Eric Goldman: Hello everyone, my name is Eric Goldman. I’m a professor of law at Santa Clara University School of Law, and today I’m here with Nirav Tolia, who is going to talk about his journeys through internet entrepreneurship and some of the ways in which his experiences could help us understand and think more smartly about how we can build online communities and, in particular, what we can do to encourage informed, productive and equitable communications online. So, let’s talk a little bit about your experience at Yahoo. Tell me what you did there.

Nirav Tolia: I joined Yahoo in 1996 as one of the first one hundred employees, and I would describe myself as a lottery ticket winner, having joined Yahoo at that time with no qualifications, really, no experience, certainly not as a technologist or as someone who was knowledge of the ways of the internet from a business standpoint. Although, even back then, I was a very heavy user of the internet and a very, very strong believer in the community aspect of the internet. But I joined the company early by virtue of knowing some of the folks who started it, because I was at Stanford and the founders of Yahoo were, of course, Stanford grad students. So, I found myself there as a surfing Yahoo, that was my title. The company at that point was predominantly made up of engineers who were writing the code that operated the Yahoo set of websites and the surfing Yahoos. We were the folks who would build the Yahoo directory. Remember that Yahoo was the first directory of websites––maybe not the first, but certainly the largest and most prominent directory of websites. Because, at that point, the internet was just getting started, the consumer internet as we think of it was just getting started. And one of the biggest challenges was finding websites that one was interested in. There was no Google. There was no real fast search engine.

So, believe it or not, the founders of Yahoo would take submissions of websites with self-submitted descriptions. In many cases, they would look at those descriptions and then visit the websites themselves. And then they developed an ontology, a kind of taxonomy of websites, a series of categories, and they would place the websites in these categories. And when you visited yahoo.com, you would find, essentially, a series of web pages divided into an ontological tree and, using the browsability of Yahoo, you would drill down into topics of interest.

So, for example, if you were looking for a site on politics, you would go to the front page of Yahoo, you could type in “politics,” and it would try with its search engine to find the pages that match that search query, or you could browse through different categories. You could start with the front page of Yahoo and then you could drill down to politics. And then, of course, there would be international politics, national politics, local politics, city-based politics, it would go all the way down. And you could keep drilling until you reached an end node of the directory. And within that node, there were a series of websites, and you could visit those websites. And, typically, there was a very high-quality bar to be included in Yahoo. So, you would believe, as an end user, that if a site were listed on Yahoo, it was a high-quality site.

Our job as surfing Yahoos was to have areas of the service, areas of the taxonomy in which we had expertise. And then they would shuttle the submissions toward us, we would submit the websites into the directory, and they would be added. And that’s how your website would be found, because everyone at that point was using Yahoo.

Goldman: So, in terms of building communities or encouraging pro-social activities online, what kind of lessons did you learn from that process of taxonomizing sites into a directory?

Tolia: Well, I think that community online goes hand-in-hand with user-generated content. So, community tends to be a set of people who have a shared interest, and then, more importantly to me, they’re able to engage in conversation about that interest. So, it’s not just that they visit a series of websites, and they have that interest in common, it’s that they themselves can connect and converse about that thing. In many cases, the submissions to Yahoo, at that point in time, we’re talking the mid to late ’90s, were not professional websites. They were amateur websites that were created by enthusiasts in different areas of interest, different topics. And these enthusiasts would then actually almost create little online communities around that area of interest. So, while we at Yahoo may not have thought of ourselves as community builders or community cultivators, we were, because we were bringing to light the websites, where enthusiasts could gather around a particular topic or area of interest and create community.

And so, in a way, my job was to find the best websites that could create community and then make sure that the people who wanted to be part of those communities could find those websites. That was the beginning of Yahoo. Yahoo then grew to become a series of different “properties.” There was Yahoo Sports and Yahoo Finance and Yahoo News. And these were less community websites as much as they were destinations on Yahoo itself. But the way that Yahoo started was simply as a pointer to websites that in most cases were created by enthusiasts and people who were creating community. Ultimately Yahoo not only had its own destinations, it ended up building or acquiring a number of different services that I would also consider to be community-centric. For example, GeoCities was a series of websites, or it was really a hosting service for websites that, in most cases, again, were devoted to creating online communities around areas of interest. That was a company that was acquired by Yahoo.

Yahoo had one of the first instant messenger applications. Believe it or not, it was called Yahoo Pager, which is a strange word when we think of messaging services today. But messaging services are natural vehicles to create community. So, whether or not Yahoo would think of itself as a community kind of website or community service, I certainly felt like most of the things that I was exposed to in that job had elements of online community. Yahoo would think of itself as a media company. But I would say, in most cases, that media was user-generated and it then engendered community.

Goldman: Yeah. And I think about just going back even to the initial function of creating the directory, you mentioned that you had a bar for what determined a site to be high quality and to be included in the index. And, of course, then there were lots of people who were trying to gain that system in order to advance their agendas, whether it was spammers or just people who are looking to mess with you. Can you talk a little bit about the challenges and opportunities that were posed by establishing that standard for high-quality services that would be indexed in Yahoo?

Tolia: It’s a great question, and I think it is a natural question, particularly when thinking about keeping communities high quality. But let’s talk about it more generally in the case of Yahoo. What was once Yahoo’s greatest strength, in many ways also led to its downfall. Yahoo’s greatest strength was that human editors were looking at website submissions and then making a value judgment based on their own expertise. The employees of Yahoo had expertise in areas and they would use their judgment to say, “This is a website that should be included” or, “This is a website that should not be included.” “This is a website that should be highlighted” or, “This is a website that is just kind of average.” So, it can be included, but it shouldn’t be highlighted. That process was extremely effective, particularly when the web was wild and woolly, which it was at the beginning.

However, the popularity of the web and the explosion in the amount of content that was created very quickly created an unscalable problem. I remember that being a surfing Yahoo was in many ways a dream job, however, it felt very much like being on a treadmill, because no matter how many sites you looked at and categorized and made this value decision on, you had a queue that was growing larger and larger and larger. If you did a thousand a day, there were 5,000 waiting for you. If you worked your way up to 5,000 a day, there were 15,000 waiting for you. And if you start to think about these numbers, how much of a value judgment can you really make when you’re trying to submit hundreds, if not thousands, of websites for submission into this directory. It’s extremely difficult.

