# Bridging AI, Ethics, and Consumer Trust with Miriam Vogel

- **Show:** Raising Health (a16z)
- **Published:** 21 Nov 2023
- **Source:** https://raisinghealth.simplecast.com/episodes/bridging-ai-ethics-and-consumer-trust-NSXVKu0w
- **Transcript:** https://podcasts.opennash.com/shows/raising-health/bridging-ai-ethics-and-consumer-trust-with-miriam-vogel-f5c19c/
- **Provider:** OpenNash Podcasts (machine-generated)

## In brief

Miriam Vogel, CEO of EqualAI, argues that AI should be governed like cars: government sets baseline safety rules and standards, while local rules can vary. She says almost every company now uses AI in important decisions, so responsible practice is not optional. Trust depends on knowing who an AI system was tested on, and on clear documentation, accountability and audits.

The tension is between moving fast and building trust. Vogel says the U.S. must stay competitive with countries that have more resources, yet she believes it wins by listening to diverse stakeholders. She sees progress in agency statements, voluntary commitments and congressional education, but says jurisdiction is unclear, deepfakes are a growing risk, and lawmakers must learn quickly.

## Key insights

1. **Existing laws already cover much AI misuse** Vogel says the FTC, the CFPB, the EEOC and the Department of Justice have each said their current rules apply to AI. Misleading AI claims, unexplained credit denials and discriminatory hiring can already be pursued without new legislation.
2. **AI success rates hide who was tested** Vogel notes AI cancer diagnosis tools can beat doctors, but nobody knows for which patients. A clinician could give a false negative because the headline accuracy does not describe the person in front of them.
3. **Good AI hygiene has four basic steps** Vogel recommends picking a framework such as NIST's, assigning senior accountability, documenting what was tested on whom, and auditing regularly. AI keeps learning new patterns and risks, so the checks must repeat on a schedule.
4. **Deepfakes already move markets and threaten wars** Vogel cites a fake image of the Pentagon on fire, to which the market reacted in real time, and a fake video of Zelensky conceding to Putin. She wants watermarking plus a public that can tell real from fake.
5. **Unclear regulatory jurisdiction is the urgent gap** Vogel says some AI risks span several agencies, and companies do not know which regulators to answer to. Resolving who has final say is a quick fix that helps even startups that want to comply.

## Questions this episode answers

### Does AI need brand-new laws, or do existing laws already apply?

Vogel says many existing laws already apply, citing the FTC on exaggerated AI claims, the CFPB on unexplained financial decisions, and an EEOC and Department of Justice statement on civil rights law. She agrees there are new questions and risks, and that overlapping agency authority needs sorting out quickly.

### Why does Vogel compare AI regulation to cars?

Most people cannot build a car but trust it because government sets safety rules for drivers and mechanics. She says AI needs national standards like that, while local rules, like speed limits, can vary by region.

### What should a company do to use AI responsibly?

Vogel says to choose a framework such as the NIST risk management framework, make a senior leader accountable, document what was tested and on which populations, and audit regularly. She says doing this builds trust with consumers.

### What is the NIST AI risk management framework?

It is a framework that Congress told NIST, the National Institute of Standards and Technology in the Commerce Department, to create. Vogel says it works for any industry, company size or legal setting, and helps organizations ask the right questions about AI risks and opportunities.

### Why is AI bias a problem in healthcare?

Vogel says AI tools can post impressive success rates, such as in spotting cancer, without anyone knowing which patient groups they were tested on. A patient outside those groups could get a wrong result that a clinician trusts.

