Ted Audio Collective.
Hello everyone, this is HBS After Hours, I'm here today with my friend and colleague Mi -Hir Desai.
Hi, Mi -Hir. Hey, how are you, Felix?
It's great to be here, although sad.
We're missing young me, our friend and colleague.
Yet again, but she will be back and we will soldier on.
You brought a topic?
I did. You know, I think media is fascinating and I know you understand media way better than I do.
And so I wanna talk about new media, digitally native media like Vice and BuzzFeed.
These things were the darlings a couple of years ago and now it's becoming less clear.
So I wanna hear what you think and talk about these new media folks, how they make money if they make money and whether they are the salvation for journalism.
And if they're not, what the heck is gonna be the salvation?
What is going to be the future of journalism?
Yeah, exactly. And what about you?
So I fear it might be a darkish episode that I have not the most uplifting topic either, predictive policing.
The idea being that you're using lots and lots of data that now the government has on every one of us and smart or maybe not so smart artificial intelligence to predict who's going to commit the crime.
And so this is kind of AI in a very high stakes sense.
That's AI. Everything that you thought about was scary about minority report seems to come true.
So that movie had a big effect on you, I can tell.
It did. All right, great.
So Felix, I thought we could talk about new media and specifically I was really intrigued by an article in New York Magazine about the rise of Vice.
And Vice is a very prominent new media company.
It's interesting because they've shot to prominence quickly.
They have unorthodox reporting methods.
They have unorthodox revenue models.
They're really quite distinct.
And of course, these have no print legacy.
These are companies that are purely digital.
And at the same time, I happened to read your new case on Buzzfeed, which I thought was really fantastic.
And in particular, Buzzfeed is again, another digital native media company.
No legacy issues, but again, unorthodox in many ways, both in terms of revenue generation and other things.
And so I want to think about these new media companies and you understand them way better than I do.
So this is a little bit of a selfish topic because I want to hear you either.
I want to know the - Well, selfish in the sense that I don't know how to think about these guys because they sound super sexy and interesting.
But when I look at their revenue models, I feel like, wait a second, is this ever going to work?
And are these flashes in the pan?
And there's a big part of me which kind of thinks flashes in the pan as opposed to the future of media.
So help me think about those new companies.
Yeah, so, I mean, the topic is I think super interesting and relevant in part because we're very unsure about the future of these new models and vice would be a good example where it's everybody startling at some point in time and we're thinking billions of dollars of value created.
And now we almost forget why we were ever excited about some of the things that they got right.
Yeah, and the article kind of traces kind of where they are today, which is still uncertain.
Yes, yeah, I think that's exactly right.
But it happens in a context that I think is very important because we're also not so sure how legacy media will survive, right?
So it's essentially this issue that we never ever figured out how to pay for journalism.
But fortunately we figured out, we can print advertising on the back or on the front of the newspaper and that paid for journalism.
With the exception of a few titles, that model has essentially fallen apart.
And have these new guys figured out something better?
I would say the new models have figured out two things.
One is. And new models, just to be clear, we're talking about BuzzFeed.
We're talking about BuzzFeed, we're talking about Vice, we're talking about, these are digital publishers that have significant news or journalistic ambitions.
And at the same time, parts of many of these organizations feels like you have an advertising agency inside the media company.
And the way the advertising and the reporting relate to one another is very canonical, very conservative in some sense and very unusual in some other sense.
And I can explain a little bit how that works.
But I think everything needs to start with, people basically never really paid for journalism.
And we had a neat trick that was called print advertising that paid for it.
Now the neat trick has come to an end.
And advertising prices online are much, much, much lower than they are in print.
And so that's the background.
What did BuzzFeed and Vice and others figure out?
Well, they looked at the development and they, I think, in part created this model that we call branded content now, where it's interesting storytelling that is commercially relevant to the brands that do it.
But at the same time, you and I get at least as much satisfaction and enjoyment out of the telling of the story as opposed to seeing an ad where - We'll walk through an example of this.