So then, of course, Google comes along. I joined Yahoo in 1996, and by 1999, Google had come along. And Google had invented this magical algorithm called page rank, which looked at the reputation of websites and determined that reputation based on the number of incoming links to that website and used that as an automated way to determine whether something was of high or low quality. And that ended up being infinitely more scalable than Yahoo, which ultimately led to Google dominating the search engine wars, and Yahoo starting to lose prominence.

So, in the early days, the ethos of Yahoo was very, very, very clear. Yahoo was a directory and our users trusted us to ensure that the directory was high quality. We felt a very strong responsibility. I remember extremely passionate debates about websites being submitted that taught end users how to make bombs. Should those be submitted? Was there an issue of free speech? I remember there were websites that talked about suicide and how to commit suicide. Should those be submitted? In almost all cases, I would say the folks at Yahoo were some of the most principled, high-integrity and highest-quality individuals that I’ve met in the industry. And I was extremely lucky to be there. However, it was an insurmountable challenge, a problem, a mountain of content to categorize by hand. And ultimately, I feel like Google came along and Larry and Sergey, they used the interlinked nature of the web itself to come up with a better solution.

Goldman: We’re still wrestling with all these problems today, knowing that content might be legal, but still is otherwise problematic and could be harmful to society. And this balance between human moderation and automated moderation. Initially, Yahoo’s vision for its directory was crafted, in which each submission was lovingly added. All the way to you describing Google as just an industrial grade machine of data without a lot of human intervention. I think we’re still wrestling with all that today. 

Tolia: Well, I think one of the big challenges is anytime a service gets successful enough to become a place where millions and, in the case of Google, now, hundreds of millions, if not billions of people will rely on that source for information, then all of a sudden the incentives for gaming that system. The incentives for being a bad actor, those actually are dramatically high. So, it’s almost, in a way, an arms race. If you are a director, if you’re an arbiter of quality, if you are the quality standard, you’re constantly having to reinvent how you determine that quality. You’re constantly having to stay ahead of the people who have so much to gain. And in many cases, those people, they’re not evil, they’re not evil minded, they’re just trying to do what’s in their own best interest. They’re self-interested. Now, there are some truly bad actors, but most of the people on the internet today, they’re just self-interested.

So, it brings up a very, very interesting question, which is how does a service ensure that self-interest and an equitability and quality and all of those things that get added to this recipe, this mix, how do those things ultimately manifest themselves for the end user? And you’re right, it’s the same challenge that we’re wrestling with today, thirty years later.

Goldman: The idea behind capitalism is that we let each person act in their own self-interest and collectively we make more efficient markets, but, as you point out, it becomes somewhat of an arms race where the people who are self-interested can overgraze a resource in a way that actually ends up being to everyone’s detriment. So, we’re going to move to our next topic in just a moment, but before I do, do you have any further thoughts about whether there is some sense of best practice, or some sort of lessons learned about how to fight these self-interested over-grazers?

Tolia: What’s a natural transition to when I left Yahoo in 1999 and joined a group of extremely bright, extremely ambitious co-founders to start a company called Epinions. And just to back up for a second, when I was at Yahoo, Yahoo was one of the most incredible places to be working in the world. I would wear my Yahoo T-shirt and walk down University Avenue in Palo Alto, and I would get mobbed by people who would say, “I love Yahoo. I love using Yahoo. I can’t believe you work at Yahoo. That must be so cool.” And they were right. But what I realized being there and being one of the first one hundred employees and seeing it grow to 10,000 employees and being there when the market cap was less than a billion and seeing the public stock go to over a hundred billion, all these incredible things, I realized how lucky I truly was and how, in many ways, as I mentioned, I was the winner of a lottery ticket. And if I were walking across the street one day and didn’t look up and a truck ran over me, I don’t think anyone at Yahoo would even notice that I was gone.

So, as I was thinking about what I was seeing, and as I was so infatuated and blown away by what Jerry and David, the founders of Yahoo, had created, I really wanted to try to do it myself. I wanted to try to create that winning lottery ticket for many other people. And I remember at my going away party telling that to Jerry Yang, one of the co-founders of Yahoo, and he sort of looked at me like I was crazy. And he said, “Do you know how hard it is to start a company?” And I said, “Well, I mean, you seem to have done a pretty good job.” And he said, “Don’t be fooled. This is one of the hardest things in the world.” Which, of course, I found out.

So, we decided, the co-founders of Epinions, that there was this incredible opportunity to take this explosion of user-generated content and to apply it to a kind of self-regulating, self-organizing marketplace of experts. And our idea was that everybody has expertise in some area. Can we find a way to be the definitive resource or the definitive destination for that expertise and create community and utility around that expertise? And to your question about what the best practices are, we thought a lot about how to architect an online system for user-generated content and online community at the beginning of creating the Epinions. And we did think about three very important things as part of our framework­­––we thought about structure, we thought about incentives and we thought about this idea of reputation.

This was the beginning of taking, I would say, a more systematic view toward user-generated content. And we looked at Amazon, which had pioneered this incredible––it seemed very simple at the time––but this incredible utility called the customer review. And we felt like that was such an incredible revolution, even though it seems so simple. Because remember back then most people were getting information from Walt Mossberg when they read the Wall Street Journal, or from Consumer Reports. And there was a scale problem there, too, those experts couldn’t possibly write about everything, and those experts may even be biased. So, we felt there was an opportunity to really even the playing field, however, we wanted to do so in a way that was scalable because we felt there were natural network effects that, if we were successful, would make this marketplace larger than anything in our wildest dreams. So, structure, incentives and reputation were three things that we thought about very, very deeply.

Goldman: And as I think back about our time at Epinions, I think about how Epinions was a little ahead of the curve, honestly, that it spotted some of the things that you describing, the structures, incentives, reputations that really were an advance forward for the industry. And, in some ways, maybe even the industry hasn’t fully caught up or it’s moved on in a less healthy direction. But can you talk about what made Epinions unique in terms of structure, incentives and reputation? What were the elements that you thought would solve those problems, compared to what you were seeing at the time?

