00:28:20.160 --> 00:28:39.599
And if we could start to sort of frame a lot of both our sort of internal education, our strategies, our human capital investments around like creating space for people to feel like, you know, psychological safety, feel safe enough to be wrong and to try new things, then the ultimate result of that will be way better.
00:28:39.680 --> 00:28:47.519
Uh, because if people fail, then they will spend so much time haranguing and try to try to iterate to just be perfect before they show you anything.
00:28:47.759 --> 00:28:51.759
I would much rather have people fail and then show everyone, oh, here's what failed.
00:28:51.839 --> 00:28:55.119
Now you could all avoid that mistake and we can all move forward together.
00:28:55.200 --> 00:29:00.400
And it's again, you know, sharing those ideas across across the ecosystem, across teams.
00:29:00.559 --> 00:29:12.160
Um, and I think that was in NASA, the success that we hit when we started really cooking again for like the really heavy innovation of sort of using AI to really uh to push boundaries.
00:29:12.319 --> 00:29:14.640
Um, it wasn't that we were just using AI.
00:29:14.720 --> 00:29:17.839
Yes, we got people over the hump of yes, AI is cool, let's try it.
00:29:18.000 --> 00:29:23.440
But it's when they started getting AI to encourage them to bring in other people.
00:29:23.759 --> 00:29:25.119
So, hey, like you are.
00:29:25.279 --> 00:29:29.599
So we hired and we said, we're not gonna hire a wing guy, we're not gonna hire a fuel guy.
00:29:29.680 --> 00:29:31.359
Um, we're gonna hire someone who's curious.
00:29:31.440 --> 00:29:35.599
And yeah, they have a PhD in fuel, or you know, like these are locket scientists.
00:29:35.680 --> 00:29:38.640
We bring them in to say, you're not allowed to talk about what your PhD is in.
00:29:38.720 --> 00:29:41.119
You have to talk about literally anything else.
00:29:41.359 --> 00:29:44.880
And that forced them to suddenly, oh, I gotta, I gotta, I gotta go elsewhere.
00:29:44.960 --> 00:29:53.359
I gotta be comfortable being being ignorant again, because I'm I have a PhD and I work at NASA in fuel, and I can't talk about that.
00:29:53.599 --> 00:29:58.640
So I think they have to be like a pseudo-expert in something else, but they have to ask good questions of the experts.
00:29:58.720 --> 00:30:05.279
They're you know, interviewers, they're now uh they're you know, question askers, they're seekers, they're not havers, right?
00:30:05.519 --> 00:30:08.319
Um, they're not bringing just what they know.
00:30:08.480 --> 00:30:13.200
Um, and that that plus AI, like, oh man, that was a powerful combo.
00:30:13.359 --> 00:30:33.920
And I think like those are gonna be the jobs and the roles, uh, the new jobs uh that emerge are the people who like can do that really well, who just like come, you know, come into a company and ask the really hard questions uh and question everything and are good at iterating quickly and finding root cause to, you know, like whatever it may be to get you to that to your product or whatever it is to that next step.
00:30:34.160 --> 00:30:36.160
To flip that on its head, right?
00:30:36.480 --> 00:30:40.319
You at NASA for sure, and you know, me at other orgs too.
00:30:40.480 --> 00:30:42.000
We work with some brilliant engineers.
00:30:42.160 --> 00:30:46.720
Obviously, I haven't worked on with rocket scientists, but some really smart people.
00:30:47.119 --> 00:30:51.759
Um, I think that there's this like um, I don't know if you've ever faced this.
00:30:51.920 --> 00:30:59.599
I'm sure you have, but there are some people who kind of look at AI and they're still very resistant.
00:30:59.759 --> 00:31:03.680
I'm not saying this as a way to like paint these people in a bad light.
00:31:04.000 --> 00:31:11.920
Rather, it's like these minimal use people that are super brilliant, they see everything that's happening and they look at Gen AI and they say, no, thanks, I'm good.
00:31:12.079 --> 00:31:13.599
I can do this myself.
00:31:13.920 --> 00:31:26.240
Um rather than immediately demonizing people saying, oh my gosh, they're gonna be left behind, I'm wondering if there was something inherent about AI that makes folks maybe hesitant to adopt.
00:31:26.480 --> 00:31:32.960
And maybe they're right, are they right about something, or do you think that maybe they were right about something that we weren't seeing at the time?
00:31:33.119 --> 00:31:38.240
And maybe a couple months later we kind of think, you know, maybe they were right about the speed of adoption.
00:31:38.559 --> 00:31:41.599
I think there is something unique to AI.
00:31:41.839 --> 00:31:51.200
I think there's something unique to the sort of ecosystem of AI in terms of like development and it's like rapid up uh, you know, increase uh in only a couple years.
00:31:51.359 --> 00:32:03.759
Um so I think, you know, maybe like we as a community of professionals just you know working in the world, like haven't maybe had time to adjust to the rapid uh growth and investment uh in AI.
00:32:03.920 --> 00:32:04.960
It's everywhere now.
00:32:05.200 --> 00:32:08.640
But the fact that it's everywhere, I think speaks to sort of this universality.
00:32:08.880 --> 00:32:14.079
Um it's sort of a trying to think of like major technological innovations that people sort of undersell.
00:32:14.160 --> 00:32:17.279
Like, oh great, we the printing press, you know, big, big thing.
00:32:17.519 --> 00:32:23.279
Yes, it increased the accessibility of books, but for a while that didn't do much because no one could read.
00:32:23.440 --> 00:32:25.759
Uh but it increased literacy.
00:32:25.839 --> 00:32:34.400
And literacy then became this like powerful force across the world, um, exchanging ideas, the enlightenment, all like that spun out of literacy.
00:32:34.640 --> 00:32:42.640
If books had not been accessible, if reading had not become sort of a thing that you could teach, then you know the the press wouldn't wouldn't have had a you know a big issue.