So I saw one just the other day, it was a Mini Cooper ad.
And the Mini Cooper did just like a most amazing maneuvering.
Like it's just, you look at it and you go, oh my God, I cannot believe that, like what the car does.
And it was playful in the context of two people you got to know and how their lives relate to that Mini Cooper.
The gold standard is when you look at one of these things, do you feel compelled?
Do you think it's okay to share it with someone you know?
Because the storytelling is of a particular quality that allows you to say, look, I'm sending you.
And I understand this is related to a brand, but at the same time, the storytelling is of a quality that feels compelling, interesting enough that we share it.
And the sharing, I think, is the gold standard of thinking about the quality.
Because if I just click on Facebook and I like something they're like, don't really mean that much.
But if I share digital content with you, that is literally like the traditional word of mouth that we've had and worshiped in marketing for a very long time.
But is that meant to be anything to do with journalism?
This is what I don't understand.
What does branded content have to do with journalism?
Oh, it's the mechanism that allows you to pay for journalism.
Right, but why couple them?
I guess this is curious to me, right?
We still have the same problem, which is branded content could be produced by you and me who decide to set up like a storytelling slash movie studio to help BMW tell stories.
Why does it get coupled with BuzzFeed, which is meant at least in part to be journalistic, truly journalistic.
Yeah, but flip your question and say, you and I have a newsroom, right?
Your newsroom is gonna be expensive.
How do we pay for it?
Well, true, but are we gonna be the best people at branded content?
I mean, presumably the point has to be that because we're a bunch of journalists, we can tell stories better, or we can formulate this better, or we have an audience that we can target in a certain kind of way.
But it doesn't seem to me to solve the problem, which is what's uniquely true about journalism that will create a sustaining revenue model.
This just feels like, and by the way, it comes with a host of conflicts, maybe a lot of other issues as well.
Yeah, so I'll say two things.
One, if we ran this online newsroom, some of the capabilities that we would want for this newsroom are exactly the same capabilities that we would want for branded content.
For instance, we would want a technology that would tell us what gets shared, what doesn't get shared, what do people pay attention to.
We would move around content if you think of it in a sort of a PC world.
We would move around content so that you always see the things that are most attractive to you.
And these capabilities are of course exactly the capabilities that we want for advertising.
And so there is a deep link, I think, in the technology that gets built and in the response that people get when they write these stories.
On BuzzFeed, and you might remember this description in the case, you can point out many, many journalistic stories that in the end look sort of similar to branded content that we had and vice versa.
Because once I'm interested in how likely is it that you're going to share a story?
That makes me think about you and what you're interested in and how you relate to others in ways that is super helpful for journalism and is super helpful for branded content at one and the same time.
Okay, so I buy the idea that they've actually figured out the branded content is a potential revenue model that actually makes sense.
Is it enough? So now...
Is it sustainable? Yeah, and now of course there's basically two big issues.
The first big issue I think is the scalability of branded content production, right?
That is a little bit like the scalability issues on the newsroom side.
So now for how many brands can you do this, how big can you get and so on and so on.
So that I think is a traditional problem.
And then what happened with the rise of video in particular was that social media became the main platform where people would find this content.
So actually very few people go to vice .com or busv .com or any one of these.
Basically this lives to a first approximation, everything lives on Facebook.
I see. So the listicles get posted in Facebook and then they go crazy.
That's right, that's right.
And so Facebook changes its mind about how it wants or doesn't want to collaborate with digital media publishing companies about every five minutes.
So the algorithm, what you see in your newsfeed constantly changes.
And that makes it super, super difficult for these companies to keep abreast of what do I need to do now, so that actually my content is seen which then means I have a reasonable chance that I get paid for by the brands that I work with.
In a way they're getting squeezed by Facebook and Google in the same way the New York Times is getting.
That's exactly right, yes, yeah.
Except the New York Times is sort of a destination in that lots of people will have the app and for most digital publishing companies that's not really true.
