This transcript was created with help from Ai. Please report errors to us and always refer to the video as your primary source.

Justin Brady: Come with me to a magical place where being labeled the pork queen is a compliment. Where being nice is just a way of life and where all leaders of the free world must pass through. Welcome to Iowa. Whether you’re a current Iowan, an Iowa expat, or an Iowan at heart, this show is for you. This is the Iowa Podcast, real Iowans, real talk. No. I’m Justin Brady from the Brady Dental Care Studio and joining me today is Joseph Nelson, the CEO of Roboflow here in Iowa to discuss all the new AI regulations that were just signed by the president. Uh, Joseph, thank you so much for being on the show. I really appreciate it. Thanks for having me. So a lot, a lot to get into and it obviously just happened. So this is a, I’m sure a lot more will come out soon, but a Roboflow, a really quick disclosure. You guys were a client a while back, so obviously I’m not a client now, but I want to get that in there. Um, yeah. So for those that love the Iowa podcast, every week we’re bringing you really incredible information. Uh, we’re bringing you great Iowa guests and so make sure to subscribe. Would you? Spotify, Apple Podcast, iHeart, it’s even on Amazon. So make sure to subscribe. All right, Joseph. So the first thing is obviously, um, Roboflow is pretty reputable company. You guys have gotten some investments from some like, I would just call them celebrity venture capitalists and in the tech world, you guys are a big deal. So let’s put, let’s just, you know, for people who don’t know you, let’s just get that out there. Obviously, President Biden has signed an executive order regulating AI and AI companies. And so I guess the big question is, uh, are you and your team concerned about it at all?

Joseph Nelson: Well, in general, the executive order was just signed yesterday, right? And the full 110 page document has just been released. And so pouring over the fact sheet and get in a sense of what’s included, what’s not included, um, just as early for me as it is, I think, uh, everyone that said, I think the intentions, uh, are largely, largely good. I think the administration’s aiming to make the United States be a leader in investing in making AI be a core competency and capturing the upside of it while minimizing the downside. Now, in terms of the ways that I think. Regulatory has a history, or maybe lack thereof of encouraging and enabling successful adoption of technology. I do have some skepticism and some and some concerns, especially as a smaller company. Um, regulation typically favors those that are incumbents, but I’m sure we can get into that overall. I think, I think, um, it’s well intended, but it’s also quite broad and it’s, you know, it’s an executive order. So it’s kind of to be determined how much of this can truly be implemented and enforced. And what exactly the specifics will be? I mean, when, when it was like, this is sort of past, they’re deliberately broad and generic, I think, to provide the opportunity for, um, individual agencies to. Enforce and determine the best way to enforce rather than extremely clear. And that’s kind of the stage that we’re at now is. There’s broad based terms, goals, and objectives, though not a ton yet of specifics of how those objectives will be borne out.

Justin Brady: So you’re hoping that this shapes AI regulation and that this isn’t like the final, uh, this isn’t the final say in it. You’re hoping that it changes.

Joseph Nelson: Well, I’m hoping that this… Invokes and instills and doubles down on investment on AI research in the U. S. I mean, a lot of parts of the executive order are about talent and investment in, um, AI’s upside potential for, for labor, for climate, a number of areas. There are also areas that the executive order calls for overwatch and validation. I would call it of training of models. And I think some of those areas, while perhaps well intended, may not accomplish their goals. But I think, you know, taking the good with the bad, I think there are some areas here where the, the executive order is aiming to make it such that the U. S. is the place to do artificial intelligence. Right.