00:32:42.799 --> 00:32:47.519
Um think of like, you know, metallurgy, nails, screws, the invention of a screw.
00:32:47.599 --> 00:32:58.079
Uh it sounds so simple, but like how much modern construction and buildings like depend on like that ability to just like put two pieces of wood together in a way that like like the sheer strength is different.
00:32:58.240 --> 00:33:00.640
Uh, like the the way it connects is different than a nail.
00:33:00.720 --> 00:33:02.000
It's not gonna come out like a nail.
00:33:02.240 --> 00:33:03.440
Nails were were huge.
00:33:03.759 --> 00:33:06.400
You know, like the I love the phrase like dead as a doornail.
00:33:06.559 --> 00:33:10.960
That comes from this idea that sorry, we're gonna random curios curiosity story.
00:33:11.039 --> 00:33:12.400
Hey, I hope you're ready for this.
00:33:12.559 --> 00:33:30.799
So you the yeah, the nails that we'd use in a door in the medieval times were nails that you would bang in and then bang, like the the, you know, the other side would have the tip pop out, and then you'd bend the tip down so that someone couldn't like pry open the door because it was usually the weakest part of like a castle or you know, some like you know, stone-built you know, thing.
00:33:31.039 --> 00:33:33.680
So doornails were like bent at the end.
00:33:33.839 --> 00:33:43.200
So they were usually not reusable because they had been bent so much that they're like, okay, it's you know, dead as a doornail is in fact, like it's the last time you can use a nail.
00:33:43.359 --> 00:33:45.359
Well, that's like, well, why would you reuse a nail?
00:33:45.519 --> 00:33:46.880
Because they were so expensive.
00:33:47.119 --> 00:33:53.759
So like nails were this thing of you'd born down the house uh and then go get the nails before you went to like move to another town.
00:33:53.920 --> 00:34:01.599
Um, because oh well, I gotta get the nails, um, sifting through through the ashes of your former house to get the nails because that was the thing that was worth keeping.
00:34:01.759 --> 00:34:09.199
So, like these technological things that were like subtle and small, and we don't think about them much today, but like they were the undercurrents of massive change.
00:34:09.519 --> 00:34:11.440
AI is one of those technologies, right?
00:34:11.519 --> 00:34:14.000
The internet, you know, was one of these technologies.
00:34:14.239 --> 00:34:20.639
And I don't, you know, I don't think that people like the people that are are resistant to it are are wrong.
00:34:20.880 --> 00:34:29.119
I think they might have like, in terms of like the fear about what's going to happen to jobs, to the economy, et cetera, like they're those are well-founded fears.
00:34:29.280 --> 00:34:38.239
Um, but I think that I would much rather have these, you know, brilliant people, much rather have them at the table, helping us come up with ideas of like, well, how it's coming.
00:34:38.400 --> 00:34:40.000
Like there's no stopping it.
00:34:40.239 --> 00:34:43.840
We might be able to slow it a little bit, but not by much.
00:34:44.159 --> 00:34:45.920
How do we make this work with us?
00:34:46.079 --> 00:34:48.639
How do we make it not an antagonistic thing, but like a partnership?
00:34:48.719 --> 00:34:51.360
Like, you know, and again, this was sort of the the approach that we took in NAS.
00:34:51.599 --> 00:34:54.239
Like, well, like I don't need it to be a fuel expert.
00:34:54.320 --> 00:34:56.639
I got a fuel expert, but how like how do we help it?
00:34:56.719 --> 00:35:00.960
How do we use it to help us come up with cool new things that that humans would never have come up with?
00:35:01.039 --> 00:35:02.320
Uh, new ideas, crazy ideas.
00:35:02.400 --> 00:35:02.880
How do we test it?
00:35:02.960 --> 00:35:06.000
How do we move faster at the stuff that we already wanted to do?
00:35:06.320 --> 00:35:14.480
And I think like that that shift, um, again, not demonizing, uh, like that shift has has to start with the curiosity they have to want to.
00:35:14.559 --> 00:35:21.199
On sort of like the AI ecosystem, part of the problem is that the, as said, you know, we had this AI winter for a long time.
00:35:21.280 --> 00:35:29.360
And then suddenly, you know, uh OpenAI releases Chat GPT and suddenly it's like, oh man, we crossed sort of a user experience threshold and suddenly the money the oh, the money's there.
00:35:29.840 --> 00:35:31.440
Now, now everyone wants it.
00:35:31.599 --> 00:35:35.119
Um, but there was sort of an AI winter, you have a lot of people interested.
00:35:35.440 --> 00:35:38.320
AI is also a piece of convergence of issues.
00:35:38.480 --> 00:35:43.840
AI is a technology that lends itself with the hallucinations, particularly the early models, to a lot of distrust.
00:35:44.000 --> 00:35:48.239
Distrust because it'll take my job, distrust because it's wrong, and I don't I have to double check all its work.
00:35:48.400 --> 00:35:51.280
So it comes to the market, there's a lot of distrust, there's a lot of fear.
00:35:51.360 --> 00:35:58.960
So you take that fear and you take this AI winter and you say, oh, well, uh let's say I'm a a I want to use AI.
00:35:59.119 --> 00:36:01.360
I am curious, I want to engage.
00:36:01.679 --> 00:36:03.599
Who do I ask to talk about it?
00:36:03.760 --> 00:36:05.360
Who can come explain it?
00:36:05.679 --> 00:36:16.880
Well, the only people in the ecosystem that had high levels of trust that weren't like, you know, sell, you know, uh, you know, selling selling you a bridge, uh, like I'm an AI influencer or whatever it is, were the PhDs, right?
00:36:16.960 --> 00:36:24.239
The deep experts who've just been living in it in, you know, in their basement uh at some company for a long time, just like pushing and pushing on the research.
00:36:24.480 --> 00:36:29.920
Those tend not to be the people who are the best at explaining, you know, what will impact the world, right?
