Latest / General Magic & The Soul of Tech / Ep. 10 | The Internet’s Last True Believer Defends Section 230 | Eric Goldman

Transcript

Ep. 10 | The Internet’s Last True Believer Defends Section 230 | Eric Goldman · General Magic & The Soul of Tech

0:01
SPEAKER_0: Hi, this is Mike Stern.
0:03
SPEAKER_1: And this is Tom "Hersh" Hershenson.
0:05
SPEAKER_0: Welcome to General Magic & The Soul of Tech.
0:11
SPEAKER_2: We believed we were building something that would change the world, and in some ways we did.
0:16
SPEAKER_2: In others, we didn't understand what we were setting in motion.
0:19
SPEAKER_2: The story of General Magic isn't over.
0:22
SPEAKER_2: We're still living inside what it started.
0:29
Our guest today is Eric Goldman,
0:31
who is professor and now,
0:34
as I understand it, Associate Dean at Santa Clara University School of Law.
0:39
He has a history in technology law that stretches back to the early 90s,
0:45
both on the private practice side, in-house, and now as a member of academia.
0:51
Now, as part of setting the table,
0:53
we have to talk about Section 230 of the Communications Decency Act.
0:58
And for listeners who are uninitiated in that,
1:00
what was the genesis of that,
1:02
the objective of that,
1:03
and your involvement in championing it and its enduring effects to this day?
1:10
Section 230 says,
1:11
just to summarize super quickly,
1:12
that websites aren't liable for third-party content.
1:15
SPEAKER_3: It creates an allocation of responsibility that somebody might be doing something wrong online.
1:22
SPEAKER_3: But in order to do that, they use a bunch of tools and services.
1:24
And those tools and services aren't liable for the bad actions of
1:28
the person who's committing the wrongdoing.
1:30
The principles of Section 230 are actually quite intuitive,
1:33
even if Section 230's label has gotten a bad rap.
1:36
Congress enacted in 1996 in response to a couple of cases that had come out in
1:41
the early 1990s,
1:43
that collectively between those two cases had suggested that an internet service
1:48
that was trying to curate conversations online would be liable for anything
1:53
that went wrong in those conversations.
1:55
And this general legal paradigm created the risk that services wouldn't try
2:00
to curate the content,
2:02
wouldn't attempt to up-level it from the standard likelihood
2:08
that people were gonna take the conversation to the gutter.
2:11
And as a result,
2:12
either the conversation was gonna end up in the gutter because
2:15
the services couldn't intervene,
2:18
or the conversations would go away entirely because the services couldn't bear
2:23
the liability for other people's actions.
2:26
SPEAKER_3: So Section 230 just cut through all of that.
2:28
SPEAKER_3: And it said simply, there's a wrongdoer.
2:31
They should be liable for it,
2:32
but everyone else around them isn't,
2:34
and that general principle has been the foundation of the modern internet
2:37
as we know it today.
2:39
SPEAKER_3: Now I had nothing to do with Section 230's enactment.
2:41
SPEAKER_3: I was practicing law at the time, as was Mike, together.
2:45
So you know,
2:46
at the time we were advising clients about how to proceed
2:50
without any statutory protection.
2:52
Once Section 230 came out of the books,
2:53
and once we understood its power,
2:55
it opened up a whole new way of advising clients and guiding them towards thinking
3:00
about how to create conversations between users and not take responsibility.
3:05
SPEAKER_1: So two points I just want to clarify real quick.
3:08
First of all,
3:09
am I right in thinking the analogy is sort of like the difference
3:12
between somebody getting a book published that has libelous content
3:18
in it versus somebody who makes a printing press.
3:20
No one would imagine that because you're a manufacturer of printing presses,
3:24
you'd be held liable for what someone's words that they put out using the press.
3:29
SPEAKER_1: Is that a kind of a fair analogy or am I missing kind of the thrust of it?
3:33
Yeah, don't love the analogy,
3:34
in part because at least you go back in history and printing press operators,
3:39
in fact, faced extraordinary liability for their conduct.
3:43
SPEAKER_3: So it was you know, always risky to publish other people's content.
3:48
SPEAKER_3: And that had been the dominant legal regime for centuries.
3:51
And Section 230 turned that long standing venerable premise on its head and said,
3:58
you can publish content, but if it's not yours, you're not responsible for it.