Tolia: Well, certainly I agree that Epinions was early, and as many early pioneers ultimately have arrows in their back, we, too, received many arrows in our back. But it was really exciting because we were pioneering, as pioneers. We were inventing, we were creating, we were not necessarily refining, we were taking concepts that were brand new and trying to put them into action. So let me quickly describe structure, incentives and reputation, and then I’ll talk about how we applied them. And I think you’re right. These are things that all UGC and community websites struggle with, even today. The idea behind structure is that if you simply give someone an open webpage, an editor that just lets them go in and write whatever they want, the quality that you end up getting is very variable. It doesn’t necessarily have to be bad, but it’s very difficult for it to be consistent because many different people have many different ideas on how they want to submit this user-generated content to any marketplace.

So, with structure, you can use forms, you can require certain things like a minimum word count. You can ensure that the formatting of the submissions and the intent of those submissions ends up being high quality and uniform, at least that was the theory. So unlike Amazon at the time, where you could write anything you wanted as a customer review on a product or service, at Epinions you would show up, there would be a subject, there would be a title, there would be a description, there would be the body, there would be a minimum word count, we would have a spellchecker built in, all of these things that, at the end of the day, were about uniformity and quality. So, when you came to Epinions, that uniformity would help you feel like, “Okay, I know what I’m going to get.” And the quality, of course, was most important. When you came to Epinions you would know that the submissions are high quality. There’s good spelling. They’re well-formed. They look the right way. We wanted to edge user-generated content closer to editorial content. And obviously you know that, even today, whether it’s digital or back then when it was mostly on dead trees, printed media, people spend a lot of time typesetting this content, so that it looks the right way. And that was one of the ideas we had. So that was structure.

The second thing is incentives. We thought a lot about why people do this. And ultimately, I think this is the one that we got the most wrong. We decided to pay people for content. It was such a novel thing to do, not really novel if you consider that most of these professional publications are actually hiring writers, and the writers make a salary. But for an online website like Epinions to pay people for writing reviews, any person, you’re not an employee, you haven’t been vetted, we would just pay you some minimum amount and some maximum amount for your review. And there were lots of interesting ways that we determined how much you would be paid, but that was ultimately our incentive system.

The incentive system then had a kind of witty little approach, which is we would pay you according to how popular your review was. And we would measure that popularity by how many times it was read. Now Eric, as you remember, this led to all kinds of things like reading circles, where reviewers would read each other’s reviews. There would be bots that would be created so that it would look like people were reading the reviews and they really weren’t, I mean, it ultimately was the wrong set of incentives. And I think, globally speaking, I would argue that most online communities that rely on a real currency, like money, as its main incentive, those create almost too capitalistic kinds of marketplaces. And I don’t believe that those ultimately work.

So, all the people out there who try to pay people to do things online, people that aren’t their employees, people that don’t have, in most cases, any intention of taking that job seriously, those systems don’t work very well. And then, finally, let’s talk about reputation, because remember, back then, you could create an account at really any website and not use your real name. In fact, no one did, there was lonelygirl13, that was the sort of archetypal username. And so, we had this idea of reputation where you didn’t have to use your real name, and it’s not just that you didn’t have to use your real name, you didn’t use your real name. However, you would build a reputation based on the number of people that would follow you or be your fans or read your reviews. We used various words for this, and that, too, ultimately lead to lots of problems. Because we felt like creating these fan bases would be valuable. And it was, except for when people started to say, “If you scratch my back, I’ll scratch yours.” And that created all kinds of strange dynamics within the marketplace.

I do believe, however, that structure, incentives and reputation continue to be three of the most important items that you need to really think deeply about when you’re creating an online community. Interestingly, there are examples of flourishing online communities that are high quality that have a very different approach, and by that I mean they don’t think as much about these things. I’ll give you one example that comes to mind: Wikipedia. What kind of structure is there on Wikipedia? It’s just an open text editor. So, on the structure side, they do actually adhere to that rule of uniformity that I talked about, as all Wikipedia pages look largely the same. However, they rely on editors to make sure that, over time, those pages adhere to certain quality standards because it’s just an open text editor when you open it up. So, there are lots of different ways to skin a cat, and we got some of them right at Epinions and, unfortunately, we probably got more of them wrong.

Goldman: Can we drill down a little bit more on this? Because I think each of the three points you raise also raise so many additional interesting questions. Let’s go back to structure for a moment. I think about something like social media today, Twitter or Facebook, where really they are just as open text editor and they’re actually encouraging relatively off-the-cuff and not particularly well-formatted submissions. It almost seems like a step backward from the type of scenario you described Epinions pursuing, where it was very tightly structured in how users navigate it. And it seems like social media almost has gone the opposite direction.

Tolia: I’m not sure about how we bring together quality with structure, but I don’t think that I would quite agree that the services that you mentioned and the other dominant social media services today would agree that they don’t think about structure, and I’ll give you some very specific examples. Twitter’s lasting contribution to structure is having a character limit. So, a tweet was designed to be a short set of words, if not a fragment. And, of course, that’s been waived now, but you still see people do these tweet storms. And you have this idea that it’s better to express a thought in as few characters as possible versus being long-winded. Both because it forces the creator of the content to think more deeply about what he or she is saying, but also from a consumption standpoint, it’s a lot easier to read tweets than it is to read essays.

So, I think Twitter actually has pioneered this very interesting structure. Now let’s talk about Facebook. Facebook would use these leading questions, like, “What are you doing today? What are you feeling today? Tell your friends what you’re up to.” These are ways of getting someone into a mindset, when they post, about what kind of content you want them to create, that structure. The fact that Facebook was really one of the first social networks that prized photos above texts, that is structure. Now, in the case of Facebook, yeah, you could do anything, it could be really low quality, and it unfortunately frequently is. However, Facebook, if we move on even from structure to reputation, Facebook was your real friends. It was real people with your real names, so they got at the quality thing a little bit differently.

Let’s actually move to a completely different social network that doesn’t even really focus on the written word, Snapchat. So how does Snapchat deal with structure? Well, Snapchat is also pioneering and very, very novel because you open the app, and the camera opens up. Well, that’s a form of structure because Snapchat is telling you, “What we want you to do when you open the app is to post a picture.” So, all of these things end up creating, I think, a kind of output, and that’s really at a high level, the chief aim of structure. It’s this idea that the creator of the service says, “I envision this type of content to come out of this service. And I’m going to nudge people.” In some cases it’s nudging, in some cases it’s actually absolutely rigid, like it was on Epinions. You have no choice. You have to do the structure that’s laid out. But in those cases, I would say that social media has done a decent job of keeping the structure. However, I’m not sure that if you only get structure right, you’re bound to have quality as a result.