00:36:30.000 --> 00:36:31.440
Because they have the blinders on, right?
00:36:31.519 --> 00:36:40.079
It's like asking the guy who designs the hammers uh for the, you know, like at Husky or some tool company to come in and talk to you at like an architecture conference.
00:36:40.320 --> 00:36:42.559
That dude probably knows nothing about architecture.
00:36:42.719 --> 00:36:46.800
But yes, the builders who build that home or what they'll probably use a hammer, sure.
00:36:46.960 --> 00:36:57.039
He could talk to you about hammer design, but there's such a chasm between like the expertise of designing a like a well-balanced hammer and the metallurgy and then architecture.
00:36:57.199 --> 00:37:11.440
But they are, you can sort of see that like chain of like AI, the guy who knows how to build a really robust LLM, is not the guy who can tell you what is gonna be the impact to like education or the economy once that LLM is deployed at scale in a business environment.
00:37:11.760 --> 00:37:16.320
And that gap of like, well, who are the few people in the middle who have done that work?
00:37:16.480 --> 00:37:28.159
It was like, I count myself like you know, one hand of people I knew before this current boom of people who's sort of trying to do that alongside me, who are asking those like bridging the technology to a real product, but oh yeah, let's try the AI.
00:37:28.239 --> 00:37:30.079
It's not perfect yet, but it's worth trying.
00:37:30.320 --> 00:37:31.360
There just wasn't a lot of it.
00:37:31.519 --> 00:37:37.360
But those, I think, that skill of bridging that gap is the is the piece that is going to make the difference.
00:37:37.519 --> 00:37:40.639
But AI has been lacking in that group for a while.
00:37:40.719 --> 00:37:52.480
And it'll take us four or five more years before we have, I think, a robust enough set of those experiences across all these uh these industries where like, oh, yeah, we we know someone who's done this, broken it, fixed it.
00:37:52.639 --> 00:37:57.119
They know how to deploy and roll out AI um safely, effectively, et cetera.
00:37:57.280 --> 00:37:59.440
Um and it can sort of bridge that gap.
00:37:59.519 --> 00:38:08.320
Um, which is why, you know, again, like if you go to an AI conference uh like two years ago, it would have, you know, you know, gouged your eyes out, you know, ripped your ears off, it would have been so boring.
00:38:08.400 --> 00:38:13.199
Uh because they just didn't know what they're uh, you know, here's a really, really super technical paper.
00:38:13.280 --> 00:38:14.639
And like no one's like, I'd what?
00:38:14.800 --> 00:38:15.760
Like I'm I'm a CEO.
00:38:15.920 --> 00:38:17.679
Like, well, what's this mean to me?
00:38:17.840 --> 00:38:21.760
Um, we are getting it is getting a little bit better, but it's still not fully there.
00:38:21.920 --> 00:38:22.880
We're not fully baked.
00:38:23.199 --> 00:38:25.519
Yeah, I was gonna say it's it's we're still cooking.
00:38:25.599 --> 00:38:27.440
Um still cooking, yes, yeah.
00:38:28.800 --> 00:38:31.599
Um it's definitely getting, I think, more accessible.
00:38:31.679 --> 00:38:32.559
Like research.
00:38:32.719 --> 00:38:47.840
Um, even like I've been in research for a while, and I say that loosely because I'm not I do not have a PhD, but um, my first job in the Bay Area was at a research company, or at least the research arm of a company, and I was surrounded by PhDs.
00:38:48.000 --> 00:38:50.800
Just I was the only one with BS, my title.
00:38:50.960 --> 00:38:55.199
Even my last job, it was uh it was being part of a research group, right?
00:38:55.519 --> 00:39:13.920
Um, so it was just like, okay, well, like you know, research is accessible nowadays, and everybody can have an opinion, and ever and I think even more so, everybody now has the tools to have an opinion that is validated back and it really reads, um, it's more approachable.
00:39:14.079 --> 00:39:16.800
I so I think this is a really fun time to be in research.
00:39:16.960 --> 00:39:22.320
Um, the really good research will always come up above like the AI-generated research, which is great.
00:39:22.480 --> 00:39:24.480
Um, and I'm very happy about that.
00:39:24.719 --> 00:39:33.360
But like now we're just getting more people who feel confident in being able to like talk about the things that they're working on or share about the things that they're doing.
00:39:33.599 --> 00:39:41.679
And you mentioned this really interesting part where you're like, oh man, there's like these genius people that just do not know how to communicate ideas.
00:39:41.760 --> 00:39:45.440
And our last podcast guest, we actually spoke about this as well.
00:39:46.079 --> 00:39:50.559
So there's something that you mentioned, and I'd love for you to explain it really quick.
00:39:50.880 --> 00:39:55.920
So you mentioned something about institutional language versus human language, right?
00:39:56.320 --> 00:39:59.760
And the different modes that like we kind of need to translate between.
00:40:00.079 --> 00:40:11.039
And I'm wondering like how these kind of these two different types of languages help in in um just like your the communication philosophy overall, and like what what this kind of means.
00:40:11.360 --> 00:40:16.000
Well, I would say institutional language is just like a subset of human language.
00:40:16.159 --> 00:40:22.159
Um it's like it's learned, uh, you can think of like onboarding at a you know at a company, like you you get the lingo.
00:40:22.320 --> 00:40:30.400
You know, like if if anyone has ever interacted with anything in the sort of the defense or security space, the amount of acronyms that you have to learn is insane.
00:40:30.559 --> 00:40:32.079
You know, it's just it's just it's absurd.
00:40:32.159 --> 00:40:34.239
And people like they just assume that you know.
00:40:34.400 --> 00:40:42.320
Um, so there is that sort of lingo in every large org about how they talk about their teams, their projects, the code names of that, you know, research thing, whatever it is.
00:40:42.400 --> 00:40:43.760
There's just a lot of that.
00:40:44.000 --> 00:40:48.639
And I think people get caught up in that and they default to that shorthand.
