The text below is a transcript of audio from Episode 27 of Onward, "Understanding AI and its implications."
Disclaimer: This transcript has been automatically generated and may not be 100% accurate. While we have worked to ensure the accuracy of the transcript, it is possible that errors or omissions may occur. This transcript is provided for informational purposes only and should not be relied upon as a substitute for the original audio content. Any discrepancies or errors in the transcript should be brought to our attention so that we can make corrections as necessary.
---
Cardiff Garcia:
Hello, and welcome to Onward, the Fundrise podcast, where you'll hear in-depth conversations about the big trends affecting the US and global economies. We are recording this on Wednesday, September 20th, 2023. And before we start today's show, we want to ask that you please keep rating and reviewing the show in those podcast apps. We really, really love hearing from our listeners, and we've heard from a lot of you, we want to hear from even more. Also, evergreen reminder that this podcast is not investment advice, it's intended for informational and entertainment purposes only. With that, let's get on with the show.
This is the AI might take our lives, but it'll never take our jobs episode. Or is it the other way around? I don't know. I'm Cardiff Garcia, Bazaar Audio. I'm joined as always by co-host Ben Miller, CEO of Fundrise, in the flesh too, not an AI version of Ben Miller, I think, but how would you know? Ben, how are you?
Ben Miller:
Hey, Cardiff, how are you?
Cardiff Garcia:
This is the AI episode. This is a big deal. I can't wait, Ben, to discuss this with you because as I understand it, you've been thinking a lot about this in the last few months and even going down a little bit of a rabbit hole yourself on it.
Ben Miller:
We have machine learning people at the company, all these software engineers. We're building applications with it at Fundrise and we're investing in it. I've tried to get my mind around it, and obviously I'm not a mathematician nor an ML expert, so I can only do so much, but I feel like there's so much here. The way I think I'd frame it, is basically everyone needs to understand AI to leverage it, everyone can use it and to anticipate how it might affect you, your kids or your career or your country. So there's a lot here.
Cardiff Garcia:
There's a lot here. There are those practical reasons that you just mentioned for needing to know AI. I also think just in terms of understanding the world, it's helpful to see how much we're already using AI. I feel like AI has gotten a ton of attention in the last year in particular because things like ChatGPT, those language models have emerged and everybody's started to use them and they've had this uncanny ability to mimic a conversation that you might have online with another human being, and that's been the thing that has really raised AI to a level of prominence in the societal and cultural conversation.
But actually AI's been around for some time and we've been using it more and more even outside of those language models, and it feels like we're in a moment now where everybody's trying to figure out just how big a deal it is. Does that roughly track with your experience of it in the last year?
Ben Miller:
I probably have some nuance in there. An example of that is our CTO, one of my co-founders would say, it's not AI, it's machine learning. And he would argue that none of it is AI and that it's all machine learning. And then you'd have to start saying, well, what's the difference between machine learning and AI and when would you cross this threshold? But what's happened is that we have used machine learning a lot. We can talk about what that is. And now we're doing something that seems different. It seems more like an artificial intelligence, the new AI, the new machine learning. So something's different. Something's happening, and implications are enormous.
Cardiff Garcia:
The differing definitions of AI is itself a point of nuance, machine learning versus other things, and in particular the fact that AI being the use of computational statistics to try to mimic certain human thought before we even get into the practical implications and the implications for you and for Fundrise and so forth. It's worth noting that AI is also really starting to challenge the very notion of what it means to be human. The idea that certain thought processes that were thought to be uniquely human may actually soon be at least mirrored, if not outright mimicked and reproduced by machines.
That leaves humanity with some bigger questions to sort through, which is why I love talking about this topic so much, because I think it really gets into some core issues of who we are and not just what we do, not just our jobs, but who are we fundamentally as people?
Ben Miller:
Oh man, people are always so good at that.
Cardiff Garcia:
Of trying to figure that out, of answering that question.
Ben Miller:
Good history of dealing with those questions well. So it'd be interesting to lay out some of the case for why this is a big deal and then get into some of the specifics and come back out to the implications again.
Cardiff Garcia:
Let's do it. Let's just start with that very question. Why do you think it's so important to be understanding and talking about AI right now? What do you think are the big implications of the emergence of such powerful AI?
Ben Miller:
Well, I'll start with a bunch of studies, because I think it's fun to see what all the talking heads are saying. So if you go out and do some research, you'll see that there's all these reports now out by PWC, which are one of the big four accounting firms or Goldman Sachs and McKinsey. They're saying things that are along the lines of AI is likely to increase global GDP by 26%, according to PWC. So that's $15 trillion. Goldman Sachs is saying increase of 7%. So that's 7 trillion per year by the way. Per year.
Cardiff Garcia:
That's a massive number. I'll just pause to put that in context. We feel like the US economy had a pretty good year these days if it's growing at two to 3%. Imagine now almost tripling that, that is radically improved economic growth. That is transformational stuff where your living standards double in a very short amount of time in a matter of a few decades as opposed to half a century or something like that
Ben Miller:
Or faster. The big range of these numbers really the question essentially how powerful will the AI become? Because it's still going up the curve rapidly. GPT-1, two, three, four, each one is step change better and then how fast we adopt it. So those are questions that we don't know the answer to today. But some of the assumptions these companies make are that it could automate away 40 to 70% of workers activities.