## Facts

- **Miriam Vogel** [0:01:47]: "So I led President Obama's Equal Pay Task Force and also helped create implicit bias training for federal law enforcement at the Department of Justice."
- **Miriam Vogel** [0:02:21]: "originally, five years ago, when we launched EqualAI, it was this confluence of events where we were first understanding that there were bias and hidden harms in this exciting technology"
- **Miriam Vogel** [0:03:41]: "Almost all companies today are AI companies, if you look at the way that they're using AI in pivotal functions."
- **Miriam Vogel** [0:05:18]: "There are competitors out there that are going at this full speed with much greater access to resources and investment than, for instance, we have in the United States."
- **Miriam Vogel** [0:06:42]: "We're seeing AI do better than the doctors. That said, we don't know for whom."
- **Vijay Pande** [0:07:03]: "certainly children generally don't go through clinical trials and we dose them like small adults for better or worse"
- **Miriam Vogel** [0:09:24]: "we have an AI summit and we came together with some of the leading thinkers to decide on together what are some of the absolute must-haves if you're going to be a responsible AI actor."
- **Miriam Vogel** [0:00:28]: "Now, speed limits, however, are not necessarily federal or nationalized. That is something that can be region, neighborhood specific."
- **Miriam Vogel** [0:11:09]: "if they were talking about their AI capabilities in ways that were exaggerated, well, FTC has purview over you."
- **Miriam Vogel** [0:11:47]: "If you're giving financial scores and not able to explain what the bases are for denying somebody an opportunity, you've got the CFPB."
- **Miriam Vogel** [0:11:47]: "They then went a step forward a year later and had a historic, unprecedented joint statement with the Department of Justice saying, The civil rights laws on the books are absolutely applicable, whether it's AI use or not."
- **Miriam Vogel** [0:14:06]: "Congress mandated that NIST, the National Institute for Standards and Technology at the Commerce Department, create this framework."
- **Miriam Vogel** [0:14:36]: "And what they did is they made it industry agnostic, size agnostic, and law agnostic."
- **Miriam Vogel** [0:15:33]: "You need to have an audit system. You need to know that your AI will continue to iterate. It will learn new patterns."
- **Miriam Vogel** [0:18:15]: "We've seen pictures go around the internet of the Pentagon on fire. We've seen images and video of President Zelensky conceding to Vladimir Putin."
- **Miriam Vogel** [0:18:54]: "With the Pentagon example, we saw the market respond in real time to that disinformation."
- **Miriam Vogel** [0:19:29]: "as was part of the voluntary commitments we saw that frontier model companies make with the White House last month, where they committed to having some kind of watermarking or other labels to ensure that deep fakes are identified as such."
- **Miriam Vogel** [0:22:26]: "I think the average age of Congress is in the upper 60s."
- **Miriam Vogel** [0:22:56]: "Senator Schumer has started this insights forum where he's bringing on some of the leading thinkers from Silicon Valley, from civil society and elsewhere to come educate himself and several of his peers."
- **Miriam Vogel** [0:22:56]: "I think when I started this work five years ago, there were quarterly AI hearings at best."
- **Miriam Vogel** [0:23:33]: "that it should not be used in nuclear armament, that it should be used in some educational spaces, making sure there's watermarking when you're talking about deep fakes"

## Transcript

**[0:00:01] Olivia Webb:** Hello, and welcome to BioWeights World, a podcast at the intersection of bio, healthcare, and tech. I'm Olivia. And I'm Chris. Today's episode features Miriam Vogel, president and CEO of EqualAI, co-host of the podcast In AI We Trust?, and chair to the recently launched National AI Advisory Committee. She is joined by Vijay Pande of Bio + Health. Miriam discusses her work and where she thinks regulation could play a role in AI development. Where do we want to model this?

**[0:00:28] Miriam Vogel:** Cars and safety. Very few of us know how to build a car. We know how to operate it when we know if we trust it. We know what it takes for us to go into a car knowing the driver, the mechanics. So government had a role to play there. Now, speed limits, however, are not necessarily federal or nationalized. That is something that can be region, neighborhood specific. And so thinking big picture, what are the standards and the speedometer that we need to have with AI?

**[0:00:59] Miriam Vogel:** and what are the speed laws where the regional interpretations are more effective and appropriate. Miriam also gave her advice on how companies can improve their AI hygiene. And I think there are some key variables that we all need to be adopting. Whatever size company you're at and whether it's an organization, civil society, government, these are best practices for all of us. So first of all, you need to pick a framework. You need to understand you are probably using AI in a pivotal way.