Justin Brady: Um, uh, so, well, there was a really interesting tweet. One of my, one of my friends is in healthcare, and he’s an AI in healthcare, and he posted that Director Iradi, Prabhakar and I could be getting that name wrong, but, uh, he posted her comments and essentially, let’s see, what was the, what was the quote here? Uh, okay. So basically what his quote, this is his quote. Uh, so he’s in medical health care, uh, and he says that. He, and he’s commenting on her comments. Okay. So he says, comparing Gutenberg’s printing press to AI and in the same breath, celebrate celebrating the regulation of it is just like what the European monarchs did in 1468. So to your point, he’s saying, you know, she’s, she’s talking about how Gutenberg was this great technology and it should be celebrated and AI is a great technology, but it could be used for bad, which is why we’re regulating it. Like this is essentially his argument is. This was the same argument leveled at the printing press why we should burn books and regulate writings. Do you see the same comparison, or is it not quite a fair comparison? I think it’s far reaching,

Joseph Nelson: um, to make that comparison so far. I mean, I think skepticism is warranted. Like, there’s pretty high level, flowery language that… Would be tough for any 1 person to say, oh, yeah, that’s something that we shouldn’t do. For example, like, oh, we should capture the upside benefits. We should protect against risks. We should better protect privacy. We should evaluate how there’s clear and fair standards. We should eliminate algorithmic bias. We should enable AI to be used responsibly. These are all noble goals. And objectives that I think it’s pretty difficult for anyone to disagree with, right? It’s, it’s kind of like if the executive order, um, pretend we’re not talking about AI, pretend the executive passes an executive order. That’s like, we should foster economic growth. Very few folks going to be like, oh, yeah, that’s not something that we should, we should invest in and have happening. The devil’s going to be the details of how, how 1 thinks we should go about fostering that growth, right? Like, if you’re generally a democratic. Party economic growth is going to look different than if you’re generally a Republican Party. And I think the same is true here for the executive order as it relates to AI. It’s like, yes, it’s very difficult to say no, I don’t to disagree or or not support advancing responsible use of AI. But if you said. What the way that manifests, um, and this is this is TBD, which is an example, one way you could argue that responsible use manifests is you could say the only way that artificial intelligence could be used in your product is if you pass a series of lengthy tests and those tests could only be born. The costs of enduring those tests could only be born out by Companies that are, uh, much larger scale, like, for the FDA and healthcare is known to be one place where there, there should be a strong regulatory regimen for the purposes of protecting, uh, citizen health, right? Like, you and I both want to ensure the medicine that we take are safe, tested and validated. That said, I think if you talked with a number of healthcare experts, the multi phase, multi year waiting period does stifle many good drugs making their way to market more quickly. And, um, you know, there’s, there’s, there’s some balance between the regulatory regime that exists presently for bringing drugs to market and one that might be most optimal for enabling innovative drugs to be brought to market. And I think that for AI, you could see a similar standpoint, right? Like think about it from the perspective of a young organization and a young startup in the context of healthcare. If you don’t have five, sometimes 10 years of runway, like Moderna was a publicly traded company before they released their first product to consumers. Right. And if you’re not, if you’re not a well funded company and you have that sort of regulatory burden before you’re enabled to bring your product to market, that actually may not advance. The use of responsible use of AI, it actually might stifle our ability to know how I should be used responsibly at all. Now, I’m not suggesting and making an equivalence of saying this EO calls for an equivalent amount of regulation of bringing health care to bringing drugs to market. I’m more just trying to draw the parallel that when we say we want responsible use of AI, one way that could be construed is limiting innovation and understanding of incredibly powerful technology. Thank you. And because I, you know, we’re in the very early innings in a very exciting way of all the ways that artificial intelligence is going to improve what it means to be, you know, frankly, even like human existence, basically, there’s no part of there’s no part of society or humanities history. We’re adding intelligence to a system reduces the quality of that system is adding intelligence to a system and decreasing the cost of intelligence. So whether that’s, you know, finding climate change, improving regulatory, finding new drugs, enabling us to, like, move goods more efficiently, like, all of these areas are going to be radically improved with with AI. And I think the more that we introduce potential unintentional burden to finding and adopting and revolutionizing our economy with all the ways AI will be impactful, you know, the, the, actually, the more we’re living in the upside of the potential. So that’s the thing I worry about with an executive order like this is well intended and broad reaching language with unintentional. Um, implementation right that that undermines the results it’s aiming to foster. Yeah. And you’re