4:02
And that opened up the door for a whole new set of conversations
4:05
that were never possible under the old model,
4:08
some of which are good and some of which are problematic.
4:10
You were outlining that absent 230,
4:12
platform operator faced one of two choices,
4:15
either not host these conversations at all,
4:18
as a way of escaping any possibility of liability,
4:21
or expose attempt to moderate them or curate them and you know,
4:25
the traditional liability would attach to the content on their platform.
4:30
SPEAKER_1: Am I getting that right or am I mistaken in that?
4:34
No, you summarized it properly but I would add there is a third possible option,
4:38
but it's really a null set.
4:40
The other option is to allow everyone to have their say,
4:44
moderate it or curate it perfectly, in a way that eliminates all liability.
4:49
SPEAKER_3: But that's a null set.
4:50
SPEAKER_3: That's not possible to do.
4:52
And so if you take that option off the table,
4:55
the other two options are it turns into a cesspool or publishers exit the industry.
5:00
SPEAKER_1: Got it.
5:00
SPEAKER_1: Mike.
5:01
Yeah, you know,
5:02
I don't disagree that in 1996,
5:04
as the commercial internet was just beginning to emerge,
5:07
that Section 230 helped create the industry and without it
5:11
the industry probably wouldn't have developed in a meaningful way.
5:15
But it's sort of to me it's like mercantilist or protectionist theories of trade,
5:19
that when you're building a nascent domestic industry,
5:22
protecting it makes sense,
5:24
but at some point you stop doing that,
5:26
because competition will make its evolution sturdier and more productive.
5:32
We're 40 years on or 30 years on from Section 230,
5:36
and we have clear and convincing evidence that the platforms like Facebook
5:42
and TikTok and Instagram and X, are purposely designed to create harm.
5:48
We know that from the discovery in the trials in California and Nevada,
5:53
which Meta and YouTube lost.
5:55
SPEAKER_0: And we know it from books like Careless People.
5:58
SPEAKER_0: Doesn't that change the whole playing field for why Section 230 should persist?
6:04
It's sort of borrowing ideas from contributory and vicarious infringement theory,
6:09
we know that the platforms are consciously designed to cause harm.
6:14
At this point,
6:14
does that mean that a different scheme and regulation that imposes some kind
6:20
of liability for that kind of action is appropriate?
6:23
SPEAKER_3: So let me start with your Section 230 angle.
6:27
Part of the reason why the cases are proceeding is because the courts have held
6:30
that Section 230 does not apply to the claims that are at issue.
6:33
SPEAKER_3: And so we could have eliminated Section 230 and the legal claims would look identical.
6:39
SPEAKER_3: It might be that the services don't exist or would be architected entirely differently.
6:44
SPEAKER_3: But Section 230 isn't holding back most of the claims that the plaintiffs are advancing.
6:49
SPEAKER_3: Now you said something really provocative, and this part is where you lose me.
6:54
SPEAKER_3: You say that the services were intentionally designed to cause harm.
6:59
Now they're definitely intentionally designed to cause something,
7:03
but to say that they're intentionally designed to cause harm to their users,
7:07
I think is an oversimplification and maybe even wrong.
7:11
But I think that the difference between what I might call product marketing,
7:15
where all businesses try to figure out what do their consumers want and try
7:19
to give them more of that.
7:21
SPEAKER_3: And designing a product that is intended to harm its users, are two really different things.
7:28
SPEAKER_3: And I would encourage us to keep it separate.
7:30
Well, I'm thinking of the tobacco analogy,
7:32
the cigarette companies knew that their product was poisonous and continued
7:35
to market it and continued to obfuscate the facts about its health effects.
7:40
When Facebook deliberately targeted teenagers with body image problems
7:44
or enabled its advertisers to deliberately target teenagers with body image problems
7:49
with content that would exacerbate that mental health problem,
7:53
isn't that a fair description of what I said?
7:56
So I think that there's some really interesting questions here and we can spend
8:00
the entire time talking about Meta and some of the choices that it's made.
8:03
SPEAKER_3: And I'll be candid, I would make different choices.
8:05
I'm guessing the two of you would make different choices as well if faced with
8:09
the evidence that they had in their possession.
8:11
But I also think that it's tricky to analyze that evidence in the abstract
8:18
without considering all of the other subpopulations that were at issue
8:23
that might also have important stakes in this conversation that
8:28
a service like Meta is trying to balance.