Cardiff Garcia:
That if you just look at the individual tasks that all workers do and you add them all up, that as many as half of the current tasks that we know, 40 to 70% as you said, could be subject to being automated away, leaving obviously still some chunk of tasks that humans can do and crucially opening the door perhaps for new tasks that we haven't imagined yet to emerge, but we don't know what they are and it would at the very least represent massive radical change of the kind where many people would probably benefit enormously, but also perhaps if not handled the right way, many people might suffer because disruption is never easy. It can lead to progress, it can lead to some amazing miraculous things, but it's not easy on everybody.
Ben Miller:
That's the big implication that we can talk about at the end, giving some more stats, because these are outrageous. I think Goldman's the one who said that they expect AI to have equivalent productivity boost that was 500 times the personal computer, so that's a good one.
Cardiff Garcia:
So with all that said, I think Ben, we should boldly move forward and really try to get at what's going on with AI having established some of the big potential societal implications, and we'll be discussing more of that. But first, why don't we just start with an attempt at explaining what AI is and how it works.
Ben Miller:
What the heck is AI? Okay, so there's a lot of ways to answer this, and there's a spectrum of how much trouble I get into with one audience or the other. So is the question, how does it work? Or is the question, what does it do?
Cardiff Garcia:
Why don't we reverse that? Why don't we start with what does it do, which is something that we can probably try to get at in normal human language terms before we go into the harder challenge of explaining how it works, which might be a little bit more complicated, but let's establish the stakes first. So what does it do? How would you describe it?
Ben Miller:
Okay, let me take a shot. I'm going to start with machine learning for a second, because I think AI is so loaded or in a way you start with where we were a few years ago and ladder our way up to it. AI and machine learning, the best definition I've seen is automated pattern recognition.
Cardiff Garcia:
Okay. And that's machine learning specifically? Which is, by the way you're making the distinction between AI and machine learning is I understood it, machine learning was a subset of AI. It's a way of describing partly what AI does rather than a completely distinct thing in and of itself.
Ben Miller:
Well, there's a debate around that where I would argue with my CTO, and obviously he knows way better than me, is that there are some emergent properties of generative AI that I think might transcend machine learning, but he would say no. We can get into that later. But the point is, is that, let's start with machine learning. Start with basically what is already happening? What do you use as already machine learning? I think you can name a few, but basically the machine learning categories, there's classifying things. For example, you can say this credit card transaction is legit, this credit card transaction is fraud. That's a pattern. Somebody bought 20 things on the internet in less than one minute. That's a pattern. It recognizes that as, for example, labeled as fraud. It's good at predicting things.
It can extrapolate patterns. So that could be weather, could be stock markets or if you talk to a machine learning person will say everything's classification and everything's prediction. So this may not be mutually exclusive in a clean way, but good at recommending things. Movies, songs, TikTok, those are all algorithms that are machine learning, recommendation engines. Computer vision, that's like self-driving cars or optical character recognition where it can read some numbers on your check when you're depositing a check. So those are examples of machine learning. The point is this stuff's been around and it's very powerful and you don't even see it as AI. You don't think like, well, this thing that has read my check is AI. I'm not looking at Netflix and say, oh, Netflix is an AI company. I don't think that's how people think of AI.
Cardiff Garcia:
Targeted personalized advertising, by the way, is similar here. That's an example of machine learning. Something like translation services, automatic translation services also fall into the category of machine learning. I do like the idea of describing AI as a so-called prediction technology. And essentially what that means is that, as you said, it's pattern recognition and then from these enormous data sets that it is trained on, that an AI is exposed to and trained on, it can then try to recognize patterns and then use those patterns to make predictions. And you can see examples of this already in things like you said, movie recommendations, but also in things like medical diagnoses. So it finds patterns in symptoms that suggest a certain diagnosis and then it makes a prediction of it's 78% that you have a certain kind of cancer or something like that.
As you said, movie recommendations. Some people use it for stock market predictions or weather forecasting, but I love this idea that these are examples of machine learning that we have already started to use in many of our day-to-day lives. We just don't recognize it. And now the computational power that's available to people is so enormous that we can see that this is likely to increase, this trend is likely to increase. I like that as a depiction of what AI or specifically machine learning actually does. But yeah, did we leave anything out before we get into how it does it?
Ben Miller:
A bunch.
Cardiff Garcia:
Okay, great. Excellent. I can't wait.
Ben Miller:
We're just getting started. Basically everything you said is true, but there's basically been, I'm going to say it's a break or a breakthrough, a step change, essentially what changed over the last 10 years, but basically over the last 10 years since 2012, laddering all the way up to ChatGPT, there's been a change in the algorithms, a change in the architectures, a change in the compute with Nvidia and the GPUs. And essentially that series of breakthroughs have accumulated to what I think is a tipping point. And what I described, and I think you can loosely call generative AI. And generative AI is different in a lot of ways than the more traditional ML, is still a form of machine learning, ML, machine learning. But what it does is different, a couple really important things.