Justin Brady: not alone in that concern. A lot of people are wondering if this is a backdoor to regulatory capture and a lot of, you know, bigger companies are saying, uh, because this isn’t, you know, you’re not the first to bring this up, right? Which is that, uh, large companies can get through the FDA process. Small companies and small innovators that may very well have the next gen solution cannot get through the FDA. And so you can make the safety argument, right? We don’t want crazy street drugs treating everyone for certain diseases. Obviously nobody wants that. But at the same time, even like yourself, you guys are a very small startup and you were disrupting a lot of what bigger companies could do. And you know, I could see why a big company would want to go to their congressman and say, Hey, Hey, These little guys aren’t regulated. They’re not very safe. You should have extremely long and expensive processes to get, you know, to, to, to slow down that runway for these small guys. Um, so to your point in this executive order, the, the department of health and human services has been tasked to create a program to evaluate potentially harmful AI related healthcare practices. So what you said earlier, you know, like who’s, who’s actually creating intentionally harmful AI related healthcare practices and who, who defines what harmful really is?

Joseph Nelson: Not only that, but I think who determines what falls into scope and out of scope for a given use case. And I think that, um, there’s a strong history of Introducing regulatory, um, regulators ever increasing the things that they’re able to regulate in a sense, like, if you’re, if you have the opportunity to oversee something, then, of course, you’re going to view it as. Your your right and responsibility to oversee those things. And so I think that, um, you know, for example, let’s talk about health care. Like, if we’re talking about drug discovery, it’s like, yes, that’s very probably clearly up the fairway. But like, you know, maybe if we’re talking about health care and we’re saying it’s in the context of we have we have users that will use our products for example, for doing, um, content moderation, uh, invalidating or invalidating whether there’s a presence of We’ll say, like, in some content, like drugs or alcohol or things like this, um, or maybe you make a claim that this is something similar to, um, you can make a claim that that’s similar to mental health care or consumption of products that are not healthy and all these sorts of things. I mean, I guess, I guess my point is I have a healthy sense of skepticism of. What’s in the air bounds and what isn’t, and by the way, this isn’t just like a, a self interested, like, oh, okay. Like you’re, you’re 1 company. Of course you want to enable, uh, AI to, to, to be used. It’s much bigger than, than like small company, big company. This is like the way innovation happens. There’s just so much to explore with artificial intelligence and machine learning that, you know, it’s the most powerful technology Of our lifetimes, and I think that’s what, uh, society is coming to, to, to realize and agree now with that in mind, like, you know, Andrew is famous for saying is the new electricity. Imagine if only a subset of companies of the 4 biggest companies of the era are able to say, and experiment and see. You can use electricity for the amount of ways that we would have limited consumer electronic growth. And exploration and it’s just simply like, you know, you can make it, you can make it as simple as the following. There are more people outside of, you know, the big 3 tech companies than there are inside of the big 3 tech companies. No longer no matter how big they are. Right? Correct. There’s more people outside than inside. And given that there’s, there’s more innovation and responsible development and exploration that can be taken place That fosters innovation and fosters things like, um, cheaper drug discovery, better, better ways of solving cancer. Like if you’re, if you’re a startup, that’s trying to spin out of a university lab, cause you think you found an innovative way to, um, tackle a form of novel cancer, you know, in a world where you need to abide by a series and set of regulations, maybe you don’t do that, right? That future doesn’t exist. Right. So I think, um, what I’m suggesting is that. Um, we’re very early in a positive way, and we should be looking at that with optimism for all the ways that we know AI and ML is going to improve and shape the way that we live. Yeah.