What really makes these services so difficult to monitor, to regulate, to ensure that they keep the quality bar high, is that these three things, structure, reputation, incentives, they’re sort of interdependent. And they’re all coalescing into some output. They’re not unique or independent of each other, they’re not mutually exclusive. They come together to create something that creates an output. For example, I would argue, in the case of Twitter, which unfortunately Twitter has some of highest-quality content that you’d ever find. And then it’s also got some of the lowest-quality content. Twitter, on the reputation front, you can use any name you want, and now they have verification, the little blue check mark that says that you’re a verified user. But for many years they didn’t have that, and you could use any username you wanted. And, ultimately, followers––which, I wonder about that word, even––followers was one of the ways that you would build reputation. As a result, your incentive was not to tweet the highest-quality content, but to tweet the content that you think would get you the most followers.

So, these things are all interdependent. I mean, I would say that Facebook has done maybe the least work on structure, but on the reputation front, Facebook is your real friends. Facebook required a real email address from a university to join initially, and that kept the quality bar extremely high. However, because you knew that these were your real friends and because you were building the friend graph yourself, and because you could only be in someone’s graph if they actually had chosen to accept you, because remember, you also have this either symmetrical acceptance of a request or asymmetric, right? Followers tends to be asymmetric––I can follow you without you even knowing who I am. Whereas friending is a symmetric kind of reputation builder, which is: I ask you if I can be your friend, and you have to say yes or no. So that’s where Facebook relies mostly on quality. Because, on Twitter, you can have people follow you that aren’t very nice, and they can actually respond to your tweets in a way that’s quite rude. And that’s one of the biggest issues that we find on Twitter. And it’s because of the structure, reputation and incentives that all come together on that platform.

Goldman: I want to come back to incentives and reputation in a moment, but one more thing about structure. When I think about Epinions versus Twitter, you said some of the ways that Twitter structures the conversation, things like a maximum word count or a character count, as opposed to Epinions that had a minimum word count. Or Twitter just having the box that just says basically, “Whatever you want to say, go for it.” As opposed to Epinions, as you said, had much more structured prompts. It almost seems like it’s solving different problems, like it’s lowering the barriers to people expressing themselves. And I wonder if that’s ultimately a healthy approach. There might be room for both, but I’m curious about your thoughts on if we have taken a step back in the kinds of reflective qualities that the Epinions structure might’ve produced when we see something like Twitter?

Tolia: It’s a great question. I think, ultimately, you have two macro trends that are bigger even than the question of online community and user-generated content. And those two trends are, firstly, you have mobile devices taking over the world. On a mobile device, you are much more likely to both create content that’s shorter and consume content that is shorter. So, Twitter is a mobile-native application, and you see that from day one. I mean, it was modeled after SMS, which is a mobile-specific perfect technology. Epinions was a service that was created when there were no real mobile phones that were being used. It was all desktop computers and some laptops. So that’s one big macro trend. And I do believe that we are less conditioned and less able, as a result, to consume longer pieces of content. Our attention spans are actually shorter. We have less patience in general to care about quality. We want to just flick through a timeline and get to what we’re looking for as fast as possible. Those are all trends. It’s hard to understand whether those trends are in effect because the social media services are being built in such a way to take advantage of those trends that are already there. 

And then the second thing is we’ve now created a world where everyone feels like they have a voice––everyone. That wasn’t the case when Epinions started. The idea of Epinions, this idea that there were amateurs who had expertise, that was something that people were a little puzzled about. No one’s puzzled by that today. In fact, we’re at the other extreme, there are people you’ve never heard of that are the influencers, right? They have no credentials, they have no qualifications, they just have large fan bases.

So, I think it’s an interesting question to ask, which is: Are the social media services the way they are because of these two macro trends? Or are these two macro trends there because the social media services have pushed us in these directions? I think probably the answer is a little bit of both. But, Eric, as you and I think about our children, the reality is, they are going to be interacting with user-generated content and online community on phones, not on desktops or laptops. And they are folks who have shorter attention spans, and are more impatient, and are used to scrolling like crazy until they get to what they’re looking for.

So, I think if I were creating Epinions 2.0, and, you remember, there were many Epinions 2.0 kinds of things that we created over time. if I were creating the next version of Epinions, now twenty years later, I would say to myself, “Without ignoring the fact that we’re in a mobile world where people have short attention spans, how can I still raise the quality bar?” So maybe there’s a different approach. Maybe something goes into some kind of beta area before it gets pushed out and is read by everyone. But I don’t think it would be very difficult to swim against this tide of short-form content in snippets on a mobile phone. Extremely difficult. The closest I’ve seen that people have done that is the emergence now with user-generated content and online community of newsletters. The most important example I can think of, of longer-form, high-quality content that still creates online community and is, in most cases, created by what we might think of as amateurs or everyday people versus professionals is things like Substack, which are quite interesting to look at today.

Goldman: Substack, and an entire genre of other payment platforms like Patreon are OnlyFans, they all have that incentive issue. So, can we come back to this incentive piece for a moment? If I understand your perspective, you would say now that paying for content in the Epinions environment was ultimately a mistake?

Tolia: Yeah, I felt like it was a mistake for a couple of reasons. I don’t think, generally speaking, that Patreon and OnlyFans are making mistakes. In fact, I think they’re onto something, but they’re doing something very different than what Epinions did. And these are the two very specific ways that they’re different. The first is Patreon doesn’t pay its creators, the end users who consume the creator’s content pay Patreon, and then Patreon pays the creators the majority of what those end users have paid them. In the case of Epinions, it was a very strange kind of marketplace because it’s not that readers of reviews would come to Epinions and then pay Epinions, and then Epinions would go and pay the majority of that to the content creators. Now, in a way, we had abstracted that, and said, “We’re going to generate advertising revenue based on the content that’s created. And then we’re going to share that advertising revenue with our creators.” But it was not a direct relationship between the consumer of the content and the creator. Whereas OnlyFans and Patreon, there is that transactional relationship, that customer relationship between the entity that’s creating the content or providing the experience and the person who’s paying for it.