So one, and I'm going to say this is rough, so engineers bear with me. But what it does is if you give it a lot of data, it can see the patterns and structures in that data and then generate similar new data that's like the data you showed it. So that's generative AI. So let me give you some examples. So if you showed it 10 million or 100 million faces, pictures of faces, it could then generate or make pictures of faces that were original. Or if you showed it all the words on the internet, it could then generate writing. So it's the generative aspect to it that is what makes it so different than what came before.
Cardiff Garcia:
The fact that it can generate something genuinely new from all of the stuff that it is analyzed, but it is generating something that is new, that is original, like you said, a new face. Or in the case for example, when you're interacting with ChatGPT, it can respond to you based on what you say to it, based on what you ask it. It can essentially understand the words that you're feeding it in your question. And then based on the relationship between those words and how those words have interacted and all the huge data sets that it has analyzed in the past, it can then use all that information also to respond to you in a way that is appropriate. And frankly for those of us who've been using it the past year or so, really quite impressive and even a little bit spooky.
I remember when we all first started using, in my case ChatGPT, how impressive it was. And in fact, Ben, as we were discussing what to do for this episode, I actually was asking ChatGPT for some helpful explanations that we can use. And I thought I'd give you just one quick example on generative AI. I basically said in fewer than 150 words, what is a good definition of generative AI specifically? Right away it said, generative AI refers to a subset of artificial intelligence models and techniques designed to create new content or data that's coherent and contextually relevant, often resembling authentic human generated output. These systems can generate various forms of content such as images, text, music, or videos. I'll stop there. And then it gave an example or two similar to the one that you just gave.
I think that it's just such a powerful new thing. And I agree with you, this at least feels like it represents something that's different in kind and not just in degree. It's not just some incremental new advance or the computational power got a little better. It is a step change as you described it.
Ben Miller:
I worked hard on trying to develop the right analogies to help bridge this gap of understanding. That's why I really think the pattern recognition, the essence of it is pattern recognition, and I want to try to dig into that for a minute to say, what does that mean? Even the idea of pattern recognition, I think people don't always know what that means without some examples. So I said, let me just try to go down into this, because how is it that ChatGPT or other generative AI models like doing this and showing it pictures of faces, the patterns and images are easier for people to imagine. You can see what do noses look like? What do eyes look like? What are shapes of head? What's hair? There's clearly patterns on how a face is organized. And basically everything is pattern recognition. Everything we do, everything in the universe, how the human brain works is pattern recognition.
And that's the part where it's not really clear to people that's the case until I try to pull back the curtain. So instead of talking about AI, you should talk about the human brain for a minute because you sure taking that for granted, how is it that your brain is able to recognize faces and a language, what's happening inside the brain? Because you can recognize something if you try to parcel a bit, you can recognize it, you can remember it, you can then recall it, and then you can figure out what it means. Those are in way four different things happening inside your brain. And what's underneath of that is when you're learning something, when you're basically memorizing something, it's actually a pattern that's being put into your brain.
It turns it into a bunch of synaptic connections, like when you learn something, you make a new synapse, a connection between neurons. And things that are related have more connections inside the brain, and when your eyes see those things, lights up that map. That's not just true of a thing like a face, but it's also true of a concept, a concept like algebra or the periodic table, but also, hey, my kid is in a really bad mood. Oh, I see the pattern. They're hungry. All these things are patterns. And I think when people say, oh, GPT was able to map out all the patterns of language, I think that feels obscure. It feels opaque to people. What does that mean? What's happening there?
But once you recognize that language is a reflection of how we think, the patterns of our thinking, even the words fruit is a pattern of five letters and you see that pattern and you then know your brain sees that as a symbol and turns into the word fruit, and then fruit is relationally connected to vegetables, but it's definitionally different and you have these connections and inside of fruit is apples and bananas and oranges. Your brain starts having these groupings or connections and things that are not connected like astrophysics. There's no connection in your brain between astrophysics and fruit, and that connections, relationships, groupings, similarities and matching, that mapping process is super similar metaphorically to what the generative AI or GPT, what the generative pre-trained transformer's doing.
Cardiff Garcia:
This is such a powerful point because a lot of times when people try to define AI as we've just done, we almost talk about it as if it's something completely new or completely original, but actually in some ways the human brain is itself a prediction technology. It is a pattern recognition technology. And here I want to be careful because we can almost take this too far and act as if there are no remaining mysteries about the human mind, which is not what we're trying to say here, but this is a very important part of how the human brain, the human mind actually works, how it memorizes things, how it recognizes things, how we learn to interact with each other.
There is a very heavy component of obviously pattern recognition of prediction technology, and so in some ways what AI is doing with a lot more computational power behind it, is in fact trying to mirror or reflect or imitate or whatever word you want to use, something that already has happened in the human mind. It's not entirely clear I think, how that would relate to other elements of being human. Creativity, persuasion-
Ben Miller:
Emotions.