Justin Brady: And like to your point, one of my clients, because, uh, people in the podcast, uh, this is my Iowa podcast. But you know, people know that I do other stuff, right? This isn’t my full time job. So to your point, there’s another client I have that had a amazing medical device, FDA cleared, really cool device, very small company in Iowa. I mean, I might actually have them on in the future. Um, but I, they had this new device sitting. It’s non invasive, non drug, and they had a new device sitting on this table. And I was like, man, that’s really cool. Why don’t you guys push that out to the market? Like, let’s, let’s do it. And he’s like, you know, Justin, we’d love to, but we don’t have like the few million to clear that through the FDA. I’m like, are you insane? And he was like, yeah, all the testing we need to do because that’s understood. We should do that. You should do safety testing, but we have to do all this reporting. We have to do the regulatory process. We have to wait a long time, sometimes years, and we just don’t have the runway to do that. So there was this amazing life saving technology sitting on a shelf. They were just not going to push out to commercialization because they didn’t have the capital to do it. And I’m like, how many, um, amazing tech innovators are out there that could solve a disease or solve a big problem, but they can’t because in some cases, I’m not saying all, but in some cases the regulation is prohibiting them from doing so. So like, I want to get, and obviously you’re not a lawyer, but um, Like, what, what exactly are regulations here intending to do? Because, um, if I build, if I build a new product and I 3d print some new widget that no one has ever seen before, and I go smack someone over the head with it and they die, no one’s going to come and say, well, there are no regulations around that widget he made. So therefore we don’t really know if he killed that person. They’re just going to say, you built a tool, I don’t really care what the tool is, but you went. You obviously intended to kill that person and you did so, therefore you’re going to go to jail. Right? So don’t current laws on the books actually protect civil liberties? Don’t they actually, um, don’t they already achieve most of what this bill is trying to achieve? Or are there some actual areas that this, this, uh, executive order is pushing into that current laws just simply did not account for

Joseph Nelson: what you’re getting at with your example. Is I think, uh, extremely, um, shrewd and that is regulation that’s targeted at a use case, uh, rather than a technology is more likely to be successful in protecting public interest. So, for example, there’s a number of of regulations on the books today. That prevents, um, discrimination in the process of applying for and receiving, uh, housing. And this is this is a good example to bring up because the executive order calls out, um, algorithms that maybe approve or deny. Decisions for receiving housing as a specific example of, of AI. And I think that the way to regulate something like this is to continue to lean into presently the, the outcome, the, the outcome of there should not be unfair preferential treatment. For someone based on a set set number of standards is how we know if a system is fair or not. It’s not whether or not the technology uses AI. It’s right. The outcome results in exactly equitable treatment of all that are participating in the system. And you take that same form of thinking to your 3D printer example. It’s not whether or not someone used a 3D printer. It’s what they used the outcome of that 3D printer for. Or we look at AI in the context of. Again, in the context of health care, it’s not that I inherently being used for drug discovery is as good or bad. It’s. Well, you shouldn’t be producing and creating things that intentionally introduce farm. And as I mentioned, there’s, of course, already a number of ways by which we can say, Hey, you had an harmful intent that shouldn’t be allowed. And so I think that, um, this is the, this is the. Great thing to call out is that, um, I think a skeptic would say that the executive order is a response to, uh, other countries, especially the United Kingdom, um, and and countries across Europe and certainly the EU. Uh, wanting to quote, unquote, take the lead in regulation. And to me, it’s kind of funny to say you want to be the leader in regulation globally. I think that this is this executive order is aiming to, uh, for the Biden administration. I think they view that view. I think the Biden administration views being the global leader on A. I. regulation as an important place to lead. And so I think that this executive order is basically being able to enter the administration into that conversation. But in doing so, it basically you net out with great. Let’s talk about a what instead of a, you know, the, the why or, you know, the actual use case by use case. Like, maybe we can use another parallel, right? Databases are an incredibly transformative technology. Um, and when they came out and Oracle and my sequel and some of these things. Revolutionized in the way that we’re able to access store and query data and fundamentally changed, um, how software could be built. And if I said to you, Justin, hey, there’s a new technology coming online that allows me to store every single thing, every single record of a place that you’ve been, or a document you’ve written and access and retrieve that record in sub second latency for the purposes of. Making, um, decisions or applications about, um, something you’re going to do, you can say, oh, wow, that technology, that sounds way too powerful. Like those databases, those need to be banned or regulated or better understood. And that’s kind of what’s happening here to some degree, when in reality, what should be done is, well, how are you using that database? What information are you, uh, like the information that you’re storing, you’re using it to what ends? And that’s the same thing for AI, right? It’s. I mean, the age old, like, um, you don’t regulate the use of a hammer you like the existence of a hammer. You regulate how to use, um, a hammer. I mean, AI is an extremely powerful tool. That’s a new form of intelligence. But at the end of the day, it is still a tool that we can wield. Right? So we should, we should be, we should be thinking through it in the lens of, Okay. How do I wield this capability? Not the fact that the capability exists in of itself. Yeah. And let me get one more specific thing about databases to AI present. The way that regulators have been talking about regulating AI, uh, in terms of like its capability has been in terms of model size and parameters. Which is extremely misguided, so the way that model intelligence is often measured is, um, you know, by its capability and its capability is, yes, indeed, typically correlated with the size and a number of parameters of that model, um, parameters, meaning how many things basically it takes into account when making a decision. So if you take more things into account, It follows that you might be more capable of a technique while doing something like you can’t, there’s been some suggestions that a good way to regulate is if you’re a model above some parameter count. Then you need to start reporting on results to the government. Or if you’re using compute above some threshold, then you need to start reporting to the government. This is exactly the type of regulation that is. Somewhat equivalent to regulating by counting numbers of lines of code. Like if you’re, if your application, that’s a good point. If, if your software application exceeds X number of lines of code, then you need to start talking to the government. And as, as you and I talk about that, hopefully that becomes abundantly clear that it’s like, well, it’s, you know, it’s not really about the number of lines of code or the efficiency of your code. It’s more about what is that code doing and for whom and to what end, because you could use a big model for. Relatively benign purposes, or you could use a big model for relatively harmful purposes, but the model size itself isn’t really the relevant portion of that. It’s how are you using the model? And that’s where regulation, um. Can be and perhaps already is targeted. So ultimately I think focusing on, um, you know, the fact that you say AI means that you need to go through a new set of standards ends up being misplaced versus how are you using ai and is that a place where there’s either existing laws that, that protect the equitable and, and fair and just treatment of people? Or is there a new place that we need to investigate to ensure that we live in a society that is right and just.