8:30
They're trying to figure out how do we cater to these heterogeneous subpopulations,
8:35
some of whom are deriving substantial benefits from the design choices that
8:40
a service like Meta was making.
8:42
And so the problem isn't that there was some obvious low-hanging fruit where
8:47
the services necessarily could redesign and make every other user better off,
8:52
the more likely scenario is that there were some subpopulations that were vulnerable
8:55
and other subpopulations that were benefiting.
8:58
And we don't really have a good principle for how a company ought to proceed in
9:02
a circumstance like that.
9:03
So that gets to a question on a theoretical level,
9:06
which is premised on the following idea that a given technology inherent to it is
9:12
a likelihood of certain social impacts, right?
9:15
SPEAKER_1: And the analogy I often use is you can't make a kinder, gentler gun.
9:22
SPEAKER_1: I mean, a gun is a tool for destroying things.
9:24
SPEAKER_1: And in certain circumstances, it may be the perfect tool.
9:26
And in other circumstances,
9:28
it's gonna have not just unfortunate,
9:30
but really you know, devastating antisocial impacts.
9:35
But one thing is for sure,
9:37
if you give a lot of people guns, you can expect things to be destroyed.
9:41
So if we accept the premise that certain technologies have tendencies in terms
9:45
of their effects in society, how do we deal with an emerging technology like AI?
9:50
SPEAKER_1: How do we understand what it is likely to do in terms of its social and political impact?
9:57
SPEAKER_1: Do we let it run its course?
9:59
Do we try to anticipate what those are and have regulatory and legal regimes to try
10:05
to head those off?
10:07
SPEAKER_1: How do we proceed?
10:09
Not necessarily just with respect to AI,
10:11
but how do you,
10:12
someone who's thought very deeply about these issues,
10:15
approach it at a conceptual level?
10:18
SPEAKER_3: Yeah, I love the idea that technology has a bunch of intrinsic biases in it.
10:24
SPEAKER_3: The example I always give is scissors.
10:26
SPEAKER_3: As you may know, scissors are generally designed for right-handed people.
10:30
SPEAKER_3: So they don't work very well for about 10% of the population.
10:34
SPEAKER_3: It's not like the scissor manufacturers are intending to be biased or catering to the market.
10:38
It just so happens that the design choices that they make have a bias,
10:42
and for the people who are in the 10% of population that isn't going to be able
10:46
to use the scissors very well, it feels crummy.
10:49
SPEAKER_3: Something like guns are super tricky to discuss.
10:52
SPEAKER_3: I'm not a gun owner, don't plan to become one.
10:54
SPEAKER_3: But there are legitimate reasons why people own guns.
10:57
We might decide that guns are appropriate for law enforcement,
11:00
we might decide that guns are appropriate for things like hunting.
11:04
SPEAKER_3: And then there are a bunch of illegitimate uses of guns, of which we see too much of that today.
11:09
SPEAKER_3: It's not the gun that's necessarily the problem.
11:11
SPEAKER_3: It's the way that people use the technology.
11:14
And so part of the questions then are how do we teach people about the proper use
11:18
of technology and how do we think about the fact that there will be people
11:21
who will misuse the technology and set a set of boundaries around that?
11:26
SPEAKER_3: That might not be the responsibility of the technology provider.
11:30
That might be something that's a social problem that's outside the scope
11:33
of any design choice that the manufacturer can make.
11:36
Do you think it's also outside the scope of government to enforce certain boundaries
11:41
or do you see it as rather a something that society at large needs to tackle?
11:46
SPEAKER_3: I mean, you know, I can't do the Spanish, but why not both?
11:50
SPEAKER_3: It might be that it takes a whole-of-society response to new major technology evolutions.
11:54
SPEAKER_3: It's not the responsibility of any one player in the ecosystem to figure out all the problems.
12:00
SPEAKER_3: That's unrealistic, frankly.
12:01
SPEAKER_3: And so we're gonna need, you know, a combination of teaching users how to be responsible.
12:06
We're gonna need the technology manufacturers to think about how they can improve
12:11
the safety and efficacy of their offerings.
12:13
SPEAKER_3: We're gonna need the government to regulate misuses of the technology.
12:17
And we're gonna need a general overall social compact or understanding
12:21
about what do we want from people?
12:23
SPEAKER_3: How do we set those norms and then enforce them?
12:26
SPEAKER_3: Who's the right mechanism to enforce it?