Cardiff Garcia:
Emotions, persistence and other things that come with being human. By no means does it capture everything. But that is a very, very important part of what it is to be human. And now more and more it looks like we're mimicking it awfully, awfully well using AI. And that's why by the way, I used that word spooky when I at least first started interacting with ChatGPT, because nothing like that had really existed before. It just didn't feel the same. It felt like a big step change. And so I feel what we're doing now and what you've just done in laying out the way that AI works, is trying to explain just why we got that spooky feeling and why it is that this is something different and not just what we've had in the last 10 years of machine learning, but a little bit better. It is something of a different category and not just a marginal improvement.
Ben Miller:
There's one other thing that's really important about it that is not obvious unless you're closer to the tech, is that before those machine learning models that we talked about for finance or for recommending movies and stuff, they really needed structured data. They had to be taught very literally with labeled data say or supervised machine learning, they had to basically be shown what the patterns were in order to then recognize the patterns in the wild, in actual application. And the difference with generative AI or GPT, is that you're able to give it unlabeled data or unstructured data then can intuit the patterns, and that's huge. In some ways that's just as big as everything else we've talked about, because the fact that it can replicate the patterns without being told what the patterns are, the implications for that are absolutely insane, and that's why I think it's more like AI than ML.
Cardiff Garcia:
It seems to me, based on what you just said, that there really is a difference between, for example, the prediction technology stuff, the decisions that AI is going to make, and then the judgment in terms of how it is applied, which I think that judgment part will remain in the realm of the human. In other words, the people who actually know how to use the AI best are going to have an advantage, and it's also going to be necessary. The understanding of the values that you apply to whatever the AI is doing, the judgment that you apply at the end of it, I think is going to still matter. It seems to me like we're getting at that distinction between what will AI be amazing at and really rapidly help improve our lives versus the things that people are still going to have to do and perhaps want to do. It might become more interesting.
Ben Miller:
It's not as clear to me yet on that point. There's some that's really clear. This has always been true of software, so AI just makes it more true. A lot of the work you do is repetitive and before software was good at doing repetitive tasks that were literally repetitive but not sort of repetitive.
Cardiff Garcia:
Routine.
Ben Miller:
Yeah, you're an accountant, look at a spreadsheet, move it into a trial balance. It's not literally the same because there's different labels and sometimes the decimal point is in the wrong place, whatever it might be. And so things that are loosely repetitive, software actually couldn't bridge that gap and what GPT this new model can do or architecture, it can be a lot looser, so it can take more of the routine away. In an organization the amount of stuff that's actually routine is enormous. So that's I think really clear. But the opportunity as a builder or an investor, it's like trying to build out that application that can then take the routine work away, whether it's accounting or finance or looking at cancer screening. There's a lot of work in getting something that's right on a lot of routines for a lot of people, and that's I think, the main clear piece of work that generative AI is going to do over the next five or so years.
That's the clear stuff. And exactly where it will top out. I think that's much less clear. The one thing I think it is helpful is to know that there's a pattern to technology. One of the patterns is new things show up. They have this exponential curve and then they start to slow down. There's an S curve. So right now, GPT-4 is better than GPT-3 and 10 times bigger. But we'll start seeing diminishing returns at some point. Usually it's almost 80/20 Pareto principle, the first 80% of the gains in AI will happen rapidly, and the last 20% will take 20 years. And we don't know where that threshold is. That's the thing that's open at the moment.
Cardiff Garcia:
It's worth noting that technology often works in fits and starts like this. And the analogy that a lot of technologists use and people who study AI have used, is to the emergence of electricity in the late 1800s. Well, it took a long time and the introduction of new systems before, for example, that led to a radically new way of making cars in the factory. It took decades in fact, even though the fundamental technology was developed much earlier. Why? Because when we first get these new technologies, we all start trying to essentially apply them to existing ways of doing things, and it takes some time, it takes a lot of risk taking, entrepreneurship, some first movers, to really start coming up with a totally brand new way of using a transformational technology.
And that can just take time and it can be slowed down by a lot of things, not just the natural risk aversion of people who might want to use it, but nobody else is using it. So then it just seems like it's too much because it's so much downside, and what if it doesn't work? Then you look crazy. There can be deliberate attempts to slow things down sometimes if you think that technology might be dangerous, there might be a regulatory approach to try to slow it down or at least to try to understand it as we go along. That can slow it down. There's going to be natural resistance from a lot of people. There's a lot of doom and gloom type stuff around AI. A lot of people thinking we're headed for a dystopian wasteland because the AIs are going to take over all the machines.
The real Terminator 2 thing, that is something that people actually debate. I will be candid with you, I find all that to be quite outlandish, but that could just naturally slow it down versus a kind of approach where you experiment and you keep going on and on and on. I don't know exactly what the right way of handling it is. I'm just pointing out that that can slow down the transition from the technology exists to it is now changing our lives. That can take many, many years, sometimes decades.
Ben Miller:
As an entrepreneur who's tried to do this in a small way and also done it internally with organizational change, it's much harder to change people's routine than it is to actually conceive of the inventions. Incredible. So changing people and changing people's mind actually is really the hard part. I think Max Planck, the famous physicist, said that progress is made funeral to funeral.
Cardiff Garcia:
Good line.