So that’s number one. Number two, in most cases, Patreon and OnlyFans are catering to folks who have an existing reputation and, in most cases, have decided to create an occupation, maybe not a full-time job, but a mostly full-time job out of doing this kind of thing. And Patreon or OnlyFans is just another channel for them in their full-time job. If you remember some of the greatest review writers in Epinions’ history, they were not professionals. They were not looking to quit their jobs and write professionally. Now maybe Epinions made the mistake of paying them so much that they thought they could do that, but that was not their intention in joining the service. Their intention was to share their expertise. But then we messed everything up by deciding to pay people for sharing their expertise. Where, in most cases, and you see this in Wikipedia and you see this all over the web today, people are perfectly happy to share their expertise because it gives them joy. That’s the currency that they’re receiving. That’s the incentive. The incentive is that they are sharing their expertise with the world. They’re helping people. And then, if it gets to be big enough, they find ways to monetize that. But I feel like Epinions taking the central role in that monetization, that was a pretty big mistake. I mean, it almost caused the company to go bankrupt and die.

Goldman: Well, and I’d like an analogy or compare/contrast with Wikipedia, because if you would look at Wikipedia and say, “That can’t work.” It can’t work that people would invest all that labor, and Wikipedia’s rules don’t allow them to monetize their work whatsoever. And in many cases, the Wikipedians don’t even develop a high profile, enough reputation to monetize their reputation outside of Wikipedia. So, the money just isn’t there for Wikipedians. And you would say normally, from an economic standpoint, that people will prefer the site that’s paying over the one that’s not. And yet, you’re suggesting it was actually almost a reverse.

Tolia: Well, let’s talk about Wikipedia, because I think it’s particularly important in the context of this question you’ve asked, which is about quality, right? When Wikipedia started, people said, “My gosh, this is a joke. This will never be Encyclopedia Britannica. This will never be a real encyclopedia. And today, it is the reference standard. It’s more timely. It’s more complete. It’s more up-to-date. So, let’s run through our structure, incentives and reputation, and think about what Wikipedia did. I think what we’ll find is going to the other extreme is what yields success. So, structure––Wikipedia has a set of rules that you need to follow when you submit things, but the text editor is completely open. It doesn’t rigidly force you into those rules. In fact, they have editors, other volunteers who will then enforce the formatting correctly.

So that’s where they went on structure. It’s the simplest website in the world. I mean, I don’t think they have specific mobile UI, they don’t have a specific desktop UI, it’s just a text editor that looks like a very simple page, in one font, with some formatting, with very few pictures. I mean, it’s a throwback that is one of the most important services in the world today. So, that’s structure. Let’s talk about incentives. As you mentioned, Wikipedia goes so far as if the people on Wikipedia, the Wikipedians, the community that is about creating the asset, if they believe that you have an incentive to gain by adding content to Wikipedia, they will either note that or they will take your content down. So, they have gone to the other extreme. Instead of saying, “We will provide these incentives,” whether they’re explicitly or implicit, whether they’re monetary or whether they’re ego-based, they have said, “You know what? You won’t actually have incentives here. Your incentive is to build this thing that’s larger than you.”

And that brings me to the last, which is reputation. You don’t use your real name on Wikipedia. You don’t really get attribution on Wikipedia. You have to go behind the curtain to understand who wrote these pages. And I would say less than 1 percent of the people who enjoy Wikipedia every single day actually go behind that curtain. They don’t click through, they just read the content. So, Wikipedia said, “Oh, your reputation is not important to us. What’s important is the reputation of Wikipedia. So, if you’re up for contributing and making sure that the reputation of Wikipedia is here, then you can contribute, if not, we’re not interested.” So maybe an interesting topic to consider is if Wikipedia is the model for a high-quality output of online community and user-generated content.

Goldman: Can we talk a little bit about reputation? You haven’t mentioned the web of trust. Could you tell us a little bit about the web of trust and how that was the structure for Epinions’ reputation model?

Tolia: The words that you choose when you’re architecting one of these systems are so important, and “trust” is a word that we used at Epinions, and it’s such a loaded word. I mean, it’s a little bit like the follower word, right? Before we get to “trust,” “follower” is a word that, to me, harkens to, “I’m a fan. I’m someone who’s following you.” And that has a particular connotation. And it matters to people that they have followers, so that was actually quite a genius word to use. In the case of Wikipedia, we would ask the question: Do you trust this reviewer? What a loaded question, right? Because, ultimately, do you choose to trust someone based on the quality of their content online, which they are writing behind some pseudonym. I mean, can you even make that judgment? We might’ve asked a simpler question such as, “Do you think this person writes high-quality content?” But it doesn’t quite have the ring of, “Do you trust this reviewer?”

So, we would ask that question and, if people would click “yes,” you would create something as an end user called a web of trust. You would have this web of people who trusted you. Ultimately, those writers who had the largest webs of trust were the ones who received the most promotion. And promotion was simply that their content went to the top of the list and the people who didn’t have a very strong web of trust in the community, their content would sink to the bottom. And the problem was, once again, it was a brilliant kind of idea, it was a very catchy set of terms. Interestingly, we tried to patent it. Eric, you may remember. And we got some feedback that the term “web of trust” had been coined by one of the inventors of the internet, if I remember correctly. So, we had a prior use issue with it. But ultimately it led to what we think of as, again, this kind of circle of trust, not web of trust, where people would say, “Hey, look, I will say that I trust you if you say you trust me.” And then we’ll sort of do this quid pro quo kind of thing.

And then, all of a sudden, we can’t tell where the trust really is. So, what would we do differently? Well, maybe you would get twenty points of trust and you could only give out twenty. And after you gave out twenty, after you trusted twenty people, you can’t trust anyone else. And maybe you can’t trust anyone until we know that you’ve read thirty pieces of their content. Or you have to write why you trust someone. But again, once you’re off to the races . . . And I think this is an important point, the reason that it’s so critical to think about these frameworks before you get started. Once the system starts, the system takes over. You’re no longer in control. Imagine what it was like when we told our Epinions community, who we were paying to write reviews, “We’re not going to pay you anymore.” Or when we introduced something first called “income share,” which was obfuscated and difficult to understand. And essentially, just an excuse for us to pay a lot less to our reviewers than we paid them before, because we were going to go out of business. We weren’t generating any revenue and we were still paying for this content.