Like what they have is exposure on Facebook and exposure maybe on YouTube.
Could have anything, the deeper question you're raising in all this is how do we fund journalism going forward?
And there's been a lot of these transitions recently, right?
So obviously Bezos bought the post, this is a gentleman in California who just bought the LA Times.
I think the Denver Post, I think the headline they actually printed was please bias or something like that.
Exactly, yes. Because they're unhappy with their current owners.
And they're unhappy with their current owners.
So there is this radical notion that actually this doesn't belong in the for -profit form.
These should all be foundations.
And there's a little bit of the ProPublica kind of a thing which I really believe in in a way.
But is that the answer?
That basically this is not a for -profit endeavor and then we should have like billionaires buying newspapers and producing and providing that kind of content?
Is that the answer?
Yeah. So I think there are two models.
Here in the States I think we tend to get rich people, rich foundations who provide journalism.
I think in Europe you would see much bigger efforts to go to some system of government funding where there's gonna be attacks of some sort that then gets redistributed to different media.
Is that happening? Part of it is because you have so much public television in Europe to begin with.
Oh, sure, that is true of course, right.
It's less foreign than it is in the American context.
But those are basically the two things that are out there.
And both of them I think deeply, deeply problematic for the reason that what happens if Amazon has a really serious conflict with?
The Washington Post.
The Washington Post.
Who's gonna win? But also in the government funding model, there is I think even in political systems where you have less concern that it's individual lawmakers having big influence, what's the right way to think about accountability in that model and what's the right way to think about innovation in that model
and how everything gets financed?
So yes, if your first impression is we're fresh out of ideas, I think that first impression would be right.
But maybe it's gonna be okay.
Maybe it works this way.
Which is you got a couple of national papers doing national good reporting, New York Times, Washington Post, Wall Street Journal.
The real question is can you get it done at the local level?
Yes. Right, can you get that journalism done at the local level?
Has anybody figured that out?
Has anybody figured out a revenue model there where at the local level I can actually create journalism that works?
That's much more precarious, right?
And I think for two reasons.
One is you have significant fixed cost.
So if you're a truly local operation, that is really hard to do.
And then unlike buying the Washington Post, a local paper is not a shiny new object that you would say, oh my God, I'm really rich.
You will not believe what I just bought.
I bought some really local paper.
People will say, like, what are you talking about?
So there of course you have alternative forms that maybe don't rely as much on professional journalism where people can share information.
So I'm thinking like neighborhood websites where people now share about what's going on in their neighborhood.
And I'm worried about like the investigative journalist who's going to go to the Baltimore city hall and like figure out what's going on there.
I mean, that's the kind of journalism we care about, or at least I would care about, right?
Cause that's the stuff that really...
But even just, you know, from a local perspective to report, what does Washington do or not do that impacts your local town?
I think we know from business history that has a big impact on whether Washington cares about a place.
But so this is altogether fairly depressing am I taking away the right message?
Which is we haven't figured any of this out.
Buzzfeed is not the silver bullet.
You see it in the valuation.
So the valuations of Weiss and Buzzfeed, they were sky high because we thought, oh my God, these things can grow dramatically.
And I think it's fair to say that 2018 is a year, Buzzfeed has figured out a million things in really brilliant ways.
And you know, it's probably a business that could be, I don't know, half a billion dollars.
Maybe if things go really well, it's a billion dollar business.
But I don't think it's the future of journalism.
Right. Both in terms of valuation, but also in terms of what they're doing.
Well, so this was altogether depressing.
Yeah, sorry. I hear you remember Minority Report, the movie?
Waguely a little bit.
I remember the previews, not a big Tom Cruise fan.
Okay. So that topic of the movie, as you might remember, was that there was a police force.
Right. And instead of hunting people who had committed the crime, they had what they called these precogs, psychics, who were the most very successful in the film.
Basically predicted who would commit a crime, and then you would be at the place of the crime before it could happen.