00:40:49.119 --> 00:40:55.920
And I think there's also then sort of an expectation that within large orgs, you say, like, oh, like I trust that person to be the expert.
00:40:56.000 --> 00:41:00.239
So they, you know, they, you know, if they say that, great, I'll just yes and.
00:41:00.480 --> 00:41:03.920
Um so there's not necessarily a lot of healthy pushback.
00:41:04.000 --> 00:41:08.079
And pushback in orgs also tend to be viewed as like hostiles, like, oh, how dare you question me?
00:41:08.159 --> 00:41:09.360
I'm, I, this is my job.
00:41:09.519 --> 00:41:10.559
It's like, okay, great.
00:41:10.800 --> 00:41:18.639
But and I think that's where, like, that's where finding common ground that isn't that institutional language, that isn't that transactional.
00:41:18.800 --> 00:41:22.480
Here's what I gotta do, here's what you do, here's what we do, here's our policy.
00:41:22.719 --> 00:41:34.079
If you want people to start sort of being curious, you know, stepping back, taking a step back and saying, hey, like our AI policies, our whatever approach to this is like, could we, could we redo this, we rethink this?
00:41:34.239 --> 00:41:36.960
Um, it's best to start with neutral territory.
00:41:37.119 --> 00:41:46.800
Uh, and this is why I have found myself like collecting stories, like weird stories from history and things, mostly because they are neutral territory.
00:41:46.960 --> 00:42:00.639
Uh, if I'm in a meeting and I start telling a story or like telling some, you know, like, you know, like uh telling some story about you know hammers and nails in indoor, you know, indoors, uh, or or something, you know, absurd and uh strangely disconnected.
00:42:00.800 --> 00:42:02.559
At the end, that will try to connect it.
00:42:02.719 --> 00:42:07.119
But it's not like either side of that conversation knows more about that random weird story.
00:42:07.280 --> 00:42:14.320
It's a it's about a like some metaphor to collect and say, hey, we're now sort of all on the same page to the point that is trying to be made.
00:42:14.480 --> 00:42:16.159
Now let's have a conversation.
00:42:16.400 --> 00:42:22.800
Um, and that sort of bridging with sort of the neutral territory is, I think, the most effective.
00:42:23.039 --> 00:42:36.559
And this, you can see this come out, oh man, um, in PhD's research communities we talk about, but any technical field, deep experts tend to use metaphors in trying to convert something that is also from their expertise.
00:42:37.199 --> 00:42:47.599
So, like, you know, someone trying to explain like how the internet works to you know, to a gardener, be like, oh, well, like the HTTP stack, all right, it's a lot like uh, and then they'll reference something in like hardware.
00:42:47.840 --> 00:42:49.519
It's like, oh, it's like the computer bus.
00:42:49.679 --> 00:42:51.440
Someone's like, what's that?
00:42:51.599 --> 00:43:00.480
Like they, their reference for sort of what is common in their industry is just so far removed from what that gardener or whatever might be.
00:43:00.880 --> 00:43:08.960
So the best communicators are the ones who have not only lots of stories, but stories that are, again, neutral to the parties involved.
00:43:09.119 --> 00:43:18.000
And I think that has been the most interesting way to bring like CEOs or CTOs that are like AI resistant and AI experts to the same conversation.
00:43:18.159 --> 00:43:25.760
Um, because again, I'm typically when I'm giving these talks or sharing these stories, I'm not there to like tell the AI person, here's here's what you invest in, here's what you build.
00:43:25.920 --> 00:43:26.320
No, no.
00:43:26.480 --> 00:43:27.360
They will always know that.
00:43:27.440 --> 00:43:30.480
And I'm not there to tell the CEO, here's, you know, like here's how to spend your money.
00:43:30.719 --> 00:43:35.199
What I am usually there to is like, here's a risk that neither side is fully appreciating.
00:43:35.280 --> 00:43:36.880
And you guys need to start talking about.
00:43:37.039 --> 00:43:45.840
Because you have like the finance side has a has a part to play in mitigating this risk, the technical side has a part to play, but you both got to be talking about the same risk.
00:43:46.079 --> 00:43:54.960
So I'm sharing stories from history of when people you know blew up the wrong thing or bought the wrong technology and it, you know, like it, you know, killed hundreds of people.
00:43:55.199 --> 00:43:57.039
Like terrible, you know, terrible stories.
00:43:57.119 --> 00:43:59.760
Like now, hey, you see how it's important.
00:44:00.239 --> 00:44:03.039
To do X, let's take that lens and apply it to AI.
00:44:03.119 --> 00:44:05.519
Let's, you know, like that metaphor, that allegory.
00:44:05.920 --> 00:44:08.800
What are the similar risks in the AI ecosystem?
00:44:08.960 --> 00:44:10.400
Okay, now, now we get it.
00:44:10.480 --> 00:44:11.440
Okay, now talk.
00:44:11.599 --> 00:44:20.000
And I think that has been, to me, the most successful way is just is you know, is to bring those sort of really sort of out of left field stories.
00:44:20.159 --> 00:44:24.320
Um, because, you know, then no one feels particularly empowered to say, like, well, you're wrong.
00:44:24.400 --> 00:44:27.039
That story is, you know, historically, you know, inaccurate.
00:44:27.119 --> 00:44:28.159
I was like, it doesn't matter.
00:44:28.239 --> 00:44:29.039
That's not really the point.
00:44:29.199 --> 00:44:30.320
And that that has worked really well.
00:44:30.400 --> 00:44:36.239
So, you know, for all the listeners out there, curiosity, you know, these stories really will help you in a lot of ways.
00:44:36.559 --> 00:44:42.480
You made this incredible crazy switch from NASA to travel.
00:44:42.719 --> 00:44:49.760
You know, and we brought up this part about institutional language, which no longer exists in this at least during this change.
