Ep. #9, Constraints, Creativity, and Competition

Adam Zimman: So I was thinking about this conversation this morning on creativity and automation and, what is the kind of differences or how do we define things? And I coincidentally, last night, went to a talk, a panel conversation that was hosted by UC Santa Cruz, and they actually were talking with a member of the Engineering Computer Science department, a member of the Astrophysics department, and a member of the Psychology department were having a conversation about creativity and community.

And it was really kind of interesting the way that they spoke about the two being so intertwined and so, that was part of the things that I wanted to hit on today. But I continue to find it interesting and also a bit frustrating that I am apparently caught in the infinite well, that is supported on one side by "everything should be AI," and the other side of, "AI is the devil and it should be destroyed at all costs."

And trying to sit somewhere comfortably in between those two chasms seems extraordinarily difficult these days.

Kim Harrison: Yeah, it's not an easy one.

Heidi Waterhouse: I think that's always true of things people want us to be enthusiastic about. I have friends who are like, no, I have never read or watched Lord of the Rings movies because too many people were enthusiastic about them.

Adam: That's fair. But I think of it from like, a technology perspective. And, you know, all of us have been doing this for a bit, right? And I think back to Cloud, I think back to the Internet, there was--

Heidi: Nobody had to make us excited about it.

Adam: No, no, that was the thing that was just occurring to me was that, I feel like-- Going back to when the Internet started to become A thing like, I think there were people that didn't embrace it at first, but there wasn't like a vehement "oh, this is destroying our society" kind of perspective. And I feel like with AI, it's just, I get it. There are reasons, right?

You know, obviously there's, I think that there's a lot that's being done that's being lumped into the "this is AI" bucket. And you know, I still contend that one of the strongest reasons why AI has struggled in this capacity was that there was a wholesale pivot from all the tech bros that were like all in on blockchain and all its variations to, "oh, wait, no one's ever going to use that. Let's pivot to AI."

And so like all the kind of obnoxious excitement and you know, interestingly, like the complete and total disregard for environmental constraints with regards to power consumption and resource utilization, it's just like they wholesale pivoted to AI. And I think that there are things that are being done in terms of the way that AI is being rolled out and that have like any kind of fiscal conservatism has gone out the window. Any type of rationalization or consideration of. Is this what we--

Heidi: Like, what do you mean, a trillion dollars?

Kim: Yeah.

Heidi: That's not a real number.

Kim: Yeah.

And I don't think the PR behind AI has been, I don't want to say it's not smart, but the PR stories behind it make sense if you want to sell it, but do not make sense if you want to make fans amongst the people.

Adam: Well, and I think that that was part of the, you know, what I talked about with regards to like this kind of like blockchain pivot. Right. I feel like that was very similar.

Kim: Yeah.

Adam: Where there was a lot of like Web3 talk of people who were, you know, "Web3, Web3 crypto, crypto, crypto." You know, like that was like the only words that would come out of their mouth, even though they didn't have any meaning.

You know, people were doing stuff with blockchain and like had convinced themselves that they had value. But you look at where they are now. Like, I think you know, the primo example of this is obviously NFTs, right. No one's really talking about their NFT investment portfolio right now. Haha.

Kim: Wait, wait, I'm pretty sure there's some documentaries on Netflix about all the scams.

Adam: I mean, it's unfortunate because I think that there are aspects of AI and I think that the positive side is, you know, I think back to the conversation we had with Betty where, you know, she talks about that she was never a programmer, right? She's always been a marketer and like, but she's got ideas, she's got things that she would love to see exist and the way that AI has been able to lower that barrier for her, and I think that's very real.

I think that, you know, I saw this, you know, kind of early days at GitHub as well, where the introduction of GitHub was again, like one of these inflection points where the barrier to entry got lowered for so many people. Right? It was this notion that, you know, Heidi, you've talked about, of this democratization of technology where suddenly people who never had access because of the step function or the burden of cost to be able to use something was so high and all of a sudden there are aspects that are of that that are getting lowered.