**[0:01:29] Miriam Vogel:** You're listening to BioEats World from A16Z.

**[0:01:37] Vijay Pande:** Miriam, thank you so much for joining us on BioEats World.

**[0:01:39] Miriam Vogel:** Vijay, it's great to be with you.

**[0:01:41] Vijay Pande:** Can you pinpoint the moment where you decided to delve into AI and what was the catalyst?

**[0:01:47] Miriam Vogel:** Well, it has been something I've explored my entire life waiting for this moment. Not really. Historically, I've worked in law and had a lot of work, obviously, in the tech space involving IP and otherwise. I had also worked in policy, and a lot of the policy was to both support innovation and also bias reduction. So I led President Obama's Equal Pay Task Force and also helped create implicit bias training for federal law enforcement at the Department of Justice.

**[0:02:21] Miriam Vogel:** And originally, five years ago, when we launched EqualAI, it was this confluence of events where we were first understanding that there were bias and hidden harms in this exciting technology that also had the opportunity to address and rectify some of these harms. We have two daughters. And when I saw their education and that they were not having access to computer science at that time or other critical thinking skills of the kind that you would need in this AI economy, I realized we need to do more to make sure we're not raising a lost generation and that we're ready to compete.

**[0:02:57] Vijay Pande:** Can you tell us more about your work at EqualAI? What is EqualAI and what are you doing?

**[0:03:01] Miriam Vogel:** Yeah. So we've been around five years. And originally, the founding principles were to address bias and other harms in artificial intelligence. And so a lot of our work was helping people understand what that meant. Because at that point, that was a pretty foreign concept. People thought of AI as, you know, they had an implicit bias. AI is computer science, it's neutral, and it's correct. So first of all, it was a little bit debunking that myth to help people understand what's under the hood and that AI is a reflection of our society, all the good and bad, and that it needs to be treated as such when you're interpreting its decisions.

**[0:03:41] Miriam Vogel:** And so we quickly pivoted into looking at what does it mean to be a responsible AI actor and helping, first of all, the broader ecosystem understand this is not someone else's problem. Almost all companies today are AI companies, if you look at the way that they're using AI in pivotal functions. And then understanding that most companies want to be responsible. They don't want to create harm and they don't want to create liability with their AI, but it's hard to know what to do right now where we don't have norms, both internationally or nationally or otherwise, that are broadly accepted.

**[0:04:14] Miriam Vogel:** And so a lot of our work is helping companies and helping policymakers understand what is best practice right now. What is the role of an executive within the company, the board, and those throughout the company to make sure that their AI use is responsible and that it's safe and that it's broadening their consumer base, not harming people. And you can't expect companies to be interpreting and navigating that without government, civil society, and others clarifying some of those guardrails.

**[0:04:43] Vijay Pande:** Well, yeah, you can imagine the transition from pre and post electricity. Every company today uses electricity, right? It's not even something we have second thoughts about. And so I can imagine AI would be like that. And certainly when we implement electricity, we want to do it in a way that's safe and responsible. And hopefully if we do it well, it's not even something we worry about, right? I mean, it just works and no one gets electrocuted or anything like that.

**[0:05:09] Vijay Pande:** On the other hand, there's also been concerns of over-regulating. I'm curious how you think about the challenges of where the sort of the sweet spot is between the two.

**[0:05:18] Miriam Vogel:** I think it's a real challenge because on the one hand, if we do not bring in more people of color, more women into the computer science space and the broader AI ecosystem, I think we have significant problems. I think democracy as a whole is depending on us to do better. On the other hand, as you say, it's that fine balance that we are not in this alone. There are competitors out there that are going at this full speed with much greater access to resources and investment than, for instance, we have in the United States.

**[0:05:55] Miriam Vogel:** So it absolutely is a fine balance. But I think there are certain steps where there's not much controversy and where we clearly need government to step in.

**[0:06:04] Vijay Pande:** Could you give a couple examples? I think people love to sort of have a concrete sense of what you're thinking about.

**[0:06:10] Miriam Vogel:** Absolutely. If you're using AI, you need to know for whom it was created. There needs to be use cases. Like in any clinical trial, you know who it was tested for, who it's set for. If there's a success rate, you know for whom that was successful. And you can extrapolate for whom that you may need to do additional tests or have a little skepticism in the success rate. But here, right now with AI, we're talking about these amazing success rates in the healthcare space alone with diagnosing or identifying cancer.