00:44:49.920 --> 00:44:57.840
And you're stepping into an industry, like you said, where you are now the subject matter expert on AI, trying to bring it up, trying to get this adopted.
00:44:58.000 --> 00:44:58.320
Right.
00:44:58.639 --> 00:45:11.599
I think you now, as the builder, right, you're still building all the time, and you now also are kind of tasked with shepherding, I think, this organization into the modern era.
00:45:12.000 --> 00:45:25.840
And you don't necessarily have all of the same exact institutional language that they have, and you're challenged with kind of bringing them up to a matter of fact, your institutional language.
00:45:26.239 --> 00:45:29.119
So how is that how has that kind of been?
00:45:29.280 --> 00:45:36.880
You know, I can imagine that you know, there's obviously a reason why they hired you and why they brought you on right as CTO, obviously.
00:45:37.199 --> 00:45:50.559
There's also this other part as CTO where you're now an executive and you now need budget for things, and you need to go and get people to say, hey, like we want, we, or you need to say, hey, I need money for this, and they are gonna say why.
00:45:50.800 --> 00:45:53.519
And now you need to explain it in their language, right?
00:45:53.840 --> 00:45:55.920
So, what does bridging that gap look like?
00:45:56.159 --> 00:46:18.159
And I guess what are the different types of things uh or areas that you like to focus on when making these arguments for investments, either in different parts of AI, or maybe to um empower different teams that are using AI, um, or to, you know, even train ICs so that everyone has like a better skill set in AI.
00:46:18.400 --> 00:46:24.320
First thing I would say is that I I tend to start in any this this is not my first sort of change management with AI.
00:46:24.480 --> 00:46:26.000
I think the AI has gotten really better.
00:46:26.159 --> 00:46:28.880
The job is still, I think, largely the same.
00:46:29.039 --> 00:46:37.119
And I I tend to approach sort of like a lot of this stuff as like 90% carrot, you know, uh all carrot, no stick, 90% carrot, uh very little stick.
00:46:37.280 --> 00:46:49.199
Um, because people, if you want people to grow and be innovative and try new things, uh, you cannot fear tends not to be the best way, threats tend not to be the best way to get that uh that code of action.
00:46:49.280 --> 00:46:52.000
Um you won't get more deliverables, you will not get new ideas.
00:46:52.159 --> 00:46:56.880
Um you're trying to be innovative, you have to build coalitions of the willing.
00:46:56.960 --> 00:47:00.719
Um and that is a hard culture to build uh of like getting people excited.
00:47:00.800 --> 00:47:04.079
It's like, hey, like come come, you know, like uh you know, come on board.
00:47:04.159 --> 00:47:07.119
Uh, you know, I'm the I'm the I'm the strange new manager with candy.
00:47:07.199 --> 00:47:11.840
Like come to my my lunch meeting, um, uh my brown bag, and I'll I'll teach you stuff.
00:47:12.000 --> 00:47:16.159
And so a lot of it was sort of like showing them what could be done.
00:47:16.320 --> 00:47:23.360
Um, and you know, even with the ICs, like I've been having a lot of like one-on-ones and meetings and demos of like, hey, like let me show you how I code.
00:47:23.440 --> 00:47:31.280
Just like give me, you know, give me a ticket, give me a task, and like we'll do it, we'll do it together, all like 20 people in a room, and like we'll like we'll play this game.
00:47:31.440 --> 00:47:35.760
And they can see, oh wow, AI did a lot, but they also see like where I pushed back.
00:47:35.840 --> 00:47:37.920
And they can ask questions like, why did you ask that?
00:47:38.079 --> 00:47:46.239
Uh so they get to see again, sort of that what are the limits and and sort of learn and can mirror that, and then it hopefully brings up you know good patterns.
00:47:46.480 --> 00:47:59.199
Um so I think once you start getting sort of the people doing the hard work, the sort of in the weeds, and you understand like this, if they make these changes, this is what will happen in terms of speed, efficacy, et cetera.
00:47:59.440 --> 00:48:09.039
You now have at least a small data set that you can now take to management and say, hey, look, 70% of the team is not using this, 30% is this is how much more productive that 30% is.
00:48:09.199 --> 00:48:11.280
If it's not productive, okay, what are we doing wrong?
00:48:11.440 --> 00:48:14.159
Like you fix it, find, find, find something else.
00:48:14.400 --> 00:48:22.000
But usually that collision of the willing means you get more excitement, you means you can trust the data that's coming out of it because people are not lying to you about usage.
00:48:22.159 --> 00:48:32.239
Um, and then you can get real data, you can bring that, and you can make your business case of like, now let me get the laggards uh you know to come on board um in that long tail of sort of buy-in.
00:48:32.559 --> 00:48:37.440
Um and that's sort of like the how you sort of shift that culture towards towards sort of new things.
00:48:37.679 --> 00:48:38.480
It is slow.
00:48:38.559 --> 00:48:43.199
Um it is not something that can be done overnight, it's not something that can be done like you know, top down.
00:48:43.280 --> 00:48:45.920
Um you can't be like, you know, go be smart and use AI.
00:48:46.000 --> 00:48:52.639
Um you can just provide opportunities and space uh and uh maybe positively incentivize.
00:48:52.719 --> 00:48:55.360
Um but that's the really only the way the best way.
00:48:55.519 --> 00:49:03.519
Um and I think one of the things that is discounted a lot um as an expert is as an outsider, when I come in, I have no political cachet.
00:49:03.599 --> 00:49:04.320
They don't know who I am.
00:49:04.480 --> 00:49:06.159
And like, yeah, oh, you're a guy from NASA.
00:49:06.400 --> 00:49:07.119
Okay, whatever.
00:49:07.280 --> 00:49:09.519
I don't know, for example, the travel industry.
00:49:09.679 --> 00:49:10.639
I've had to learn a lot.
00:49:10.719 --> 00:49:14.159
And there are a lot of people in my company who've been in it for 15, 20 years.