Ben Miller:
It might literally be that the Gen Z generation has to become 60 years old before everybody who thought the way they had to do work was willing to basically change. They can't change. I think that's pessimistic, but there's something in that that's unfortunately true. I'll give you some more examples of things we're doing here. There's the low hanging fruit. So example of that is customer service. We get 10,000 customer service outreach a month.
Cardiff Garcia:
A lot of people.
Ben Miller:
We implemented, we have it in beta right now, an AI application, and it's able to resolve 71% of the outreach. 71%
Cardiff Garcia:
Just using the introduction of this new AI?
Ben Miller:
Does it automatically.
Cardiff Garcia:
That's a lot. That's incredible. That's what? 7,000 tickets or whatever doing some simple math?
Ben Miller:
Yeah, well, a lot of these are by phone. So all the ones that are written, it can do 71%. So in order to do the phone, we'd have to create some software that does it for the voice, and we don't have that yet. So what it's actually telling me, which maybe not obvious, is not that there's 71% of the work that's being done, but it's actually 71% of the outreach is really trivial. It's routine. That's what it's actually saying, even though that's not obvious on first blush. I lost my password, changed my password. I need this really simple question answered. So 71% of the team's work is actually not complicated, but takes a lot of time and it took that away right away and it left behind the high value work. It's a good example of the best of both worlds.
Cardiff Garcia:
I'll be honest, I will even give the technology even more credit than that, because we've been describing routine versus non-routine. To me, what's interesting about AI is its potential to start doing some tasks that are not really routine. And even in terms of customer service, it wasn't categorized as routine in the past to take a call, manage the emotions of the person who might be upset. They need something specific. You need somebody who knows how to field their complaint or field their request or inquiry, find it, get it back to them and so forth. It might be routine now that a machine can do it, but I wouldn't always categorize that as routine work. It's just that now AI is so powerful. The technology is powerful enough that it can actually do that.
The fact that you can ask a ChatGPT to help you craft an email or help you with the grammar of something or ask it anything, that was not routine in the past, that was something that would've required doing a fair amount of work. Now it might become routine. That's what to me is so impressive about it. So even in this small example of just customer service, which is obviously not the core Fundrise thing, it's an important part of what you do, but it's not the main thing. It seems like a pretty big deal to me that you're able to solve so many of the requests using this technology.
Ben Miller:
In the same way when you said, how does AI or ML work? Let's talk about how the human brain works. People actually are not that clear about the work that they're doing. So people say, how is AI going to make this decision? How's AI going to make the investment decision? I've spent and the team has spent a long time understanding data and data management industry has gotten really thoughtful and structured and what it means, because I think people have a fuzzy idea of what it means and what work they're doing. I'll just give you two parallels. A human versus machine. So if you're basically trying to make a decision about investing or any kind of decision, the first thing you're going to do is actually gather information. I saw this in a scientific journal. Knowledge workers spend 38% of their time searching for information.
Cardiff Garcia:
Imagine getting back almost 40% of your time. That's what you do.
Ben Miller:
This is loosely what people do, and I'll try to give you some of the more structured way that data engineers think about it. But you search for information, you compile it, you'll filter it, you'll organize it, maybe organize it in spreadsheets so it's labeled properly. In the engineering world, they say clean it and join it. And so you have the information where it needs to be, and then you start doing math. You add things together. And when you're doing this math, you may call it aggregations or derivations in the data world, but basically you're trying to get to a point where you're saying, okay, this is more expensive than that. Is this a higher return than that? All of that work to get to the place where you have the information in front of you, clean with the math done to it, that is like 80% of the work.
If I gave you the spreadsheet with all of it done for me, I can say, well, this is a good deal. The decision's actually the smallest amount of the work. It's actually getting the information to a place where the decision is manifest. We can talk about the decision as a piece of work, but then there's a separate piece of work, then actually reporting on that, then distributing that back out to the world, and that's the engineering world. They would say there's BI layers, business intelligence layers or APIs or things that are pushing information out. But when you say AI is going to do all this work, there's just a lot of work in there that people actually are not going to be upset about losing. And then there are some decisions you said machine learning or AI can do, one day mirror us, mimic us, but clearly they're going to do better. Large computationally intensive math, way better.
Cardiff Garcia:
Absolutely. Yeah, mimic the process, but with this enormous computational power behind it, especially in the analytical realm. And I think it's also going to be one of those things that makes a lot of workers just way more productive. If you don't need an entire team of people to do that analysis, you might want to employ a separate team of people instead, to decide what to do with that analysis or how to make it even better. But a lot of the stuff that right now feels like a barrier to doing something will be gone. And you certainly see that because a lot of people who aren't great at coding for example, can now get a lot of help with that. So if you're somebody who, for example, I'll speak for myself here, would love to be able to make some really beautiful, aesthetically pleasing and really sharply designed and thought out economic charts based on economic data.
Well, right now you have to know where to find the database, where all the data is. You have to have some skills in terms of data analysis and data visualization, and a lot of it has to be done manually in Excel or if you're coding in something like R or Stata or something like that or Python, well pretty soon you might just be able to tell an AI like, hey, I want a chart with these variables. Here's the data set that you can pull from it automatically. Here's what I want the chart to look like and bang, it does it automatically, something that would've taken me like half a day to figure out manually. That's just going to make me as somebody who's analyzing the economy and likes to communicate my analysis way, way, way sharper.