Justin Brady: Right, because this can set a precedent that, you know, everything capable of hurting another individual should be regulated. And if I can define what is potentially harmful, I can regulate pretty much everything. Even, I would imagine I could regulate paper, you know, paper factories and paper mills, because if someone is motionless and I have a stack of paper, I could technically paper cut them enough to You know, death by a thousand cuts, therefore we should regulate the paper companies. So that is, you know, obviously the concern is that, no, we have, we, we have laws that regulate intent and action. We don’t need to regulate every little tool. Now. I know there are good reasons to regulate certain tools, but, um, you know, it, it creates kind of a bizarre precedent where we need to regulate every little thing that could potentially harm someone. Uh, so, you know, and, and to your point, the Biden order. Does looks like it mandates companies share safety testing results with the US government before they release any official AI systems. And so, you know, if you’re sharing all your data and results with the government, uh, there are, you know, two things are a concern. One. Who’s being influenced to, who’s being influenced to influence these test results? Are they going to want certain data, certain AI tools coming online? That’s obviously a concern, but the other one is, is obviously cost. As well. And the other one is what really defines harmful. That’s still, I don’t see that anywhere in the, in the bill. Um, so one of the concerns is you mentioned earlier about other countries looking at regulating AI and how maybe Biden, the Biden White House wanted to kind of set precedent and they wanted to be the first to do this. Um, if, if regulations do get a little too heavy handed, Uh, over in the United States. Let’s just paint that picture. Like, what if it gets too heavy handed? What if these regulations kind of bogged down innovation? What’s stopping you or your co founders or your company from simply outsourcing your development or your AI work to another country?