So, you start these systems, and you don’t realize at the time that, if you’re lucky, lightning strikes and you have some success. But then you realize that something you put into the system from the beginning is not leading to the right output, it is so difficult to change. I think that’s one of the reasons why it’s almost impossible for Facebook to copy services. I mean, the next company I started, Nextdoor, there’s now something called Facebook Neighborhoods, and, of course, Facebook is quite well known for building competing services. And, in most cases, the competing services don’t get off the ground. It’s because you have to architect one of these services or systems from the ground up for it to really have the right kind of output. And when you have an existing service, it occupies a space in the consumer’s mind and it’s very difficult for that consumer to think of it any differently.

So, as an example, if Facebook says, “We want you to post pictures like you do on Snapchat.” People will say, “But I’m used to status updates. Why would I go to Facebook to post pictures? I’m used to status updates.” And so, this is actually one of the challenges. But in the case of Epinions, using words like “eroyalties,” which was what we used when we were paying people for writing reviews. Expressions like “web of trust,” those were very novel at the beginning, and we loved them because they were catchy, and people responded to them. But when we realized they weren’t quite right, it was very difficult, if not impossible to put the genie back in the bottle.

Goldman: I want to come back to the trust piece in just a moment, but this notion about switching gears on your community, I think it’s pretty interesting. It’s almost about what value proposition aggregates an audience. And then if you change that basic value proposition, it seems like that audience is going to say, “But that’s not why I was here in the first place. You’re asking me to do something that didn’t attract me in the first place.”

Tolia: I think you’re exactly right. It’s a question of expectation management, but I think it goes deeper. Any online community that’s successful feels like it’s owned by the community, not by the commercial entity that started it. So, you want the community members of your website to feel like they own it. That’s why they care. That’s why they spend all the time there. That’s why they do the things that you want them to do to create this incredible marketplace of user-generated content. So, when you change something, almost as if you’re God in that system, people don’t like it, they weren’t consulted. I mean, we went through this at Epinions, we created councils and we had conference calls and we had interviews and it was never enough, because, ultimately, we had succeeded beyond our wildest dreams in one way, which is our users felt like Epinions was theirs. But then it was almost impossible to change those things. So, at the low end, it’s expectations management. At the high end, if you’re really successful, it’s about ownership. It is about ownership, and you are changing the core of something that someone else feels like they own.

Goldman: Yeah. That’s a pretty powerful statement, honestly. This notion that, to succeed, you want your community to feel like they own it. And it reminds me of an experiment from about a decade ago. I don’t know if you remember this, where Facebook actually tried to put its content policies up to a user vote. Do you remember when they tried to do that? And they got almost no participation from their users. So, they realized, “There’s no point in us building a system if the users aren’t going to engage.” So, I wonder, is that a contrary example, that Facebook has architects so differently that we are never going to feel like we own it. And therefore, if they ask us to try and own it, it’s actually fighting against the value proposition.

Tolia: No, I think in the case of Facebook, something as obscure and vague as content policies is not what the users on Facebook feel like they own. On Facebook, what you feel like you own is a mouthpiece of communication to your friends and family members. That’s what you feel like you own. So, as an example, let’s take the competitor to Nextdoor that Facebook is building, Facebook Neighborhoods. If, all of a sudden, you start to see in your feed posts from people you don’t know, but they happen to live in your neighborhood, you may ask yourself two very interesting questions, which are completely different than anything you’ve experienced on Facebook, or prompted by the fact that it feels like a very different experience. The first is: “How does Facebook know where I live? How does Facebook know that I live in this neighborhood? And they’re showing me content. Does this mean that, in some way, Facebook is taking advantage of information it has about me?” The other thing you might ask yourself is: “Why am I seeing something on Facebook from someone who isn’t my friend?”

And so, what I mean by the ownership of the experiences, it’s less literal than “I own this car,” or “I own this computer.” It’s about an experience that’s owned. What I own about Facebook is not the service itself, but I own the experience of seeing things from people that I have a symmetric friend relationship with. And those people can live around the world, or they can live where I do. But where I live shouldn’t be important to Facebook. So, if Facebook then changes its core value proposition, it’s the ownership of the experience that I feel like has been changed. It’s an important concept. It’s not a literal one, when it comes to ownership, but what it essentially means is you’ve built this thing and now I have curated it because of my own contributions. And I have this experience that I feel like I own, I have my friend graph. You don’t even think of it that way, because “friend graph” is a very Silicon Valley term. But then when all of a sudden it changes, it’s very, very difficult to reconcile.

Goldman: It makes me think about my experience when I was a contributor at Forbes. They gave me a platform to publish, but I was bringing a lot of the audience to the table. And then they put up an ad system that basically froze people out of blocked ads. And my readers were just flaming me in my inbox saying, “You’re asking me to unblock these ads.” And I’m saying, “I’m not, Forbes is.” But it was Forbes’ interposition with my readers that actually drove a wedge in my relationship with them. And that’s what, ultimately, I think, put me over the edge to point where I said, “This is not working for me. They’ve changed the relationship between me and my readers in a way that ultimately doesn’t work to my advantage.”

Tolia: So, maybe that’s a better word, “relationship.” Ownership is also a loaded term, because it makes you think of something that is about possession. But this is really more about the relationship that a platform has with its end users. And when the platform changes in substantial ways, that relationship can become very difficult. And look, the really insightful thing about Facebook is people don’t trust the company based on surveys and what we read today. However, Facebook is exceptionally good at evolving its product by essentially telling its users, “Hey, we know better than you do on this, and we’re going to show you that over time. And you will trust us and stay with us.” And they’ve done it over and over and over again. Remember when the newsfeed was invented by Facebook, which is one of the greatest innovations ever in the history of social media? So many people didn’t like it. There were all kinds of revolts. And I give Mark Zuckerberg so much credit, because, like the true visionary that he is, he held strong. Even though that relationship that Facebook had with its users was changing a lot, but it was changing for the better.

Goldman: I want to get to Nextdoor, but I did want to mop-up this issue about trust, because so much of the debate today about the internet is about how we know whom to trust online. What are the indicia that someone is trustworthy? And happens when we misdirect our trust or when we accept content that isn’t trustworthy and treat it as if it is? I just remember back in the Epinions days that this notion about trusting another author was always uncomfortable to me, because I don’t trust people categorically. I might trust somebody as being an expert in consumer electronics, but I’m not likely to trust them as equally about vegan foods. So, this notion about trust seems so contextual in nature. And I’m wondering if you have any further thoughts about your lessons from trying to force these trust relationships on top of the interactions on Epinions. What did you learn about today’s concerns we have about who we trust online? How do we know we can trust?