But it does mean of course on the flip side that some of the jobs that in the past were quite valuable, some of those are going away, but some other jobs become more valuable and there'll be the introduction of new jobs and people who in the past could not have done that can now do that. And so there'll be a greater emphasis on a new task. So anyways, I just went off on a little bit of a tangent because that's like a personal example, but I think it's a quite powerful example of what you're talking about. And also Fundrise itself of course does its own data and real estate and market analysis. So I can imagine ways that it might be helping you in that regard as well.
Ben Miller:
You teed me up because you said some things that are patterns of technology. It takes away the routine stuff like a washing machine and a dishwasher does that too. This pattern of basically technology can take things away that are work and basically can do things with a higher and higher quality. And the other thing is that it lowers the cost of it. Not everybody used to have a car and everybody used to have a cell phone. Even machine learning, Fortune 100 companies have been using machine learning for 10, maybe 20 years. The arc here is that it democratizes access. So you just said, I can now do very high quality design through AI enabled software. I don't need to have a full-time designer.
Cardiff Garcia:
When you say design, you mean like real estate design, designing a home, designing a neighborhood?
Ben Miller:
Designing for, or I'm designing presentations. I'm designing websites. I can design all sorts of things. Design team is 11 people here. There's 11 full-time designers and they're full up and half the time I have to design myself. There's a running joke internally about that. So this democratization, technology democratizing access, you go back hundreds of years, the gun, the cost of that provisioning in armies became so cheap that everybody ended up with guns and it basically caused an overthrow of governments. Basically the whole Napoleonic revolution. So this arc of democratizing was something that was elite people could afford or elite companies, that's underneath of all of this too. And it's not just AI in the narrow sense, it's all this machine learning that I cannot bring to bear in my organization, which before I could not, is sort of part of this change. You said use R or something like that, but if you could add some ML for your economic analysis-
Cardiff Garcia:
It'd be way better. You just ask it in plain language. I want a chart to look like this. This is what I want on the Y axis, this is what I want on the X axis, here are the variables. Here's the database. Boom, go.
Ben Miller:
No, no. I'm saying even more advanced like, hey, tell me the correlations between these data sets and what's causal. Hey, right now we're having these arguments about is there going to be a recession or not, load up these 20 different measures and give me your probability and how do you think about it?
Cardiff Garcia:
Absolutely, it can get better and better and better. It's a real advancement in terms of somebody's ability to do something that in the past they wouldn't have been able to because they didn't have the right training. And I want to make the point also that this can apply to people who don't have great jobs right now or a little bit further down the income spectrum, because it could be that a really fundamental skill that they don't have is what is stopping an otherwise very talented person from advancing. Or even in the case, let's say of Fundrise where you might have somebody who knows real estate really well, has a really strong analytical mind, who is very charismatic, could be a great manager and so forth, but for whatever reason isn't a great communicator, isn't a great writer of emails even.
Well suddenly if that task is something that can be taken over by machines, well it's okay that they're not a great communicator, because they have help now, they've upskilled basically, and now they're more recognizable for the talented person they are instead of having this one thing that was holding them back. You can imagine something like that happening for other parts of the labor market.
Ben Miller:
The way technology plays out, there's a pattern to it. Let me explain that. So if you go back, invent computing, which would become mainframes, that's a wave technology. And then after mainframes come personal computers and then the internet and then mobile and then cloud, and now AI. If you go back to those waves, they actually play out in a similar way. Now we're in the middle of one and it seems like it's novel, but actually, hey, this has happened before. This is a broadly understood pattern in the tech world that not as well known outside. The inventions start usually in deep science, in the hardware, really inventing better and better compute, better and better algorithms, which are like the math underneath of all this. And those things unlock new potential.
Obviously the integrated circuit, the chip unlock new potential and that deep tech or hardware typically then moves its way up. The tech stack, this is still mostly invisible to people who are not engineers, but in the tech stack you have things like operating systems and databases. People have described different ways, but basically it's like where the engineers spend all their time building things. And on top of that are applications. Here's an example. So things went to mobile, if a mobile hardware, the iPhone or the Android, and inside there are like Qualcomm chips that basically let you to send data over a spectrum. That's hardware. Then you have the operating system, which is Android. Then you have these applications like Uber or Foursquare, different applications that unlock the whole idea of mobile.
So what's happened in AI is that Nvidia is the primary hardware driver here, it's the chips that allow for parallel computing. Then they have these platform technologies that are now breaking out, which is the GPT, the generative pre-trained transformers and things basically that most people don't need to know anything about. And the big opportunity next is the applications. And there actually aren't that many yet. We've invested in a couple. One that we invested in called Canva. They democratize design software. That's why I keep talking about design, because it's on my mind. So you could go on there and use it to design world-class designs.
Cardiff Garcia:
I actually think I used it once for a project cover letter or something like that because it makes that really easy. It has beautiful templates, it's very customizable and all that stuff. I know what you're talking about now.