Now, this is sort of a strange twist, but I think as futuristic that movie was when I first saw it, I think we are actually pretty close to doing pretty much what happens in the movie.
And the question is whether that's good news or bad news, because I confess I'm undecided.
Yeah, that's exactly right.
So part of why I wanted to talk to you about it is because you can tell a story that is very frightening, where oh my God, government is looking over my shoulder all the time.
If you are close to an average American citizen, which I'm not suggesting you are, but there's a 50 % probability that you're in an FBI database today.
Right. So we are vacuuming up data in unbelievable ways.
CCTV is now typically equipped, or very often equipped with facial recognition software, and so on and so on.
We know how vehicles move.
And so the question is, if we use all of this just to identify likely suspects after crime has happened, I think it's just a more sophisticated version of policing.
And that's kind of what's been happening, right?
Like it is like, let's review the CCTV tapes, which car drove through this intersection, ex post, and then you identify suspects.
Yes. That feels fine.
That feels fine, except there's about 50 police departments in the United States who use this very data to predict where crime is going to happen, which maybe we're still okay with.
And then a little more controversial, also predict who's likely to commit the crime.
So we now have lists of likely suspects, where the suspects actually haven't really done anything.
It's just, we think that you're in the category of the person who's likely to commit the particular crime.
So on that first one, my understanding is the results are pretty dramatic.
So they're using it to prevent crime geographically, and it's working, which is in and of itself kind of amazing, right?
Yes. So I think the evidence is a little mixed.
So we have a few cities that have run these tests.
And so basically what they do is they allocate police cars.
They say, you know, how often do you drive through a particular neighborhood to signal police presence?
And there is in some cities, there is evidence that this is working and the terrorist crime...
And just to clear what this is, you reallocate the police cars driving around in response to the predictions of the algorithms about where there's going to be opportunistic crimes.
Yes. So it turns out that urban crime in particular is highly, highly concentrated.
You have cities where in 5 % of all the locations, we have like almost all the crime happens in very few places.
And again, like a little more controversial, the same is true for people, right?
Most people, most of the time, are not engaged in criminal activity.
It's a few people that are much more likely to be prone.
And it's interesting because some of the data they use is, in a way, this is kind of sci -fi, but in a way it feels like no kidding, right?
I mean, like, so, you know, like a hot days and days when payday loans come out or...
You know, what I was a little bit surprised by was, man, I would have maybe watched...
Not that sophisticated.
Yeah, like maybe I've watched too many episodes of The Wire, but like, you know, don't we just know like where these things are happening?
Don't those old kind of detectives just have the intuitions?
I don't know, lesson to me was there's some added value from very seemingly simple algorithms for predicting where crime is.
Yeah, I cannot say that I have significant concerns as long as you predict where it's going to happen.
Right. I think I don't know exactly if I feel the same once we start predicting people.
Like you are the type of person who's likely...
So we have American cities now where police will show up at your door, not because you have done anything, but because you appear on a list with likely suspects for future crime.
But just to be clear, that's been going on for a century, meaning they were informal lists in police people's minds and this is maybe better, Felix, right?
Because look, these cops were basically every time a crime would happen, they're shared like, let's go talk to these people who this fits their MO or their biases, the police's biases.
So isn't this a little bit better than that?
Because it was happening anyway.
So I think better and worse in one particular way.
I think one concern with using artificial intelligence to make these kinds of predictions is that artificial intelligence, as always, is only as good as the data that you use as inputs.
And so we're using machine learning in order to improve these predictions over time.
And one concern that people have is if the basic data is biased.
For instance, we've known for a very long time that whites and blacks are equally likely to smoke mariana, yet the likelihood that you will be stopped and frisked and punished very different across the two races.
So isn't the scale at which we employ this and the black box nature of the algorithm?
That feels different to me from, you have a police department where people know, yes, Mr.
Miller drinks and he's likely to beat up his - This is just sensing which is just self -reinforcing, right?