**[0:06:42] Miriam Vogel:** We're seeing AI do better than the doctors. That said, we don't know for whom. So worst case scenario that, you know, you think about what keeps you up at night. For me, it's the doctor, the nurse, the physician's assistant giving a false negative to a patient because they just have no idea that the success rate is not representative of the person who's

**[0:07:03] Vijay Pande:** standing in front of them. Now, I think the comparison to clinical trials is a natural one because we often have some of the same limitations in clinical trials. Typically, there are underrepresented groups there. And certainly children generally don't go through clinical trials and we dose them like small adults for better or worse. I think part of the challenge

**[0:07:23] Miriam Vogel:** is right now deciding what the biggest risks are. I think the hard part is deciding where to act first. And then I think part of the challenge is also some of the things that need to happen are not big headlines. They're basic constructs. It's definitional. What is documentation going to look like? What needs to be documented? What needs to be tested and when? And just even clarifying the definitions of what is tested so that those down the line who are using the AI system understand what it was that was tested with what metrics and with what standards, making sure that that's international.

**[0:07:59] Miriam Vogel:** So there's another wrinkle because you can't just have as hard as it is to get Congress to agree, well, that's not enough because AI doesn't have boundaries or borders. And so how can we expect them to be built as such? But just clarity on expectations for documentation. And then there's the next layer of when it goes to the consumer. I think at the end of the day, one of the biggest challenges and opportunities we have now is to build trust with consumers in the AI system.

**[0:08:26] Miriam Vogel:** And how can we do that if we're not documenting what is baked in in a way that's meaningful? Yeah. Well, one of the things I think

**[0:08:33] Vijay Pande:** is intriguing is that in any human endeavor, there's always going to be bias in some degree. It's almost impossible to remove it. And in AI, in principle, if we can document this, we can actually understand the biases in ways that we probably couldn't even with human beings. My understanding is that you have specific projects working on at EqualAI to reduce

**[0:08:53] Miriam Vogel:** this unconscious bias. Yeah. So our work is really focused on what does it mean to be a responsible AI actor? In fact, in the next few days, we have a white paper coming out. A lot of companies have talked about their AI principles. What we haven't seen is, first of all, how do you operationalize them? How do you make these principles have meaning in day-to-day operations and translate that across your enterprise? What we also have not seen is a broad consensus of key stakeholders.

**[0:09:24] Miriam Vogel:** So what we did with one of our projects is we have an AI summit and we came together with some of the leading thinkers to decide on together what are some of the absolute must-haves if you're going to be a responsible AI actor. So our white paper shares that consensus on making sure you have accountability, making sure that you have documentation, what that means to have appropriate documentation, so that we can break down beyond the platitudes of principles, what it looks like for those companies that have dedicated significant time and resources and thought to making sure that they make good on these

**[0:10:03] Vijay Pande:** commitments. In this Goldilocks game where we don't want maybe zero regulation and the dangers, but we also don't want too heavy. How do you see that balance? Who's overseeing this? How does this

**[0:10:15] Miriam Vogel:** come to be? Another example that I would look to in terms of big picture, where do we want to model this? Cars and safety. Very few of us know how to build a car. We know how to operate it when we know if we trust it. We know what it takes for us to go into a car knowing the driver, the mechanics. So government had a role to play there. Now, speed limits, however, are not necessarily federal or nationalized.

**[0:10:41] Miriam Vogel:** That is something that can be region, neighborhood specific. And so thinking big picture, what are the standards and the speedometer that we need to have with AI? And what are the speed laws where the regional interpretations are more effective and appropriate. And so if we take a step back, we can look at the fact that there are already a lot of laws on the books. And so I think that's the first place we need to start.

**[0:11:09] Miriam Vogel:** We've been fortunate that government has stepped up. I think most federal agencies in the U.S. have now talked about how they're thinking about AI to some extent with their own use, but for the most part, how they're regulating it, how they plan to investigate, prosecute, and oversee other activities. The FTC came out several years ago talking about their purview and that many companies who did not think of themselves as within the FTC purview, well, if they were talking about their AI capabilities in ways that were exaggerated, well, FTC has purview over you.