Ben Miller:
And so Canva is application focused on design. They plugged into the platform, which is OpenAI has a platform you can plug into. We plug into it as an API. Uses GPTs, the generative pre-trained transformers that uses technology basically to do things, they want to do things with art and design and visual media. So that's an application that's likely to have explosive growth because everybody who's doing any kind of communication, design is something that's very fundamental and the applications of AI to design is a huge business or the application of AI, for us real estate, we're working on that. Somebody's going to build an application to do accounting for sure.
You go down the list and those applications are the things that take away the routine work, but you end up needing to do a lot of specialized work as a company. For us to do real estate, it's going to take us a lot of work to take this broad foundational models that exist out there into specialized models with specialized data and specialized context to get it to do what people expect it to do when they're using it.
Cardiff Garcia:
In real estate specifically, I'd actually love to hear more about that. How would you think about using AI to determine, for example, I don't know, where your next investment might be or what part of the country you're now interested in investing and other decisions like that, investing decisions for Fundrise, how would you think about applying AI to those decisions? Or maybe you don't want to give away the story just yet.
Ben Miller:
I'm not going to give it away, but the product work is where I spend my time and that is where technology meets the user. I have a lot of work on that that's happening in real time. I'd rather share it by launching it. But another area to invest in that's related to AI is, people would call it the picks and shovels strategy. There's an AI gold brush and you can try to invest in and be the winner of an application layer like Uber is for transportation or you can pre the technology underneath of that application. Like Twilio is the technology that allows voice in app. A lot of companies use it. It's a very big company. There's a lot of picks and shovels technologies that are platform underneath these things that are enabling software, that will become very, very valuable and critical. If you go back to the PC world, there was Intel, which is the hardware.
There was Microsoft's operating system and then there were applications that sat on top of it like Adobe or your browser. There's this middle layer, this platform layer that is invisible to the consumer largely, but it enables the apps. If I'm an application, I'm building something for real estate or something for design or I have all this data, there's common tools you need and those tools are common across different companies. Those are the common elements underneath of almost all of this. And there are certain companies, there's a company called Databricks, which we invested in. They're like a core infrastructure company that is very common to do this type of work. It's one layer below what people are familiar with, but it's different and very good investment strategy. So you can invest in the direct applications or you can invest in the software that enables these applications.
Cardiff Garcia:
Excellent. Well, Ben, at this point we've now talked about what AI does. We've talked in some detail about how it works. We've talked about its applications widely and in terms of Fundrise specifically. Should we close with a discussion of what the potential big picture consequences of AI could be for the economy or even for society more generally?
Ben Miller:
Yes. It's like the industrial revolution. I almost think of it as the personal computer was the steam engine and AI is electricity. We're still early in the revolution that's happening. And then there's some things that are for sure that people don't know about and some things that are really speculative. Let me just give a couple that are for sure that are mind-blowing. We've talked about a lot of applications. We haven't talked about one that I think is, for me, the most incredible. And that's the implications to health. As we said, the models or the architectures can get a bunch of data and find the patterns in that data. So the thing that it can do is it can learn the grammar of molecules, the language of cellular biology.
We could talk to our own cells, so there's obvious things like gene and protein sequencing and drug development and cancer screening. But the consequences to being able to talk to our DNA and talk to our cells and talk to our human bodies and talk to all biology and program it, I can't even fathom the implications of that. But they are so big.
Cardiff Garcia:
I agree. There are some wonderful possibilities when it comes to human health, when it comes to medicine, when it comes to finding ways for people just to live possibly longer, but hopefully also better lives, healthier lives. And this is all related to my biggest hope for AI, which is different from saying that I think it's necessarily going to happen, but my biggest hope for it is that it doesn't just become what economists know as a general purpose technology, which is to say that it has applications all across the economy in every different sector there's some way to apply AI so that people, workers and so forth can be more productive and make better more various things for others to buy as well and to improve our lives.
But that it also becomes what's known as an invention in the method of invention, which is that AI ends up transforming not just every industry across the economy, but also the way that we think about innovation itself, because it is so powerful and because it can unearth previously disguised patterns and things that we would love to be able to analyze and just didn't have the computational power to do until now, that would be truly magnificent and it would have hopefully a self-sustaining quality when it comes to making the economy better, more productive. So that is my positive hope. There are some potentially negative consequences that we should discuss.
But before we get to those, let me throw it back to you, Ben, in terms of hoped for outcomes, the truly wonderful, miraculous potential of this technology, what else besides health and different ways of doing innovation, do you think we should emphasize?
Ben Miller:
This is again, a very hopeful or speculative one.
Cardiff Garcia:
Blue sky thinking right now, pie in the sky stuff, whatever cliche you want, that's where we are. That's the realm we're in now.
Ben Miller:
So the internet, it had an unintended consequence of fragmenting society actually, in a way that now you look at what does the internet do? And it basically creates more connections as opposed to having four TV channels, everybody gets their information from, it's information from everybody else. So they cause social fragmentation, which is why some people are critical of social media. And I think that AI could actually have the opposite. It could actually drive more cohesion and consensus because there's going to be less debate about facts, for example, in data, there's only going to be a few AI models, foundational AI models because they're so expensive. Maybe five, maybe less, maybe three.