00:49:14.239 --> 00:49:26.880
They know way more, but all the weird little gotches of like how, oh, well, how how they do this, you know, weird accounting to like, you know, to do bookings or you know, taxing this, but it's a reseller rate over here, like it's absurd.
00:49:27.039 --> 00:49:30.880
And I was like, I, you know, AY would I would never guess this, AI could never guess this.
00:49:30.960 --> 00:49:33.280
That expertise is still very valuable.
00:49:33.440 --> 00:49:35.360
And it's got to re respect that.
00:49:35.440 --> 00:49:38.000
It's like, I'm not here to blow up your life, I'm trying to make it better.
00:49:38.239 --> 00:49:43.360
But coming and say, hey, you can help teach me, I'll help teach you, and we can go together.
00:49:43.440 --> 00:49:44.719
Again, a coalition of the willing.
00:49:44.960 --> 00:49:51.440
And if you don't respect that, if you don't sort of acknowledge that you you don't know everything, then you won't, you won't get the buy-in.
00:49:51.679 --> 00:49:57.760
It's even better if you can train that person who's excited, that first, you know, that star student who's like, oh man, I really want to try.
00:49:57.920 --> 00:49:58.719
Thanks for helping.
00:49:58.800 --> 00:50:01.039
And then say, great, you teach the next class.
00:50:01.199 --> 00:50:02.800
Like you host the brown bag.
00:50:02.880 --> 00:50:04.480
I'll schedule it, I'll make everyone show up.
00:50:04.639 --> 00:50:09.280
Uh, I'm still a CTL, but you host it because they would much rather hear it from you.
00:50:09.360 --> 00:50:13.599
And if there's anything really hard, I'll step in and like, oh, actually, here, you know, here's that's a hard question.
00:50:13.679 --> 00:50:14.639
Let me take that one.
00:50:14.880 --> 00:50:18.719
But having it come from within, I think buys, gets gets better buy-in.
00:50:18.960 --> 00:50:21.360
People like, you know, to see their peers learn and grow.
00:50:21.440 --> 00:50:23.840
And it feels, again, it makes them feel like, oh, I can do this.
00:50:23.920 --> 00:50:26.960
Um, and I hope, you know, any of my teammates listening to this, they they get that.
00:50:27.039 --> 00:50:31.360
And if they are excited about new stuff, please volunteer for the for the next brown bag.
00:50:31.519 --> 00:50:37.440
Um, because you know, radical candor, like I want, you know, I want people to feel like they can grow and they are going.
00:50:38.079 --> 00:50:43.039
That is the only way I think we all will survive this coming storm because we're moving too fast.
00:50:43.199 --> 00:50:49.599
The system is moving, the ecosystem, the tech is moving too fast to not be constantly growing and learning.
00:50:49.760 --> 00:50:51.840
It's the only skill that will matter at this pace.
00:50:52.000 --> 00:50:55.280
You're now building AI systems at a massive scale, right?
00:50:55.760 --> 00:50:57.360
So here's the question.
00:50:57.599 --> 00:51:00.800
Every parent with a curious kid is actually wrestling with.
00:51:00.960 --> 00:51:01.360
Everyone.
00:51:01.519 --> 00:51:03.920
So lots of this is a lot of pressure on you.
00:51:04.159 --> 00:51:14.719
So your tenure adult years old, your 10-year-old comes home from school, plops down at the kitchen table, and says, Dad, my teacher says AI is going to do all the thinking for us.
00:51:14.800 --> 00:51:17.119
So why should I bother learning the hard stuff?
00:51:17.280 --> 00:51:18.159
What do you tell them?
00:51:18.239 --> 00:51:20.239
And more importantly, what do you show them?
00:51:20.400 --> 00:51:20.559
Yeah.
00:51:20.719 --> 00:51:22.559
This used to be easier because the models were worse.
00:51:22.639 --> 00:51:23.840
Uh, it's harder now.
00:51:24.000 --> 00:51:26.960
Uh, but I would say, I I am worried about this with my kids.
00:51:27.039 --> 00:51:29.119
I have a you know, three and a half year old and a one and a half year old.
00:51:29.199 --> 00:51:32.800
Like, I don't know what school will be like when they actually are like doing real learning.
00:51:32.960 --> 00:51:42.159
I would say it's the same thing I tell a lot of people who are professionals, is that I would say, okay, let's sit down and I'm gonna show you AI.
00:51:42.239 --> 00:51:43.840
And we're gonna, oh, look how cool it is.
00:51:43.920 --> 00:51:45.119
Look at all these fun things we can do.
00:51:45.280 --> 00:51:48.000
And then I'm gonna start showing these, like, well, where is it wrong?
00:51:48.320 --> 00:51:53.519
And I think the challenge you'd have like with you know, five-year-olds is that they don't know what's wrong yet.
00:51:53.679 --> 00:51:55.760
They don't know enough to like, oh, that's the wrong answer.
00:51:55.920 --> 00:51:56.960
A 10-year-old should, right?
00:51:57.039 --> 00:52:00.000
They should have some things that maybe like they just know about the world.
00:52:00.159 --> 00:52:05.039
Um, so showing it that AI can make mistakes is a good first step.
00:52:05.199 --> 00:52:12.480
And I think if you can sort of, if you want to gamify it, it's like, okay, well, now the game is like, how can we make the AI wrong?
00:52:12.719 --> 00:52:19.599
Like, what can what what questions can we ask of the AI to trip it up, to make it wrong, uh, to make it come back wrong.
00:52:19.840 --> 00:52:22.800
And that sounds insidious, um, but it's not.
00:52:22.960 --> 00:52:28.239
Because what you are really trying to do is to get, again, the framing of the AI isn't always right.
00:52:28.320 --> 00:52:37.519
And your job is to come with curiosity, to come frame the questions, come with like you have to know, go find enough information to ask a hard question.