And especially if the inputs are arrests because then the inputs are, well, we arrested this guy.
And so he's a likely guy to re -arrest, but we arrested him because of the biases we might've had.
And so that becomes really depressing.
But I think your general lesson is the similar to mine, which is, this is to me just a reminder of how these algorithms are sophisticated they are without human guidance and without carefully thinking through inputs are really not that great.
And can actually be very, very destructive.
So to me, it was in a way I came away from thinking about predictive policing, not like, oh, it's amazing the machines are going to take over, but more like it's all the more important, the human guidance that goes into these things.
Yeah, here's something I thought about.
Why don't we require police departments that use these algorithms or the firms that use these algorithms to actually reveal how the algorithms work, similar to what we do inside.
It's interesting. If you and I publish a paper, we give the data away, we give away the code that we have written, and the purpose is not so much that we think, everybody's trying to tell a story that's not really true.
The purpose is that the community is often smarter than the individual person.
And so many of the biases that are described in the popular press, you and I even as semi -sophisticated statisticians would never fall prey to these biases because we were correct for it.
Many of these are classic statistical problems that we have dealt with for a very long time.
So say compare wages of men and women, well, we only observe wages of women conditional on women working.
And we've always known that that's an issue for that comparison.
Same thing in crime.
If we arrest many more African -Americans than whites, there is a very simple statistical way to deal with that.
But if the algorithms are black box, we don't know.
Yeah, that's interesting.
So I mean, the key creates a couple of possibilities when you talk about the community knowing better, right?
I mean, I guess what you have envisioning is some kind of like an open source.
I don't know if that's the right analogy exactly, but police forces put out algorithms.
They get judged in a much broader way and kind of the way we would judge scientific research per year metaphor, right?
And then of course it has many advantages, which is there's learning across communities, meaning your community figured out something I can learn from it and then I can adopt it.
But then we get validation.
The thing that's really appealing to me about your suggestion is, you get much more sophisticated, kind of you can get cutting edge statisticians thinking about these problems in real time and trying to affect algorithms.
That's pretty beautiful.
The one thing that gives me pause and where I don't know how to solve the issue is if the algorithms are public, of course there is a response to knowing what the algorithm is, right?
So if I know, oh, you know, airport security, they're not going to check someone who's shorter than, I don't know, 4 .5 because statistically speaking, you're not likely to hijack a plane.
Of course, as an organization that is thinking about hijacking planes, I would send people who are shorter than that.
And we know that burglars are using social media to figure out things like when people are at home or not at home, right?
I mean, so they are using technology.
They're not actually trying to outfox the algorithms, which is a particularly sophisticated thing to do.
But they are doing a piece.
Something similar, yeah.
I want to just go back to your thing about using it for people because the most troubling part to me, if I understood what's going on correctly, is for parole decisions, they're using these algorithms to predict recidivism, meaning the likelihood of, you know, you committing a crime again.
I think that's right, yeah.
That really bothered me.
Why is that? Well, because this goes against my notion of what parole decisions are supposed to be, which is, I guess I think, let me try to put it in a more sophisticated way.
My instinct is in that setting, there's a lot of omitted data.
There's a lot of unobservables.
I kind of want the parole board to look at the person and say, I see this person and I see what they look like and I see the way they're behaving.
And I don't know, as I say this, I'm not sure it's right, honestly, Felix, but I was really bothered by the idea that let's get rid of parole boards and let's just do this by algorithm.
That sounds terrifying to me.
You know, now no crime has been committed.
We're just actually guessing about your likelihood of committing another crime.
That sounded terrible to me.
And it's apparently widespread.
It's widespread. But it's interesting to me that you think this is terrible while showing up at someone's door to mark police presence before the person has committed a crime.
That seemed okay to you.
No, no, no. I thought, in your example, it was before you committed a crime.
Right, right. Oh yeah, no, those are analogous.
I agree. I thought you were doing it in the sense of post crime committing.