Tolia: Well, two things. One is you essentially answered the question when it comes to Epinions, but I would say in a very general sense, “trust” is such a loaded word that means so many different things to so many different people, that I would use that word at peril, particularly explicitly. Now you can say, as the creator of one of these communities, “I want to build a trusted community.” Or, “I want to make sure that trust is one of the pillars of this community.” But when you start to ask your end users whether they trust something or not, or in the case of Epinions, whether they trust someone or not, I think that’s fraught with danger.

The second thing I would say is this is a perfect segue to Nextdoor. We created Nextdoor in 2010, so it’s over ten years old. I would say that the thing that we’re probably most proud of with Nextdoor is that we feel very strongly about earning our members’ trust. And we have done a number of different things along the structure, incentives and reputation fronts to not just make that something we say, but something that we have done. When we did things to try to build trust, we never used the word “trust,” because that was a mistake that we made at Epinions. But we did try to build more trust in our system and with our members, and many of the mechanisms that we used that we felt would do that were poo-pooed quite a bit. In many ways, they hobbled the system and its ability to grow quickly, but we felt like those were the right investments to make. Ultimately, now with 99 percent of American neighborhoods, in one and three households, and eleven countries around the world using the service, I think it’s been working pretty well. But we never said, “Do you trust your neighbors? Do you trust Nextdoor? Would you trust this particular person?” We tried to do things that would increase the level of overall trust. And I can give you some of the specifics.

Goldman: Yeah. Let’s drill into that. And I have some other questions about Nextdoor. You said you did some things that people advised you not to do that looked like they were going to slow the growth, but that built the company for the longer term. What are some of those things?

Tolia: At the very top of the list are two things, both of which were very, very antithetical to the way that services were being built back in 2010. The first is all content on Nextdoor is private. So, this idea that none of the conversations on Nextdoor are indexed in Google, or can be accessed by anyone outside of the network, that’s very, very unique because most of these services grow because the content can be discovered outside of the network. So, we built a close network, because we said to ourselves, “Look, we need people to feel that these conversations online are as similar as possible to the conversations that they would have offline in their neighborhoods.”

When one has a conversation with a neighbor, that conversation is not broadcast to anyone who wants to read it or listen to it or hear about it, that conversation is between neighbors. So, let’s create a private social network for the neighborhood. And those were the words that we used. We didn’t use the word “trust,” but we said, “Nextdoor is a private social network for the neighborhood.” There was no discoverability of content. You never had to be afraid that something you posted on Nextdoor is being read by some random person across the web. Now, who is it read by? That was the second thing we did. We felt like it was critical, once again, to proxy what happens in the real world. When you speak with a neighbor, you know the person your neighbor, because you’ve walked down the street and you know where the person lives. Or you’ve bumped into the person at the corner park, or you know the person because that person’s children go to the same school that your kids go to.

So, in the early days of Nextdoor, we said, “You cannot join the system unless you verify your address.” Because, in our world, your address was an objective identifier of whether or not you lived in the neighborhood. So, those two things, restricting access until you prove to us that you lived in the neighborhood and, even then, ensuring that all the content that was created was kept within this closed network, caused a lot of interesting debates in our board meetings. I mean, the third thing we did was, from the very beginning, we said no to people who said, “I don’t live in that neighborhood, but my parents live there.” Or, “My kids go to school there.” Or, “I visited there, and I really want to read what people are saying there, not because I’m being pernicious or because I want to do something harmful, but I just want to know the best pizza place there, because I like to go there on the weekends.” And we said, “Sorry. If you live in the neighborhood, you have access. But we’re not going to let people do drive-bys in neighborhoods that they don’t live in.” And that not only was very antithetical to what our end users wanted, it also really cut down on the network effects in the system because there were only network effects within a neighborhood.

If you think about it, growth is the lifeblood of any consumer internet company, particularly a social media service. And, from the very beginning, think about the friction we created, no discoverability on search engines or across the web, no ability to join and even see the service until you verified your address. And finally, no ability for membership to spread from one neighborhood to another, in a viral way. Doing these things were incredibly, I use the word “antithetical,” I mean, it was just counter to all the best practices. And we weren’t doing these things to be stubborn, or because we felt like it’s just better to think differently or something like that, we were doing these things very specifically because we felt like we needed to build a system that was based on trust. And the way we could do that was to as well as possible understand what goes on in the real world and then create an online environment that proxies those customs. So that was really the beginning of Nextdoor, which was very different than anything that had been created until then.

Goldman: And I like the way you frame it, because I think in all respects, Nextdoor mirrors the physical space of neighborhoods. But the mirror also shows us some of the ugly sides of our neighbors. Can you tell me a little bit about how Nextdoor has been dealing with that?

Tolia: Well, there’s sort of an obvious ugly thing and then there’s a non-obvious ugly thing. So, let’s talk about both of them. The obvious ugly thing is people are passionate and they have different points of view, so they can get into fights. They can get into fights in the real world, they can get into the fights in the online world. So, in a neighborhood, when your Nextdoor neighbor puts a Trump sign in his front lawn, and you don’t like Trump, that might be an issue for you. And the same thing can happen online. It’s a mirror of the things that are going on offline when people have political conversations on Nextdoor.

For the longest time, on a wide variety of those kinds of topics, we’ve said, “This is not content that’s suitable for Nextdoor.” Nextdoor, in many ways, could be an unbelievable place to talk about politics because you know where people live, and people tend to vote based on where they live. But we have said that national politics really has no place on Nextdoor, and local politics is a different story. But you do have disagreements and you’ve got neighbors that get into arguments, and that’s the obvious challenge that we knew that we were up against from the beginning. There’s a non-obvious challenge, and it’s because I would say it’s not intentional. Nextdoor has received many allegations of racial profiling. And I will describe what racial profiling is. It’s essentially when a neighbor posts on Nextdoor that they see a suspicious person walking around the neighborhood. And they assume that person is suspicious because of the person’s appearance. And, in many cases, the person’s appearance is based on skin color, based on the way they dress, based on the way they wear their hair. And, in almost all cases, it’s a person that “looks different” than the person who’s posting. Unfortunately, in almost all cases, the person that is being posted about is from an underrepresented minority, an African American, a Hispanic person, someone who tends not to be Caucasian.