00:52:37.679 --> 00:52:43.360
Um, and that means reading to kill a mockingbird or whatever it is, because you got to come in with a question that maybe be, oh, well, what about this?
00:52:43.519 --> 00:52:47.920
What about this weird reference that uh, you know, Jim says this and Scout says that.
00:52:48.079 --> 00:52:51.199
That is, I think, the way that I would approach it with it with a 10-year-old and with a kid.
00:52:51.280 --> 00:52:59.760
Um, I would probably have may send a note to the teacher like, hey, could I give you some better resources about how to think about AI in your classroom?
00:52:59.920 --> 00:53:08.880
Um, because if they're saying, you know, it's like it'd that'd be like a math teacher saying we don't need to learn math because we'll all have calculators in our pockets one day, which we now we do.
00:53:09.119 --> 00:53:10.559
But math is still valuable.
00:53:10.800 --> 00:53:16.000
And when you get to higher level math, no one is using a calculator because higher level math is is constructs.
00:53:16.079 --> 00:53:22.480
It's it's you know, it's all letters and Greek alphabet, and it's it's crazy, but you still need to know the the the core level stuff.
00:53:22.719 --> 00:53:26.000
Those, that level of pursuit will still exist in the future.
00:53:26.159 --> 00:53:29.119
And you know, like the calculator is a tool, but it has limits.
00:53:29.280 --> 00:53:30.800
AI is a tool, it has limits.
00:53:30.880 --> 00:53:39.119
Um, so let's figure out a way, teacher, parent, student, of finding those limits and then being creative about like how we sort of learn from both, right?
00:53:39.199 --> 00:53:45.280
It's you know, having both, you know, one foot into the AI world and one foot out, uh, and being critical about you know which way we lean.
00:53:45.519 --> 00:53:45.840
Cool.
00:53:46.000 --> 00:53:46.559
Well said.
00:53:46.800 --> 00:53:51.840
And honestly, I think one of the one of the things I'd be really I would love to do this.
00:53:52.000 --> 00:53:54.079
Maybe I should, maybe I should go volunteer somewhere.
00:53:54.239 --> 00:54:01.360
Um, I would love to like go to a classroom and just like pen test AI right there and just make it say a bunch of funny things.
00:54:01.599 --> 00:54:20.000
And I bet like not only would I create the the coolest generation of like pen testers, um, but I think showing showing them that like AI lies and AI isn't uh the smartest person in the room would really level set some of their expectations.
00:54:20.320 --> 00:54:21.760
And I think that's the problem.
00:54:21.920 --> 00:54:30.239
Right now, there's so many inflated expectations and around AI where it's like we've made it so perfect and it can't be wrong.
00:54:30.559 --> 00:54:34.800
Like it's just not beneficial for us if it's wrong.
00:54:35.119 --> 00:54:46.000
And I think uh with this next these next coming generations, I think showing the mistakes that it makes and showing how bad it can be is gonna prepare them for when it's really good.
00:54:46.559 --> 00:54:58.719
Um to be really like still skeptical, still curious, still, you know, knowing that yes, it may be right and it may be even more right than them every once in a while, but it's not gonna be right every time.
00:54:59.039 --> 00:55:00.800
Let's say AI gets really good at coding.
00:55:00.960 --> 00:55:20.239
Okay, so we need fewer software developers, but we might need more product managers because AI is still pretty bad at like sitting down with a customer and like really like chatting with them, sitting over their shoulder, asking, like, oh, trying to understand what their needs are, like putting that in context of the the product market fit and the business, you know, and scalability and cost and margin.
00:55:20.320 --> 00:55:21.360
Yeah, it can help you with that.
00:55:21.519 --> 00:55:25.760
But it's actually still pretty bad because it can't leave the computer and go be curious.
00:55:25.920 --> 00:55:27.440
It can't ask those questions.
00:55:27.599 --> 00:55:32.320
Um, you can't force a customer to go sit in front of your AI and while it while it peppers it with questions.
00:55:32.559 --> 00:55:45.519
So, like even in software where we're seeing this massive uptick, yes, it's maybe sort of like collapsing AI sort of software jobs, but I would say it's actually creating more software jobs that like dabble in product.
00:55:45.599 --> 00:55:51.840
And it's enabling more product managers to dabble and prototype things with AI before they like try to build it at scale.
00:55:52.000 --> 00:55:56.480
And it's making more engineers aware of like, well, oh man, I have to do this a million times a day.
00:55:56.639 --> 00:55:58.480
Oh, like I how do I think about scale?
00:55:58.559 --> 00:56:00.400
You know, because my starp is growing way fast.
00:56:00.559 --> 00:56:06.480
Like there are new jobs that are emerging that are adjacent or wholly new in these fields.
00:56:06.639 --> 00:56:10.159
So, yes, AI is, I think, you know, like changing the way we we think about it.
00:56:10.320 --> 00:56:17.679
But if you're not being curious enough to see how it, how it could be transformed, how there's new opportunities opening up, then again, you're not you're not being curious enough.
00:56:17.760 --> 00:56:19.599
You're not you're not asking enough questions.
00:56:19.760 --> 00:56:28.960
Um and I encourage anyone who's is sort of super dire about AI to think about all, you know, try to think of like all the new jobs uh that could be created when we sort of level up one step.
00:56:29.119 --> 00:56:32.639
Yeah, we'll need more fewer software developers, but that means we have maybe more companies.
00:56:32.800 --> 00:56:36.159
You know, the same amount of software developers trying new things, building new things.
00:56:36.320 --> 00:56:38.719
The ecosystem becomes more frothy and more interesting.
00:56:38.960 --> 00:56:43.840
So you said your favorite example of interdisciplinary innovation is bread and cancer.
00:56:43.920 --> 00:56:52.320
Um, a Japanese engineer who built bakery scan to identify pastries, and somehow that led to better cancer detection across Japan.
00:56:52.719 --> 00:57:01.199
So I feel like you know more about this than me, but like the burning question on my mind is like, who was in the room when someone was like, you know what would really help oncology?