**[0:11:47] Miriam Vogel:** If you're giving financial scores and not able to explain what the bases are for denying somebody an opportunity, you've got the CFPB. You also have had some joint historic statements. The EEOC came out a few years ago with the Equal Employment Opportunity Commission saying that they, too, are looking at AI. They then went a step forward a year later and had a historic, unprecedented joint statement with the Department of Justice saying, The civil rights laws on the books are absolutely applicable, whether it's AI use or not.

**[0:12:21] Vijay Pande:** Well, I think that's actually a very compelling framework in that, as you laid out in the beginning, AI reflects our culture and our civilization and our laws and what we do and just reflects that back. And hopefully, we have built up the sort of regulatory framework to address the ways that people screw up or the behaviors we want to support. And then, do you think there's anything new then that has to be put in for AI or does it fit within existing

**[0:12:50] Miriam Vogel:** regulatory frameworks? There are basics that are covered. Absolutely, there are new questions. There's also unprecedented risk and there's areas that have multi-jurisdiction purview. So I think in those areas, we need to resolve and quickly who has jurisdiction, who has final say. companies need to know as they're building and deploying to what regulations they need to have their eyes focused and to what agencies they need to be responsive to. I think you make a great point

**[0:13:20] Vijay Pande:** that startups and other companies will want to be compliant, but they just have to understand compliant with whom. And understanding the areas of jurisdiction, I think, is a critical sort of important immediate step that needs to be understood. Yeah. So what would you consider would be good AI hygiene, you know, especially either for startups or for larger companies?

**[0:13:40] Miriam Vogel:** Yeah. Thank you for asking. I think good AI hygiene needs to be a common term. It needs to be an expectation. And I think there are some key variables that we all need to be adopting, whatever size company you're at, and whether it's an organization, civil society, government, these are best practices for all of us. So first of all, you need to pick a framework. You need to understand you are probably using AI in a pivotal way.

**[0:14:06] Miriam Vogel:** Are you using it in your HR systems? And are you using it in your financial, any financial determinations, any healthcare evaluations, any pivotal ways? And if so, or to be proactive, establish what framework you want to use. So for instance, a really good model that I suggest to all the companies I work with and all the organizations is the NIST risk management framework. We're very fortunate that Congress mandated that NIST, the National Institute for Standards and Technology at the Commerce Department, create this framework.

**[0:14:36] Miriam Vogel:** And what they did is they made it industry agnostic, size agnostic, and law agnostic. So wherever you are, whatever size organization you're in, this can be useful to you at asking the right questions to make sure that you've thought about potential risks as well as opportunities. So first, you need to pick your framework. Second of all, there has to be accountability. There has to be somebody in your senior ranks who authorizes, supports, and stands behind this work.

**[0:15:02] Miriam Vogel:** It can be taxing. It can have costs associated with it. It will. And you need to have senior level buy-in to make sure that these are commitments that your company is willing to stick to. Third, I would say you need to have documentation. And the hard part right now is there's not international consensus on what the documentation should look like. And so you should be very clear and explicit on what you've tested and when. Who are the populations that are over and underrepresented in the data that you're using?

**[0:15:33] Miriam Vogel:** To the extent you have that with more traditional forms of AI, if you're talking about generative AI, a little harder to assess that. But you still can talk about with your success rates, the population that was included in your success rates. You need to have an audit system. You need to know that your AI will continue to iterate. It will learn new patterns. It will develop new risks. and problematic patterns. And so you need to develop and establish a routine cadence.

**[0:15:59] Miriam Vogel:** And you need to make sure that these practices, your process is known across your enterprise. So people know what risks you're looking for, who to contact when they identify a problem. And the upside is when you do this, you're building trust. You're building trust in your company, you're building trust with your consumers. And at the end of the day, I would say that's the most important thing we can do, all of us, making sure that our AI systems are trustworthy and making sure that they are known to be trustworthy.