So, when the first allegation occurred or when the first set of major allegations occurred, I was stunned because I am a person of color, and I would never be part of creating something that would racially profile. But as we dug a little deeper, we began to understand that the issue was one of unconscious bias. It was an issue where neighbors were posting not to be racist, but because they truly were trying to help their fellow neighbors. But they didn’t understand that simply noting that people look suspicious based on the way they looked versus based on their actions was a form of racism, whether it was an unintentional bias or intentional.

So many companies would say––and we did this in the beginning as well––“Let’s do a statistical analysis to understand what percentage of content could fall into this category of unintentional bias or racial profiling.” And when we did that analysis, not surprisingly, it was less than 1 percent of 1 percent of 1 percent of our content, because mostly what’s going on in Nextdoor is people recommending babysitters. And they’re asking about lost dogs, and they’re talking about school fundraisers, but suspicious activity is something that people post about. And we felt, ultimately, that even one post that caused someone to feel like they had been racially profiled was too much. So, we rolled up our sleeves and said, “Can we do something about this?” We have been creating all kinds of systems, largely around structure, incentives and reputation that discourage people from just posting as quickly as possible, and encourage people to stop and think before they act. Because, in most cases, the best way to combat unconscious bias is simply to make people slow down and stop and think about what they’re doing.

And this, again, is so counter to so many of the things that are happening in the world today. And even the things that you and I have talked about, because we’re in a world that wants instant gratification, that wants to post as quickly as possible, that wants to remove as much friction as quickly as possible in the process. And what we are essentially saying is, “Before you post something in crime and safety on Nextdoor, think about what you’re doing. And if you’re going to say that there’s a suspicious person, help us understand why that person is suspicious based on behaviors, not simply based on appearance.” So, it’s a kind of two-sided coin. I would say, of all the things that I’m most disturbed by that occur on Nextdoor, racial profiling’s at the top of the list. But the other side of the coin is all the work that we’ve done to build this online system, the work that we’ve done around trying to reduce racial profiling, is some of the work that I’m proudest of.

Goldman: And when I think about Nextdoor, to me it’s at the vanguard of the opportunities that we have online to change the dynamic of people treating each other worse because they’re online, because they can hide behind anonymity, or they lack the physical presence that might inhibit antisocial behavior. Nextdoor seems like it has an opportunity to actually make us be nicer to each other. And I think about the kindness filter, in addition to the tools you mentioned about anti-racial profiling as basically ways of trying to be nicer to each other online than we might otherwise be in person. There’s no one who’s sitting on my shoulder saying, “Goldman, you’re crossing the line.” And you’re saying something that someone’s going to find to be a product of unconscious bias or interpret as mean. But Nextdoor actually mediates the conversation in a way that acts as that little good angel on my shoulder saying, “Hey, you can do better.”

Tolia: Well, and it’s all based on that very, very powerful word, which is “neighbor.” What does it mean to be a neighbor? What does that really mean? To us, it’s really sort of tied up in the golden rule, it means that you treat other people the way you want to be treated. It means that you know that your words have impact. And you know that because you will say things and then you’re going to drive and walk around your neighborhood and people will see you and understand that you are behind those words, because on Nextdoor, number one, you have to use your real name. Number two, they know that you live in the neighborhood because you verified your address. And number three, at this point, with 99 percent of American neighborhoods using Nextdoor, chances are that your neighborhood is on Nextdoor and mostly everyone is reading what you’re saying.

So, you can think about that as the kind of angel and the devil on the shoulder telling you not to do bad things and to be careful, and you can also think of it as an opportunity to engage with people that you might not see because people’s commutes are longer today. And because our neighborhoods are not nearly as active during the day, because in America we have such a pro-work culture. People don’t really sit out on the front porch as much as they once did. And this is one of the reasons why we don’t know our neighbors. I mean, there was a very, very simple statistic from the Pew Research Center that we held onto so deeply in the early days of Nextdoor, and that was that 30 percent of Americans could not name a single neighbor by name, not even one. That isn’t the way that I remember growing up. That isn’t the way that I wanted my children to grow up.

And today, when we say on Nextdoor, “One third of all US households are using the service.” Well, look, you open up your phone, you tap on Nextdoor and you will immediately see dozens of names of your neighbors, and that’s tangible progress. But it’s all back to that idea of being a good neighbor, and I don’t think this is something at all that Nextdoor pioneered, it’s just a trend that we’ve tried to be part of, which is we believe the world should be a kinder place. We believe the world should be a more accepting place. And we refute the notion that the only result of online community and user-generated content is the opposite of those things. We refute the notion that the internet has to be a wild and woolly place where people just insult each other.

We believe that technology is a force for good, and it can be. You just have to be thoughtful about the application and you have to be committed to sticking with it. Because, as we talked about, the stakes are really high and there are people out there who are acting in their own self-interest. And, instead, what we need to try to communicate to them, and what we need to try to build in these systems, is to make it easy to act in everyone’s self-interest, in our collective self-interest. And the reason community is so important to me is because, to me, the definition of community is being part of something that’s larger than ourselves.

Goldman: One last open-ended question for you. We’ve talked about so many different lessons learned, things that you would have done differently at each of the three stages, but is there any other broader takeaway lessons that you want to offer up now about how we can build online communities encouraging informed, productive and equitable conversations online? 

Tolia: Jeff Bezos has this very famous, but I believe very powerful notion, which is that it’s “day 1.” I believe that about the internet, and I believe that about online community and user-generated content, but I will say that different times within “day 1” have different areas of focus. And where I believe we should focus now is on being positive and trying to create systems that encourage people to be good. I fear that there is too much focus on the bad actors, and we’re so busy trying to remove or eliminate or punish the bad actors that we’re not spending enough time trying to figure out ways to encourage and amplify and bring along the good ones. And the reason it’s so important to do so is because you have this bi-modal distribution of lots of bad actors on one end and good people on the other end, but in the middle is where you have the largest potential community. And in that middle, we want to nudge people to be good. We want to make it easy for them to be productive. We want to show them that the internet is a force for good, as is technology, not just an abyss where people can scream whatever they want, and people end up turning off.

If I decide tomorrow that I’m not going to participate on social media because it’s just not worth it, there are too many people that are abusive, there are too many things that are unpleasant, then I’m not sharing. I’m not building community. I’m not connecting. I’m not providing examples that can uplift people. So, I think in this new era now, this part of day 1, let’s focus on the positive and let’s build systems that reinforce that.