12:28
This is standard law and technology stuff and this is stuff
12:30
that we've all been working on for decades.
12:33
SPEAKER_3: And yet each time we get a new technology, I think there's a little bit of an amnesia problem.
12:38
So from my perspective,
12:39
I think there's a lot of historical lessons that can help us answer stuff like AI.
12:43
Now, I would engage with you on the question about AI,
12:46
but I cannot do so until we define what we mean by AI.
12:50
SPEAKER_3: It means too many different things to have a good conversation without being more precise.
12:55
Last year you had a research paper turning things on its head,
12:58
a lot of people think AI is gonna doom humankind,
13:01
but your title was Generative AI is Doomed.
13:04
SPEAKER_1: What was the thesis of that paper and the basis of the arguments that you were making in it?
13:09
SPEAKER_3: The thesis of the paper, and I focused on generative AI.
13:13
SPEAKER_3: So I'm talking about things like chatbots or other kinds of AI content production tools.
13:20
That because those tools became popular during a time of techno panic,
13:25
those tools were going to get a regulatory blowback that was going to hinder
13:31
or limit their ability to come to market.
13:34
And so the examples I would give is that we see all kinds of efforts
13:38
to ban some populations from having any access to those technologies at all,
13:43
like under X age cannot use a chatbot whatsoever.
13:48
But we're also seeing,
13:49
and the thing that I think is the real threat,
13:52
the most likely thing to destroy or undermine AI,
13:57
is the partisanship that we're seeing governments thinking
14:01
that they can decide what are proper outputs from a chatbot.
14:06
And they're going to do that in a way that will reinforce their power,
14:09
almost certainly in a partisan way that will reward one ruling class to
14:15
the exclusion of all other potential rivals.
14:17
And so for example,
14:18
you could imagine the thought experiment,
14:21
you know,
14:21
asking a chatbot who won the presidential election in 2020,
14:26
that there could be a government intervention that would literally say there is
14:30
a canonical answer to that that may not be well accepted truth.
14:34
SPEAKER_3: And in a situation like that, then the chat becomes worthless.
14:37
SPEAKER_3: It's not helpful to anybody because it's just government propaganda.
14:41
SPEAKER_3: And to me, we are on the verge of that, if not already moving in that direction.
14:45
SPEAKER_3: And that puts an enormous amount of pressure on the ability of generative AI to be helpful.
14:50
And you believe that the life of generative AI,
14:54
to put it that way,
14:56
is potentially at stake in terms of various governments charging ahead
14:59
in this realm.
14:59
SPEAKER_1: I mean that's not I'm not overstating y your view in saying that, is that correct?
15:05
Yeah, I put that on the table because I think that if we understand that
15:10
if governments decide what's appropriate for chatbots to say or not,
15:13
they don't work,
15:15
there's nothing left that's valuable from them in that circumstance.
15:18
SPEAKER_3: They are just a tool of government control.
15:21
So let's hope that there will still be other applications of generative AI
15:26
that wouldn't be subject to that kind of government propaganda control.
15:30
Let's hope that there will be enough innovation entrepreneurship to find ways
15:35
to resist those outcomes.
15:37
SPEAKER_3: But I take a pretty bleak view because at the moment I'm not seeing those barriers.
15:41
All I'm seeing is they're a cheering crowd every time that the government attempts
15:45
to regulate generative AI saying please do more of that.
15:48
SPEAKER_3: And that's just seems to be taking us further into the downward spiral.
15:53
SPEAKER_0: Yeah, but we're not talking here about regulating the output of chatbots.
15:58
SPEAKER_0: We use age gating for many different technologies and products and services.
16:04
SPEAKER_0: Someone under 16 can't drive a car.
16:06
SPEAKER_0: Someone under 18 can't vote.
16:08
Someone under 21,
16:09
at least in the United States,
16:10
can't drink alcohol, and they can't buy cigarettes anymore.
16:13
SPEAKER_0: We use age gating in those contexts.
16:15
SPEAKER_0: Why is age gating so problematic for access to generative AI tools?
16:21
So I'm gonna answer your question,
16:23
and that might be the last conversation question you get from me,
16:26
because I have a lot to say on that point.
16:28
SPEAKER_3: But I did want to make sure that we didn't lose a thought.