**[0:16:29] Miriam Vogel:** So both making sure that they have our trust and making sure that

**[0:16:33] Vijay Pande:** they have earned our trust. I mean, what do you think are the most pressing ethical challenges

**[0:16:37] Miriam Vogel:** you see emerging from recent developments in AI? One of the basic challenges is how do we move society forward as a whole? How do we make sure that AI fulfills its promise of creating a better society where more people have more opportunity? I mean, you know, AI can help us hear people we haven't been able to hear, see people we haven't been able to see. How do we make sure that we have more people participating in the AI ecosystem so that our AI systems are built better?

**[0:17:07] Miriam Vogel:** We can't ignore the fact that this is a competitive space. We are not trying to achieve AI innovations alone. There are many competitors in competing countries. I firmly believe that we win by leaning into our values, that what makes our AI better is that it has gone through this complicated process of making sure that we do hear from a wide variety of stakeholders, as wide as possible. And we have to do more to understand how we can bring more people into the fold, how more people can understand the AI that they're using.

**[0:17:49] Miriam Vogel:** What does it mean to be a digital literate citizen? What does it mean to be part of an electorate where there's going to be disinformation? And how can we make sure that people can navigate what is real and what is not as it's going

**[0:18:03] Vijay Pande:** to become increasingly complicated. Could we double click into maybe a specific example of maybe some AI tool that presented an ethical quandary of maybe something that came up recently?

**[0:18:15] Miriam Vogel:** Yeah. So we've seen AI used in its best way, which is creating visuals that are hyper believable. We've seen pictures go around the internet of the Pentagon on fire. We've seen images and video of President Zelensky conceding to Vladimir Putin. Now, if that had gone to his troops without people catching it ahead of time, some people were able to notice the mouth movements were off, the eye movements did not look like his, and a slight variation in his voice.

**[0:18:54] Miriam Vogel:** If they were not able to do that in time, and if you're not in an area with great connectivity, if you don't hear or see some of these gaps, obviously that could have significant consequence. With the Pentagon example, we saw the market respond in real time to that disinformation.

**[0:19:16] Vijay Pande:** I remember that. And so what do you do about that, especially since so much of that technology could be open source, it could be done by foreign governments? What kind of safeguards could you have against something like that?

**[0:19:29] Miriam Vogel:** I think there's two ways we need to go about it. On the one hand, we have to demand what we expect to see when you're using this generative AI, this hyper effective, powerful tool. So we need to make sure as was part of the voluntary commitments we saw that frontier model companies make with the White House last month, where they committed to having some kind of watermarking or other labels to ensure that deep fakes are identified as such.

**[0:19:59] Miriam Vogel:** As we go into an election year, it's going to be really important that people understand what they're basing their information on, what is factual and what is made up. On the other hand, we can't go at this alone. We need to make sure we have an educated population. We need to make sure that people understand how AI is impacting their life, their choices, what news feeds they're getting, and their visuals, as well as the content. And so I think we need to be doing a lot more.

**[0:20:27] Vijay Pande:** Yeah, I can also imagine, actually, it's more than imagine. I know these are being built that there'll be AI security companies. And there'll be this cat and mouse game where they will learn how to identify non-label deepfakes, non-watermark deepfakes, and then maybe the deepfakers will get better at it, but then they'll iterate. But this cat and mouse game we've seen in other areas of cybersecurity already. And so I think there's also an infrastructure for that.

**[0:20:54] Vijay Pande:** It just means that this sort of computer game sort of gets taken to another level. So I want to wrap up by talking a bit about where you see the future. You know, how do you see the landscape of responsible AI evolving over the next five years?

**[0:21:07] Miriam Vogel:** Okay, in the next year, I think we've seen some shifts in government. I think people have started to understand that when they're talking about public discourse, there needs to be public-private partnerships when you're talking about AI because a lot of the significant innovation is in industry right now. Now, should it or shouldn't be is another question, but you have to meet it where it is, so to speak. And so I think we've seen both at the White House with their two sets now of voluntary commitments with tech leaders, understanding that they need to be partnering with industry to both understand what the risks are, what the technology is, and how to mold it appropriately.