Kim: Yeah, I'm seeing examples of that here in small beach town where you have small business owners who are like, I need a new menu and I want a new layout, or I need branding and I can just spend two hours creating these things and have it done and get it printed and be off to work. And one of the kids, one of Jorge's nieces pointed out, I can tell who uses it. It all has the same font.

Heidi: Oh, Ian was super mad. Like there are people using AI to write their talk tracks. And you can tell because I know what you sounded like three years ago and you don't sound like that right now. And I'm like, that's super interesting because there are so many useful things we could be doing with it. My sister works at Kaggle, which is Google's sort of moonshot AI.

Adam: Mhm.

Heidi: Well, Alphabet, whatever. And she's like, yeah, no, we have a thing that's basically like a whale dictionary. We are pretty close to understanding what they're saying to each other. It's under lock and key because they're really worried about people using it to basically honey pot the whales and kill them.

Kim: What?

Heidi: But--

The scrolls at Herculaneum next to Pompeii that were burnt, they're using AI to read those. There are so many cool uses and I don't understand why we are using the boring things. I'm like, if you don't care enough to write that memo, I don't care enough to read it.

Adam: Well, forget the memo. The book. You know, right? This is something that I'm seeing more and more. It's the book, the blog post, the article, whatever. You can totally tell when you know, things are completely generated by AI. And I think that this is where it's like, okay, well I guess that's perfectly appropriate if you think that the only consumer of that material is going to be the bots. Right?

There was a company, I forget what they're called this week that was a startup that was purchased where their offering was that they would rewrite your website so that it was more easily legible by AI.

Heidi: But that makes sense if Google has gone to search is AI powered.

Kim: Yeah, it's the new SEO.

Heidi: Like if we really are in a world where the bot is a mediation layer between us and anything.

Adam: Well, in fairness I think that this is an inevitability of our own creation. Right? Where this was part of the byproduct of the way that the Internet was structured. You know, when we moved away from AOL keywords. Haha.

I'm not saying that that was the solution that should have persisted but you know, we started to think about, how do we start to catalog all this information? How do we catalog and index and reference stuff?

Heidi: I have a favorite indexing story, but I will say Bluesky had a very lively conversation yesterday about content and how user generated content was sort of the original sin of understanding the difference between art and creativity and something that you could get other people to do the work and you could capitalize off it.

Adam: Interesting. Say more about that.

Heidi: So--

When you write a blog post, you think of it as your writing. When WordPress pours it into a format and extrudes it or somebody scrapes Reddit, they don't think of it as writing, they think of it as content, which is sort of a slurry of all of the creative and non creative outputs of humans. And content is what we sell, when we as humans create something, when we create Flickr or Tumblr or something like that, and the consumer is the host and they regard all of these things not as valuable in themselves, but as valuable in aggregate as something that they can resell.

Kim: And also taking it out of context because then you have the power, like you said, you could strip it of the various elements and just take the pieces you want.

Heidi: Mhm. And I think the most extreme example of this is like the Twitter porn bots, where you take people's pictures that they have put up because it represents a moment to them that they're interested in sharing with the world. And you strip and, literally strip the people, and make them content for somebody else to consume without knowing any of the context.

That's a very extreme example, but I feel like it sort of represents what we're struggling with when we say creativity. Because if the end of something is contextless, why does it matter who we are that made it?

Adam: Yeah. I guess now is probably a good time to introduce the primary theme that we wanted to talk about today. And this was something that's come up in conversation a number of times is this balance between creativity, and we should definitely talk about how we define that and think about that, and automation. And I think this is something that we've touched on in previous episodes, but we wanted to kind of dig into a little bit more today.

Thinking about this idea of something that we've brought up in the past is creativity. Is this, I think oftentimes associated with the creation of something new and something unknown, or the combining of things that haven't been combined before.

Some way of being able to take things that previously haven't been put together and, or done and do them in a way that all of a sudden you've got something new. As opposed to automation, where automation, as we define in the book and as we kind of articulated good automation-- Because I think that there is-- You can automate things that you have no clue about. And we get to that in a second.