16:30
You said that we aren't talking about partisan control over chatbot outputs,
16:34
and I am,
16:35
and I think that that is the existential threat to
16:38
the entire generative AI industry,
16:41
that it is so long as those outputs aren't protected by the First Amendment,
16:45
they're not treated like the editorial expression of the people operating
16:49
the technology,
16:51
governments can force the chatbots to essentially say whatever they want.
16:55
And that means that the chatbots become even worse at lying or making up information
17:01
than they are today.
17:02
SPEAKER_3: If we exclude that problem, then yeah, generative AI looks fine.
17:05
SPEAKER_3: But I don't want to exclude that problem.
17:07
SPEAKER_3: I think that's critical.
17:08
So let's start with the premise that there are a number of things that we ought
17:16
to restrict by age.
17:18
SPEAKER_3: I'm not questioning the legitimacy of that.
17:21
I am, however,
17:22
concerned that the process by which we do age authentication online looks different
17:30
than the ways that we might do age authentication in the offline world in ways
17:33
that matter a lot.
17:35
So I'm gonna enumerate five things that I think are different about
17:40
the online age authentication process compared to some of the examples you gave,
17:44
like purchasing alcohol The first is that in order to do
17:49
the age authentication online, there is going to be significant privacy exposures.
17:55
SPEAKER_3: Essentially there are two primary ways to do age authentication online.
17:59
That is to do some kind of biometric review like a face scan,
18:02
or to do some kind of government ID review like scanning
18:06
a person's driver's license.
18:08
And in each case,
18:08
the disclosure of either the biometric information,
18:11
the government ID,
18:12
conveys significant private information that is going to be captured
18:17
by whoever is doing the authentication.
18:19
SPEAKER_3: That's not true in the offline world.
18:20
SPEAKER_3: You might have to flash your driver's license, but usually no copy is kept.
18:24
SPEAKER_3: It's simply not recorded anywhere.
18:26
SPEAKER_3: You know, the person who's seen it can't even scan it.
18:29
SPEAKER_3: Because of the privacy exposures, then there's also significant concerns about security.
18:36
SPEAKER_3: Once that information is available in electronic form, it can be intercepted.
18:40
SPEAKER_3: And if it's stored anywhere, it can be expropriated.
18:43
And so now we've got a greater security risk than you'd face in the offline world
18:48
because in the offline world there may not have been any capture of that information
18:50
in the first instance.
18:51
The third issue is that the authentication process acts like a speed bump in
18:57
the ability for people to get from point A to point Z.
19:00
Now, in the offline world,
19:01
you go and you buy alcohol,
19:02
the cashier says,
19:04
I need to see your driver's license, and there's a few second delay.
19:08
There is a speed bump in that process,
19:09
and we generally accept that as part of the overall checkout process.
19:12
Now in the online world,
19:14
when we're trying to get to say a particular item of content,
19:17
that speed bump might be far more significant as a percentage of the overall payoff.
19:22
If we're only going to look at the content for a few seconds and it takes us
19:25
a few seconds to get through the process to get to that few second consumption,
19:29
all of a sudden the equation looks different.
19:31
SPEAKER_3: People aren't willing to go through it.
19:33
So what happens in practice is that a bunch of content becomes effectively invisible
19:38
because they're not willing to navigate the speed bump,
19:40
or they lack the technological ability to do so.
19:42
If it's a driver's license requirement,
19:44
for example, millions of Americans don't have any government issued ID.
19:48
SPEAKER_3: They're simply out of business altogether.
19:50
The fourth thing is the fact that age authentication online imposes costs on
19:55
the publisher having to do them.
19:58
Now,
19:58
there is an implicit cost if you've got your cashier doing a driver's license check
20:02
when someone's buying a six pack.
20:03
And that's a qualitatively different cost than the cost
20:06
of both implementing age authentication online and then actually conducting each
20:10
and every individual authentication.
20:12
SPEAKER_3: And for publishers who can't afford it, they're simply out of the industry altogether.
20:16
The last of the five things I'll mention about what's different
20:18
about age authentication online is that it is an infrastructure for surveilling
20:22
the movement of people online.
20:24
SPEAKER_3: That's the whole point.
20:25
SPEAKER_3: And the governments are going to be very interested in that.
20:27
SPEAKER_3: They will co-opt that for further control of population movements.
20:30
And as a result,
20:31
what we're doing is we're building infrastructure that the governments are going
20:33
to be very, very excited about and ways that they can weaponize against us.