Joseph Nelson: I think that’s one potential downside risk is you’re not able to capture and attract innovators. I honestly think the greater risk, though, is the United States should want to be the preeminent place where artificial intelligence advances are occurring. Right? If you believe that AI machine learning is the most forward of technology of our lifetimes, or even 1 of, let’s not take the most, let’s just say 1 of, then you should also want The United States to be leading the way and exploring, identifying and enabling those advances to have a positive influence in the world. Let me paint kind of an alternative. Sure, chat and open. I, of course, are based in the US and, um, showcase, you know, we kind of had some folks describe it like the chat moment. Uh, earlier this year, when it became broadly accessible and aware of how intelligent and capable models are getting and one, I think, filled with with excitement. Now, imagine a world where chat didn’t come out of the U. S. Imagine a world where it came out of another country where. That country published the, um, next best best gen model, and you can imagine envision a world where that model maybe came from a country that, okay, we’ll say, maybe it’s a country that’s more, uh, generally values aligned with us. And that’s maybe less less concerning. Imagine a world where, um, uh, countries that are generally more adversarial with us published it. Like, imagine a world where Russia published the next chat or China published the next chat. And I think that if you consider these, these worlds, um, what you realize is you actually contain a lot of subtle influence in terms of how new technology gets adopted and. What that technology is frankly allowed and capable of doing. And if you think of a country like the United States, we generally have a system of far more openness and. Awareness and freedom of expression. Um, and if you envision a world where QBT, um, which, you know, already has some, uh, folks call it safeguards, but, um, some ways of encouraging or discouraging some types of discussion. If you imagine a world where, um, a country that didn’t support the same level of. Free speech released a model like this, I mean, and, and, and you believe that something like chat GPT is going to be one of the future ways that we understand knowledge and influences education and the ability for folks to learn. I mean, I’m talking to friends who their kids will regularly. In a very fun way, chat with chat is almost an alternative to Wikipedia to learn and digest subjects, right? So it’s not that far to believe. It’s not that far of a leap to believe that chat influences. Uh, thinking and education and these sorts of things. So if you envision a world where, um, the U. S. isn’t the leader in putting out models like this. And in fact, it’s, uh, perhaps places that don’t share a similar set of values that we prize and maintain. Um, I think that that is actually the greater threat than, um, innovators fleeing for elsewhere. No, it’s a, it’s a, this technology is so important that I think that it carries a lot of soft power, right? Of, you know, it’s like the same reason, like, how do you quantify the value of Hollywood being, um, as influential as it is globally? And it being, you know, cool to wear blue jeans and these sorts of things. Like, I think that, um, large models actually carry a lot of similar soft power of. What’s cool and how to understand and how someone comes to learn about the world and

Justin Brady: if we overregulate or don’t do this, right? Then we’re going to get past I can tell you right now The Chinese probably are not going to regulate AI the same way we will or those in India or people in Singapore They’re not going to regulate it the same way Right? And so the countries who have more favorable regulations and don’t really have tons of hoops to draw through, or excuse me, tons of hoops to jump through, will probably get their share of innovators and, uh, you know, could, because if AI is really the next electricity, I believe you’re right. Just because I’ve talked to people like you and a lot of other intelligent people that are in the developer space. But if AI really is the next electricity and we’re saying, yeah, we’re going to really regulate heavily who can use it in the United States, we will get passed. Uh, so, uh, Joseph Nelson, CEO of robo flow, who you guys have everything you need to build and deploy computer vision models, because I’ve worked with you in the past. I know how mind blowingly amazing it is. Uh, for those that don’t know, I always say it like this. I feel like you guys give the sense of sight to software and computers. It’s, uh, so it’s really cool stuff. What’s the website and how do people get in touch with you if they want to learn more?

Joseph Nelson: We’re at Roboflow. com and I’m Joseph Donnelson at Roboflow. com. Thanks so much for the discussion today,

Justin Brady: Justin. Absolutely. Pleasure to have you. And to my fellow Iowans, Iowa expats and Iowans at heart. Thanks for listening and. Subscribing to the Iowa

Joseph Nelson: podcast.