But ultimately the way to think about automation is how do you take a process that is well defined, consistent, predictable, and be able to take away the need for manual intervention or kind of manual steps in that process so that actually you have a start button and there's a predictable output.

Heidi: So creativity versus automation. So we've defined automation and we've kind of defined creativity. But I also think that it's very difficult for us to understand why so many people think that creativity is a uniquely human attribute.

Adam: Yeah. You know, as I mentioned in this conversation that I was got to go to last night, the aspect of creativity that is not hard coupled but you know, linked with community and as a way of being able to look and say, one of the greatest ways to be able to introduce, you know, the idea of, you know, things that haven't been put together before is you introduce another individual, whether that's human or otherwise, that can bring a different perspective.

And I think that this is something where I was thinking about this in the context of, well, what does this mean for when you are working with, you know, the robot with AI? And I think that there is a aspect of that that can spark creativity.

The challenge that I see is the diminishing returns or the kind of convergence of groupthink that kind of comes about when we're all using the same models or we're all thinking of the robot as our thought partner, that over time, and I think that this can happen quickly or this can take a little bit longer, it does stop looking like new ideas. It starts looking like consistent and sameness.

Heidi: But that happens with every community. Okay. So I frequently think of fanfic because it's one of the most creative generative communities I know about. And there are samenesses that become agreed upon standards that like this character is really only interpreted this way. And anybody outside of that gets negative reactions. They get basically disciplined out of expressing a character this way because it's a community.

If they were just writing on their own, they could write Whoopi Draco all they wanted. But it's not a thing in that community. And I'm fascinated by the idea that we just sort of accelerate that when we're using an AI partner. But it's a very human instinct to say, like, let's converge on a story, let's converge on a truth, a reality, even if it's a fictional reality.

Adam: But I think that one of the things that's distinctly different right now with the AI conversation is that notion of the placation or deference, where the AI will just concede. Right? There isn't going to be the pushback.

Heidi: Mhm.

Adam: Because they're all kind of programmed to be amenable and to feed that kind of hero like outcome for individuals to make them feel validated. Right? And so I think in a community there is, we talked about this before--

One of the most powerful things in a community is someone saying no.

Heidi: Mhm.

Kim: Yeah. And those different perspectives.

Adam: Yeah.

Kim: I mean, we've got into this, in the United States, the whole DEI conversation. But truly bringing people from different backgrounds and lived experiences brings different perspectives. And those perspectives. Yeah. Like it's--

Adam: Who knew?

Kim: Yeah.

Heidi: Next up, fire.

Adam: Haha!

Kim: But okay, so being active participants. And maybe this is the thing, right? The AI cannot actively participate the way that other-- Like we're a community, we choose to be here. We have some level of understanding and respect. And so it's safe for us to push back and challenge and build upon.

Adam: Yeah, I mean, I think that there's also you know, another interesting theme that, you know, came up in this conversation last night that's top of mind for me was thinking about collaboration and the distinction between collaboration and competition where I think something that is not uniquely American, but I think is common to kind of American culture is this competitive-- Can you do better than your peer or can you, like when you are in a community, right. Can you kind of one up people?

Is that the way that, you know, we get the best ideas is by looking for ways to improve upon or beat the others in the room or the others in the space? As opposed to, there are a lot of other societies and communities in the world where it is thought of much more as a collaborative effort. And this notion of how do we build something together and how do we move things, you know, kind of forward as a community that is you know, it seems to me like a healthier perspective, but not always the one that comes to the forefront.

Heidi: All right. I think this is a question about constraints and creativity. Some of the poems that we consider most meaningful in the world are very constrained. They're sonnets, like Shakespeare's sonnets. They're very constrained. They have a rhyme scheme, they have a rhythm.

Like you can only write certain things, but because of those constraints, you have to work harder. It's a friction that makes you better. So in that sense, I think that competition is one kind of constraint where it's framing how much friction there is in what you're doing. It's possible we don't need friction for everything.