20:36
Last thing I'm going to mention is with respect to the examples
20:39
of things like cigarettes or liquor,
20:41
we should know that the restricted products are not speech.
20:44
SPEAKER_3: But online, almost everything that's age authentication is speech.
20:47
And there's also then constitutional dimensions that are raised by that
20:51
that are not present in the offline world.
20:52
So sorry for giving you the long answer to that question,
20:56
but it's so critical because it's so easy to say,
20:59
we do age authentication all the time.
21:02
SPEAKER_3: What's different here?
21:03
SPEAKER_3: Answer is almost everything about it is different in ways that matter a lot to everyone.
21:07
But there's one part I think that isn't different,
21:09
which is the notion that you're balancing interests or you're balancing harms.
21:15
SPEAKER_1: And you've very eloquently detailed five very important considerations.
21:20
SPEAKER_1: I'm wondering what do you see on the other side of the scale?
21:23
What are the countervailing considerations that we should bear in mind with respect
21:28
to age verification, age restrictions on internet use or internet content?
21:33
Or do you take a position that it's just unworkable and therefore it's a matter
21:38
of parental or societal responsibility as opposed to something that would be subject
21:43
to legislation?
21:44
So let's start with I think what Mike was asking earlier,
21:47
which is isn't okay to have some products and services restricted by age?
21:52
SPEAKER_3: And the answer is yes.
21:53
And so for those,
21:54
I'm totally fine with the idea that we're going
21:56
to require age authentication online, just like we would in the offline world.
22:00
So for example,
22:00
if you want to purchase liquor online,
22:02
it makes complete sense that there would also be some form
22:05
of age authentication coupled with that,
22:07
so that the rules are the same online or off.
22:09
SPEAKER_3: That makes sense to me.
22:11
Now,
22:11
the question is what are the benefits that might be available to the individuals
22:16
who might be age restricted?
22:18
And how can we think about preserving those benefits and still mitigating any of
22:23
the potential age inappropriate material or the safety concerns
22:26
that might be implicit in that?
22:28
SPEAKER_3: Like that's a really hard question.
22:31
And that requires a lot of thought and care about what problem we're trying
22:35
to address and what are the tools in the toolkit to address it.
22:39
SPEAKER_3: I like the analogy that sociologist danah boyd gave about crossing a street.
22:44
SPEAKER_3: We can teach children, don't cross the street.
22:47
We could teach children that they're not allowed to cross the street
22:50
until they reach a certain age.
22:51
But really,
22:52
what children want,
22:53
and what I think is in the best interest of society,
22:56
is to teach children how to cross the street responsibly.
23:00
And so we might think about the analogy to how can we teach children to use
23:04
the internet responsibly as their age and maturity supports it?
23:10
SPEAKER_3: How can we help them cross the digital street responsibly.
23:14
And so to me,
23:15
if we ask the question that way,
23:17
what do we need to do to help teach children to cross the digital street?
23:21
It's a completely different conversation than how could they be harmed
23:24
and let's take away the things that might cause them that harm.
23:26
SPEAKER_3: That's like telling them don't cross the street at all.
23:29
I just have to point out that there are laws against jaywalking and I don't know
23:34
about in Santa Clara but here in DC,
23:36
somehow that has not succeeded in eliminating that.
23:39
SPEAKER_3: I'm sorry.
23:40
SPEAKER_3: And just to point out, those laws are usually enforced in a in a bias and discriminator fashion.
23:44
SPEAKER_3: They are on the books, to be used as a tool for the police to control population movements.
23:51
Actually I think it's a great example of we should be careful whenever
23:55
the government's controlling how people are moving around in the society,
23:58
we have to ask, are they doing that for the right or wrong reasons?
24:01
SPEAKER_0: Is there any way to know if you're wrong about Section 230?
24:05
Is anybody doing any experiments with imposing tort liability on platforms
24:10
for third party content in the same way that Australia and England
24:14
and other countries are experimenting with age gating and online verification
24:18
for access to the internet?
24:20
SPEAKER_0: Are there ways to figure out what the cost would be of changing it?
24:23
SPEAKER_0: And are people already experimenting with that?
24:25
I wish that we had better economic analyses of the questions you're asking but
24:31
the short answer is that we don't really have good empirical evidence
24:35
to answer your question precisely.
24:37
SPEAKER_3: What we do have are cross country comparative studies.