Oh no, yeah. We know who the suspect lists are.
That's totally okay.
Okay, so that was on the - No, no, I'm showing up at your door because I think you moved to this neighborhood and in your old neighborhood - Oh, no, I think that's terrible.
You live close to - Yeah, one way to think about this is there's a big difference about, I thought you were talking about predicting suspects.
Not predicting perpetrators.
You know, those are slightly different.
I think where I'm coming down on this is after a crime is committed, that's fine.
Totally okay, yes. If you think about the geographic issues, that's fine.
You know, by that I mean, there's no personal dimensions to it, right?
We just happen to show more patrol cars in that area.
Anything that happens ex ante before a crime, that to me, I don't know.
I'm not comfortable specifically with Pearl.
Because the whole nature of the system is rehabilitation and I don't know, that really, really bothers me.
I don't want to even try and understand why.
I mean, this is again another great example of technology just heightening and creating an even bigger cat and mouse game than we've ever had before.
Interesting. To be continued, I'm sure.
All right, fantastic.
We've reached I think what is becoming my favorite part of the show.
Recommendations. Alex, what do you got for us this week?
As you know, today was one of these amazing summer days in Boston.
It was sunny, not a cloud inside.
Every day of the year is like that.
It was in Boston. Yes, almost, almost.
And so of course, I did one of my favorite summer things to do, which is eating ice cream.
And I don't know if you have noticed, there is some sort of a revolution going on in ice creams in that the flavors and the flavor combinations, there is an explosion.
Like nothing compared to the 31 flavors obviously, but I don't remember really ice cream being as fantastic and as interesting.
So jalapeno, avocado, chocolate, cinnamon, chili, pepper, churros con leche, toasted coconut, and then combined in really interesting ways.
So if you think ice cream is not for you, if you haven't had an ice cream in a little while this summer, got to go out.
And do you want to actually advocate for a brand?
What's interesting is I think the phenomenon is not really related to any one big brand.
I mean, when you see the big brands, of course, they're picking up some of the flavors that prove particularly popular.
I mean, here in New England, you can even go out to these farm stands and you will see some of these really unusual flavors that just a couple of years ago, you wouldn't have seen.
So rather than following a particular brand, my advice would be be adventurous, both in where you pick up ice cream and then in the flavors that you choose.
Fantastic, that sounds great.
And very apropos for this time of year.
So first, I think on ice cream, I just wanted to give a quick shout out to both Jenny's and Ample Hill's, who are my favorites.
Okay, how about you?
Me here, what's your recommendation?
My recommendation is kind of the opposite, which is, it's a beautiful - The opposite of ice cream.
No, it's not you make me think.
What could that be?
No, in a beautiful summer day, what should you be doing?
And the answer is you should be binge watching.
Ooh. With your pint of ice cream, a Netflix true crime piece called The Staircase.
And it is, I think it's around 10 episodes altogether.
It was first done the first seven episodes and then the last three, but true crime is obviously, these are real events.
In this particular case, it concerns a woman who died at the base of a staircase inside a home.
And it is unclear whether the husband killed her or whether she slipped.
And it is so good because it's so ambiguous.
You really don't know, you can't tell.
So it's not like some of these others where it's like about like finding a killer, it's just about what the heck happened.
And the great thing is at the end of one episode, you're like, he totally did it.
And the end of the next episode, you're like, I don't think he did it.
And it goes back and forth.
Wow, over 10 episodes.
Yeah, and it's really nice.
And it's obviously also a great look at kind of criminal justice and a whole bunch of other stuff.
And the theories, like one is about her falling, another is about him killing her with a poker.
Another one is that she was attacked by an owl.
I mean, it's just crazy.
It's just completely crazy.
And it's fascinating.
So that's my pick, which is the staircase on Netflix.
All right, thanks for listening.
And this has been HBS after hour.
Thanks for watching.
Thanks for watching.
I'll see you next time.
Bye. Bye.