I think this is where we're sort of stuck, is like, we don't need to have a zero sum understanding of the world for everything. And I think that's what competition's toxic outcome is, is that like, if I lose and you win, there's only one winner.

Adam: But can you have constraints without competition?

Heidi: Probably. But I do think that the way American society is structured right now, that's the constraint we're operating under.

Kim: I want to take us sideways a little bit. One of the people that we, a couple of people that we know, Catherine Hicks and Ana Hevesi, I apologize if I'm saying their names incorrectly, wrote about this A Cumulative Culture Theory for Developer Problem Solving.

Heidi: Right.

Kim: And actually dug into data around this. Like the idea of, you know, the one hero developer who has all the answers and solves everything versus exactly what you two are suggesting, like could we be working collaboratively and how do you find solutions, how do you innovate and what does that look like?

And they found that the group collaborative environments typically turned out better outcomes. They were actually innovating stronger ideas faster, they were problem solving faster and everybody was less stressed out on top of it. And they actually looked at data, like they actually dug into this. So I mean I think you are onto something. The two of you are onto something.

Adam: Sounds like we should get some experts on this.

Heidi: Yeah.

Adam: I mean I think that this is where the crux of the conversation around AI is kind of where I'm struggling in terms of the outcomes or the future that I want to see. Right? Is that I truly believe that there are so many things in our world that would benefit from automation and reduction of toil for individuals. And I also, you know, like I talked about earlier, this notion of lowering the boundary or the burden for someone to use technology in a way that they want or need.

But how do we make sure that we don't, you know, seclude ourselves in our own little Yes Bubble that is, you know, just making everything kind of steered towards an outcome or an output that is part of that training data on the LLM as opposed to something that is creative or uniquely new or uniquely us.

To your point earlier, Kim, of, you know, I think it's not lost on many people that AI generated images or websites or something like that, they all look the same. And you know, for the person who's doing something for the first time, they may not notice that.

But if it goes unchecked then do we end up in a situation where all of a sudden that aspect of creativity and design starts to look really consistent or like normalized?

Heidi: Haha. Like Corporate Memphis.

Kim: Yeah. I mean this also reminds me though of like when WordPress and Squarespace and you know, Wix, all these platforms that helped us build, build websites and do design things because you had I'm in marketing, suddenly you didn't have to know Photoshop or hard code a site, you could just throw it up and you knew like ah yeah, that's one of the 12 main themes which, for 90% of us was fine because really you just want the person to see the information and then the 10% that truly wanted to stand out and do something unique, they would take the time and the effort and really think about how they could create something completely different and unique and showcase their efforts.

And then you would know like, oh, okay, that's a designer versus they just needed a website to sell their books or shoes or whatever.

Adam: Yeah.

Heidi: I think my AI concern and I'm not going to say like the reason I wholly reject it because I agree most of us are going to have to figure out how to fall in the middle. But it's so bad at context and context is so expensive and it really makes you admire how much context humans are carrying around in their brains all the time.

Adam: Yeah.

Heidi: And I'm fascinated by what it is that they want to be computing. I was at a really cool talk at Open Source Summit where they talked about how databases needed to be less like time series and more geographical. So the question, "should I take an umbrella tomorrow," "weather in Minneapolis," and "severe storm warning" are all the same question, but they have no obvious linguistic links.

They are a cluster. And how do you cache that? How do you say return the weather for this location? It was a super interesting talk which I understood about 10% of because I'm not a databases person. But the idea of having these semantic geographical storage clusters to be able to do caching was super interesting. And I'm like. But I never hear anybody talk about data center caching. It's all compute.

Kim: That's a great example, Heidi. And I think again this, I see a lot of PR happening. There are different people who have different talk tracks and different motivations for why they want to talk about the things they're talking about. And so a lot of what's making it into the larger, you know, tier one news cycles has more to do with the people at the top of these companies that need to get more funding versus, Heidi, what you're talking about, which is real world use case for like oh, actually that's interesting.