There'll be Chinese ones, but we're not going to trust those ones. There's not going to be that many. And so I think that because there won't be many, an unintended consequence of it is more agreement because there'll be so few AI platforms, it's going to inherently drive more unity.
Cardiff Garcia:
It is definitely something to hope for. In terms of potential negative consequences, there's one very obvious one, which is there is a chance at least that at first the people who really gain from AI are whoever owns the companies that make the AIs. And that could, especially if AI does truly prove to be transformational, lead to a massive increase in wealth or income inequality. And that is something that would be exacerbated if AI does lead to widespread job loss and that the future jobs that would replace those jobs, if those don't emerge right away, then you're going to have a wide swath of society that is either unemployable or whose wages are going to be severely depressed. I want to be clear about something. I'm normally somebody who worries that too little dynamism is a much bigger problem than too much dynamism.
But it is the case that in the past when there has been a lot of dynamism, there have been segments of society that have lost out. This by the way happened at the very beginning of the industrial revolution when you had the Luddites and you had people attacking machines and they kind of had a point. Not that new jobs wouldn't eventually emerge and that the gains would be widespread. Eventually that was true, but it took generations for that to happen, because at first the gains, the surplus from the new technology was taken basically by the capital owners who didn't spread it out. And you needed broader societal changes, new institutions to make sure that everybody was able to prosper from the new technology rather than just one tiny little group of elites.
And so that is just something to pay attention to if AI does prove to be this sudden faster than we think rapidly transformational thing with the power to improve so many lives. But if we leave behind too many people who lost their jobs for too long, then in some sense the system might end up eating itself because there will be this big constituency that fights back.
Ben Miller:
On a very personal level, Fundrise launched innovation fund to democratize investing into these companies because they're not public. OpenAI is not a public company. Databricks was not a public company. Canva is not a public company. I think that everybody needs to be able to invest in these companies. And then just to get a little bit into the weeds, we are not taking a 20% carried interest. We're not charging this very massive toll, 20% toll to enable it. I think that's one of the reasons why we started our tech fund. That's a practical thing I'm trying to do. And then on the hopeful, again, I think we didn't mention this, but AI's ability to transform education and actually be part of the solution of re-skilling and helping people learn what they need to learn, I think is going to be part of the solution.
Cardiff Garcia:
And to tailor it to students specific needs, there's real promise there.
Ben Miller:
And even to change the way we think about education, you think about how my kids are taught and how I'm learning machine learning, how I go learn things, so different. We know that there's a few areas like education, healthcare, government that have been huge sources of ballooning costs in our society and we don't know how to change that. And I think AI is likely to change all three of those things considerably. And there hasn't been a chance of changing government in my lifetime. And now you're talking about what is government with AI? I think it could be something radically different that's probably going to take Max Planck's generational change.
Cardiff Garcia:
One final thought or maybe one final thought from each of us here. We started the conversation discussing how AI might change our understanding of what it means to be human, and in particular if AI with its powerful pattern recognition ends up doing a lot of the things that we once thought of as a uniquely human thing to do, which was the ability to analyze things, the ability to see patterns out there in the world. Does that mean that in some sense we as people have been diminished? I would say I don't think so. I would say that we won't be any more diminished than we were when physical machines, physical robots ended up taking over some of the things that we thought we had to do forever. We were plowing things manually for I don't know how many thousands of years before we found the ability to automate that away and then what it means to be human ended up changing.
So maybe that purely analytical part of what we do will be less necessary in the workplace, but maybe our personalities, our emotions, our ability to persuade, our persistence, as you noted some of our other emotional range, all of that might change. And I am very curious to see just how. Because AI more than maybe any other technology of our lifetime at least, you and I aren't super old, but we're not super young anymore either, I'm sorry to say, my friend, I think has the ability to radically change how we understand ourselves and not just how we understand our place in our professions, in our careers and in the world at large.
Ben Miller:
Every generation or every 50, 100 years, we're challenged by something. It was nuclear war was a huge risk in the 50s, all the way through to the 90s. And so we are entering a new era and then we can respond with fear or respond with hope. That's the personal challenge for all of us, and I'm hoping we rise to the occasion.
Cardiff Garcia:
And I think that is a good optimistic place to end the chat. Ben, this was super fun. This was a blast.
Ben Miller:
Yeah, Cardiff. It's a pleasure.
Cardiff Garcia:
You've been listening to Onward, the Fundrise podcast featuring Ben Miller, CEO of Fundrise. My name is Cardiff Garcia of Bazaar Audio. We invite you again to please send your comments and questions to onward@fundrise.com. And if you like what you heard, rate and review us on Apple Podcasts and be sure to follow us wherever you listen to podcasts. And in particular, I think Ben and I would emphasize that for this episode we'd really love to hear your thoughts, because we really are wading into murky territory here, not just because we're talking so much about the future, but because the things that we've just discussed were so dense with information, with complication and with other things that are hard to understand. So we'd really love to hear your thoughts. So please send us an email.
Finally, for more information on Fundrise sponsored investment products, including relevant legal disclaimers, please check out our show notes. This podcast was produced by The Podcast Consultant. Thanks so much for listening, and we'll see you next episode.