00:57:01.440 --> 00:57:03.199
Croissant recognition software.
00:57:03.440 --> 00:57:04.079
I'll be quick.
00:57:04.239 --> 00:57:12.800
Um, engineer pre is a sort of deep learning era in the US, like he's over in Japan where pastries are like really big thing.
00:57:13.199 --> 00:57:18.800
Like a grocery store chain is like, hey, we have real trouble like scanning like our 300 different types of pastries.
00:57:18.960 --> 00:57:24.159
Um, can you build like an image scanner at checkout that just like tells us how many, like how many bread rolls, which type of bread roll?
00:57:24.320 --> 00:57:27.599
Honey glazed, you know, pecan, whatever, like hundreds of options.
00:57:27.760 --> 00:57:30.559
And this is again like pre-our current AI capabilities.
00:57:30.800 --> 00:57:31.599
He's like, sure.
00:57:31.760 --> 00:57:37.360
He spends like six, seven years building it, almost bankrupts his company because he can't get he can't get the contract delivered.
00:57:37.599 --> 00:57:38.800
He finally cracks it.
00:57:38.960 --> 00:57:45.440
But like in Japan, it's such a big deal how fast this was and how how like quickly they were able to like scan all these.
00:57:45.679 --> 00:57:46.639
Like it, it blows up.
00:57:46.719 --> 00:57:48.159
Like local news is covering it.
00:57:48.239 --> 00:57:53.440
They're like, oh man, bakery scan, this is the coolest, hottest new thing at this grocery chain.
00:57:53.679 --> 00:57:54.800
News is covering it.
00:57:54.880 --> 00:57:56.159
It's like getting talked about.
00:57:56.320 --> 00:58:00.719
And it was covered so much that a doctor saw it on the local news.
00:58:00.960 --> 00:58:06.639
And the doctor was like, so you had one person being curious, like, hey, yeah, I think I can tackle this with AI.
00:58:07.039 --> 00:58:09.840
Let me invent a whole new field of image processing.
00:58:10.000 --> 00:58:14.480
And the doctor was like, hey, bread rolls sort of look like cancer cells under a microscope.
00:58:14.639 --> 00:58:15.679
Maybe we could work together.
00:58:15.760 --> 00:58:16.960
So he just calls him up out of book.
00:58:17.039 --> 00:58:17.679
He's just curious.
00:58:17.760 --> 00:58:19.360
He he reaches out to the engineer.
00:58:19.440 --> 00:58:20.880
He knows his name, it's on the local news.
00:58:20.960 --> 00:58:26.960
He's like, hey, do you want to get together and and figure out how to maybe this bread roll thing can be used to detect cancer?
00:58:27.119 --> 00:58:28.639
Lo and behold, he was right.
00:58:28.800 --> 00:58:29.360
It does.
00:58:29.519 --> 00:58:38.079
And it like, it works wonders uh in the era of oncology for several years before you know advanced AI in like deep learning caught up.
00:58:38.159 --> 00:58:40.239
Um, but like they were they were doing it the hard way, right?
00:58:40.320 --> 00:58:50.159
It was custom tuned uh for this particular problem, but it saved countless lives because one guy on a couch was curious about a local news segment about bread rolls.
00:58:50.320 --> 00:58:58.559
Um, so if you aren't thinking like that, uh then you are missing all of the opportunities that are gonna come up in this sort of AI era because there's so many more now.
00:58:58.719 --> 00:59:02.400
Um it's at everyone's fingertips, not just one engineer and one doctor.
00:59:02.559 --> 00:59:05.599
Everyone can be sort of tapping into new places, trying new things with it.
00:59:05.679 --> 00:59:08.400
So yeah, it's it's my favorite story because it's it's so absurd.
00:59:08.480 --> 00:59:09.840
But again, save lives.
00:59:10.079 --> 00:59:11.280
James, thank you so much.
00:59:11.360 --> 00:59:12.400
This has been a pleasure.
00:59:12.639 --> 00:59:16.239
Um honestly would listen to you tell stories all day.
00:59:16.320 --> 00:59:17.760
Uh, you're a great storyteller.
00:59:18.000 --> 00:59:21.199
Um, and it likely makes being at your org a fantastic time.
00:59:21.440 --> 00:59:22.239
Thank you so much for having me.
00:59:22.400 --> 00:59:23.599
It's been really great conversation.
00:59:23.679 --> 00:59:25.039
Um, love talking about this stuff.
00:59:25.280 --> 00:59:25.599
Always.
00:59:25.760 --> 00:59:29.920
And then um to wrap up, uh, where can people find you?
00:59:30.000 --> 00:59:31.360
Uh, where are you gonna be at next?
00:59:31.440 --> 00:59:33.119
Are you working on anything that you want to talk about?
00:59:33.360 --> 00:59:44.800
If you are interested in our travel software, uh it's a collection of companies, uh, Regiono, ResD, and Checkfront, uh, and a new company called Manifest, uh, are all in our sort of network and we're all sort of pushing the boundaries and growing.
00:59:44.880 --> 00:59:47.440
Um, so those are great to check them out if you are traveling.
00:59:47.519 --> 00:59:51.840
Um, and then the best way to follow me is on Twitter, uh James underscore MTC.
00:59:51.920 --> 00:59:58.000
That's where I you know wax philosophical about AI and politics and and how hopefully it'll change the world for the better.
00:59:58.159 --> 00:59:59.440
Um, yeah, catch me there.
00:59:59.679 --> 00:59:59.920
Cool.
01:00:00.079 --> 01:00:01.280
Well, James, thank you again.
01:00:01.519 --> 01:00:05.519
If this episode helped cut through the noise, like or subscribe so you don't miss what's next.
01:00:05.679 --> 01:00:07.039
Thanks for spending time with us.
01:00:07.199 --> 01:00:09.760
Until next time, stay curious.