Like this is where like I would challenge both of you, especially Adam, because you do hang out with VCs, with executives. Like certain people have a desire to sell certain stories.

Adam: Oh yeah.

Kim: But also just keep in mind if that's the thing that we constantly are talking about at the tier one and it's the large models and it's the huge environmental impact and it's the big, are we paying attention to the small models and the possibility that like all the other things that could be possible with this, that is very interesting.

Adam: Sure. And I mean, I think that this is, this is where, you know, kind of my opinion falls is that like, I do believe that AI, in terms of large language models and the idea of being able to use natural language for computer interactions, is a net benefit.

I honestly believe that there is a net benefit to the things that are being built. I think where I kind of diverge a little bit is number one, I think that there is a lack of responsible social behavior in terms of the way that they're not only being built, but the way that they're being forced and pushed upon people that don't want or don't need them.

Heidi: That's a real lack of consent.

Adam: And I think that like this is, you know, goes back to the, you know, kind of core premise that we talk about in the book. Right? This is how do we do a better job as builders of actually paying attention to our users and what they want and what they need at a particular time. Right?

It may be that a few years from now that's what they need, but I would argue that the vast majority of users don't need that kind of like AI incorporated functionality when the, you know, confidence rate is below 50%. Right?

Like that was not that was not what they needed right now. But like to your point, Kim, there's a certain PR aspect that I think the decision was actively made that if we don't get, you know, the planet on board, you know, our PE ratio is screwed. Haha!

Kim: You know, the entire context of the Internet copied.

Adam: We only need 6 billion users to become profitable in 40 years. Haha.

Heidi: Yeah. And the profit motive is an interesting driver because it's so difficult to figure out who your ideal customer is when you really have to maximize for scale.

Like there are so many things that only work at scale and AI is kind of one of them.

So we can't be-- These organizations don't feel like they can be selective about how to roll it out. Because if it's not widely adopted, it's not going to make their investors happy.

Adam: Yeah. And that's my point is that it is truly like an investment or kind of like sunk cost driven decision where, you know, and I think this is true and has been true even before AI, where, you know, like if a company makes a big bet on a product or a feature, like you better believe that they're going from a PR, from a marketing perspective, they're going to invest heavily on its adoption. Right?

Because they, they need people to use it and for it to be successful and people to, you know, in the kind of like enterprise sense, you need people to buy it. And this is why, you know, like bundling and things like that have always existed where it's like, I'm going to bundle this new thing in with the thing that you actually know you need. And that way I can claim, you know, kind of monetary value in it, you know, on my books.

But yeah, I do think that the aggressive nature of it and the way that things have been rolled out, like I still come back to that, you know, the machine that was built around blockchain. Because I think about how that dominated the use news cycle for, you know, a number of years. Right? Then it like disappeared overnight.

Heidi: Right.

I feel like every technology popped bubble leaves us infrastructure we feel compelled to figure out how to use.

Adam: Well, I mean I, I think that there is definitely like again it's this step function of abundance. Right? It's like all of a sudden there's like so much more compute and everything. I think the question that comes to mind is, you know, do we find use cases that are viable for consuming that?

And I think that the interesting part right now is that we've got a number of organizations, companies that are pushing for increased capacity and are presenting the PowerPoint graphs to the investors that show the kind of exponential or logarithmic curve of growth that is justifying this.

But I wonder whether or not is there the option where you think about like the benefit of local models, you know, is this just another swing of the pendulum of going from local to centralized? Do we figure out the governance and, or you know, kind of security and access control for being able to run personalized models in a cloud context?

So like the, you can run it in a data center because you don't want to run it on your laptop, you know, because--

Heidi: It'll sound like a jet engine.

Adam: Yeah, like you actually want more compute than that. But is it something where that becomes a tenable model?

Heidi: Mhm. Can you run it in a bucket? And does it get efficient? I think that's because we've been working so much on capacity, we haven't really had to focus on efficiency.