**[0:21:47] Miriam Vogel:** I think we're starting to see real depth and curiosity where we haven't seen it before on the Hill. With the Insights Forum, we've seen Senator Schumer organize a group of four, making it bipartisan. So again, this is progress. And what's nice about this space is it is still bipartisan. And people recognize that this cannot be a party or partisan issue. It is existential in that I would argue our democracy depends on it. And I don't think it's an overstatement to say that AI will be the type of seismic shift that all past industrial revolutions have presented.

**[0:22:26] Miriam Vogel:** I think it will change how we work, where we work, with what and with whom we work. And I think a lot of it can be positive. But it's really hard for policymakers who operate on several month ahead scenario when they're thinking long term to think about these 10 time 10 year horizons and to think about a technology that they weren't you know, they are not digital natives, to say the least. I think the average age of Congress is in the upper 60s.

**[0:22:56] Miriam Vogel:** So innovation when they were growing up was televisions and other areas that are not as great representative understanding of what we're dealing with right now. And so I think we've seen them step up and do a great job. Senator Schumer has started this insights forum where he's bringing on some of the leading thinkers from Silicon Valley, from civil society and elsewhere to come educate himself and several of his peers. We've seen congressional testimony. I think when I started this work five years ago, there were quarterly AI hearings at best.

**[0:23:33] Miriam Vogel:** And now we're having a few a day where all the different committees are educating themselves on the risks and the opportunities and the thought leadership that they need to create legislation that is going to be useful. And we've actually already seen a lot of legislation that is useful talking about where AI should and shouldn't be used, that it should not be used in nuclear armament, that it should be used in some educational spaces, making sure there's watermarking when you're talking about deep fakes and particularly with elections coming up.

**[0:24:06] Miriam Vogel:** So we have seen that Congress step up and start to educate themselves. They have a long way to go. But I think part of what we will see in the next six months to year is a greater convergence where government rethinks how it operates and has more open doors so that it can learn from those who are creating the technology and be a better partner in establishing the guardrails that are necessary. I mean, that's very exciting.

**[0:24:35] Miriam Vogel:** I think,

**[0:24:36] Vijay Pande:** especially the phrase government rethinks how it operates. That doesn't happen that frequently, right? I guess neither does a major industrial revolution. Do you think it is something like the significance of this that can actually sort of catalyze that? That sounds like it itself would

**[0:24:51] Miriam Vogel:** be a major occurrence in DC. Yeah, I think you're right. I think the seismic shift in government is a lofty ambition. I do think it's going to be required to some extent. And so it's, you know, I think we'll have to see how far they can go with it. But I think many of our basic systems and infrastructure will need to be rethought. I think education is going to need to be rethought in positive ways. I don't know for you, I have shown my daughters how to use chat GPT, not to copy answers, but to spur thoughts, to help them be more creative.

**[0:25:28] Vijay Pande:** One last question for the Miriam Vogels of the next generation that are just getting started right now. Looking back at your journey, what is one thing you wish you knew when you first were starting off?

**[0:25:39] Miriam Vogel:** When I was first starting off, I wish I knew that so many of the lessons that I've experienced, so much of what we've seen, are models for what's to come. So when I first started working at EqualAI, I expected that the solutions to bias in AI and other risks, cybersecurity, I thought that because it was AI, it would be all novel. I think one thing we can take comfort in is it's a lot of history repeating itself now at scale and maybe harder to detect.

**[0:26:15] Miriam Vogel:** The best insights we can get are from history, from where we've succeeded and where we failed.

**[0:26:21] Vijay Pande:** Yeah, well said. Very important work that you're doing. Thank you so much for joining us on BioEats World.

**[0:26:27] Miriam Vogel:** Thank you. It was great to be with you.

**[0:26:34] Olivia Webb:** Thank you for joining BioEats World. BioEats World is hosted and produced by me, Olivia Webb, with the help of the Bio + Health team at A16Z and edited by Phil Hegseth. BioEats World is part of the A16Z podcast network. If you have questions about the episode or want to suggest topics for a future episode, please email bioeatsworld at a16z.com. Last but not least, if you're enjoying BioEats World, please leave us a rating and review wherever you listen to podcasts.

**[0:26:58] Olivia Webb:** The content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see A16Z.com slash disclosures. Thank you.