Kim: We gotta find these guests. Yeah, because I feel like there already are a couple examples of people using smaller local models and like what does that look like? How is it actually useful? Because I think we kind of touched on this. Like sometimes you're doing a thing and it's fun, but like what is the output what is actually useful for you in that moment?

Adam: I mean, I'd love to find, to see if there are examples of, you know, like I know that I would love to find some really solid examples of AI in the context of automation. I mean I think that that's, that's easier. But what may be harder is could we find examples of AI in the context of creativity?

Kim: I think we could. Like for instance, like our friends at Jon Deere. Is anybody using a John Deere product with some kind of AI strapped in where like the farmers in the field are now doing something more interesting?

Adam: Well, I don't know about more interesting, but I think maybe more productive. Maybe?

Kim: That's interesting if that's your job. Right?

Adam: Yeah.

Heidi: Or the the crop layout weirdos. Like how do you design crop art, which is a thing.

Adam: Or just think about it from crop rotation perspective. Right?

Heidi: Yeah.

Kim: It could be both.

Adam: Yeah, should be both. It absolutely should.

Kim: It should be both. Haha.

Adam: No, I mean I think that these are kind of interesting things. You know, like it would be good to continue these conversations with some individuals that are more focused in some of this work because I don't think that it will get us our answers by simply asking the AI.

Kim: And I don't think these are examples that are talked about as loudly in top tier media. You know, if you're reading some niche newsletter or media group that's focused on it, I'm sure these stories are out there. But the vast majority of us that look at, you know, like breaking news every day, that's not what we're hearing about.

We're hearing about how these models are at the global level. I'm gonna take over and do all the things not the actual day to day, like help me get the crops on the ground or I'm an artist and like I still get to do art, but this automated a piece of it.

Heidi: Yeah, like how do you scale up a drawing to some kind of like island sized mural which is a thing that's happening in Minneapolis right now.

Adam: That's kind of cool. I mean I also think that, you know, I'm really interested in what are the automation areas that continue to cause toil that we're not going after with AI.

Heidi: Mhm.

Adam: I think that there's so much you think about like the way in which we still have to fill out information or fill out forms.

Heidi: Oh my gosh. Yeah.

Adam: Whether for the government or for like--

Heidi: Job searching.

Adam: Yeah. Just whatever it is. Like, I want my little AI bot that just like, can fill out the form for me.

Heidi: Mhm.

Adam: Can you fix that? Like, that's a pain in the ass.

Kim: And we know that the recruiters have bots that can read and understand everything you sent on the PDF. So why is it not automagically putting it in the damn form?

Adam: Yeah, and I think that some of those things are improving, but slowly. You know, it's funny, I think the one that so many people always come back to is you know, housework.

Like, this is a representation of toil that you know, would be great for robots to solve. And we've, we've been picking at the edges on this, but I think that-- Are there people working on that?

Heidi: They are, but almost none of them are experts in housework.

Adam: Yeah, see, this is where like, I think this is the thing that it always kind of comes back to with AI, is that AI looks like it does a really good job on things that you are not an expert in.

Heidi: Mhm.

Adam: You know, if you happen to be an expert in a particular field of study or a particular, you know, aspect of a thing and you look at the AI output, you very quickly are just like, no, this is not there yet. Anyways.

There are some problems and challenges that I think that we are benefiting from AI. I think what I'm missing or what I long for is a better balance with regards to the people that are building it, to be able to recognize that it is not going to solve everything and it is not actually the answer or solution for 100% of the challenges that we as humans face.

Heidi: Yeah, I want it to be a tool, not the tool. And I want to be able to turn on, I need an absolute answer versus I need an approximate answer. Like lots of times an approximate answer is fine.

Adam: Yeah. And I'm okay. I would actually prefer a response of "not enough data to provide a response" or "I don't know," than, "oh, well, I made this up. Or I found it on some random person's website that, you know, swears it's true."

Heidi: Yeah.

Kim: Or like my phone telling me "20% chance of rain," while it's raining. Like, yeah, buddy. Haha.

Adam: Yeah. Nailed it. Haha.