24:41
SPEAKER_3: And we can look at the different internets in the different countries and compare them.
24:46
SPEAKER_3: Because every other country outside the US doesn't have a Section 230.
24:49
And so all those other countries are examples of what does it look like in a world
24:54
without Section 230.
24:55
And I'm gonna summarize super efficiently just to say the one thing that stands out
25:00
when you do the cross-country comparisons is how little innovation there is
25:05
in other countries when it comes to enabling people to talk to each other.
25:09
SPEAKER_3: There are some homegrown user-generated content services in other countries.
25:14
SPEAKER_3: Usually they're a monopolist in that country.
25:17
They have been able to basically get the network effects to roll up that industry
25:21
and then they are able to survive.
25:23
SPEAKER_3: But you see very little innovation.
25:26
Almost all of the innovation when it comes to enabling people to talk
25:29
to each other online has come here in the US.
25:31
SPEAKER_3: Now that could be because of the First Amendment, not Section 230.
25:34
SPEAKER_3: We don't know.
25:35
SPEAKER_3: So reverse engineering all the different elements that could that contribute to that is tricky.
25:41
But I can say with confidence that the internet looks substantially worse
25:45
in other countries than it does here.
25:47
SPEAKER_3: And I think that Section 230 is a prime candidate for explaining why.
25:52
SPEAKER_0: You used the word worse.
25:54
SPEAKER_0: Worse in what ways?
25:55
SPEAKER_3: Well, there are fewer ways for people in those countries to express themselves.
25:59
And there are fewer ways for them to do so freely,
26:02
either in terms of financial access or in terms of free speech.
26:06
So if you look at the tools that people use to talk to each other
26:10
in foreign countries, a lot of them are run from here in the US.
26:14
I wanted to turn our attention to another topic that you've addressed in depth,
26:18
which is internet conduct.
26:20
SPEAKER_1: About five years ago you had an article about how to remedy that.
26:24
SPEAKER_1: If my recollection is correct, you had several dozen examples of how to do that.
26:29
I'm wondering if you would kind of detail that approach and is it essentially
26:34
a market based approach?
26:35
Is it that this is a form of kind of self regulation,
26:38
as opposed to anything that would reflect legal requirements or regulatory regimes?
26:44
SPEAKER_3: Yeah, thanks for asking about the paper.
26:46
SPEAKER_3: I actually I mean I love all my papers, kind of like I love all my children.
26:49
SPEAKER_1: Mm-hmm.
26:51
But that paper was a really special paper for me because I think I was able
26:54
to put together a set of explanations that I think was helpful to moving
26:59
the conversation forward.
27:01
SPEAKER_3: The problem I was trying to address in that paper is quite simple.
27:03
SPEAKER_3: We can think about a solution to any problematic online conduct or content as a zero or one.
27:11
SPEAKER_3: Either it exists or doesn't.
27:13
SPEAKER_3: And so that kind of binary thinking is like a light switch.
27:17
SPEAKER_3: So either we allow it or we don't.
27:19
And so what governments consistently do is that they look at a problem,
27:23
they see the light switch is on, they say let's turn off the light switch.
27:27
SPEAKER_3: Let's just turn it off all entirely.
27:30
And that binary thinking masks that there might be a whole bunch of options in
27:34
between the zero one,
27:36
the light being on,
27:36
light being off,
27:37
that might be a more helpful and nuanced way of trying to correct any of
27:42
the problems in the community.
27:45
SPEAKER_3: So think about it as a light switch dimmer as opposed to just an on off switch.
27:49
What are the options in that dimmer that might be available to remedy
27:53
or correct antisocial behavior or conduct without banning all activity entirely.
27:59
SPEAKER_3: So think about like the social media bans.
28:00
SPEAKER_3: Those are a very binary thing.
28:01
SPEAKER_3: They're like, just take the tool away from the kids.
28:04
Could there be some intermediate steps that are not take the tool away,
28:09
but still help address any of the problems need to be addressed?
28:12
So for example,
28:13
when you think about things like a particularly problematic type of content,
28:17
the standard approach is leave it up or take it down.
28:20
But there's a bunch of alternatives that are not those
28:23
that are also worthy considerations like make it less visible or prompt the author
28:28
to think about whether or not they want to make changes.
28:31
SPEAKER_3: These are things that don't ban the content.
28:34
They don't ensure that the content is going to remain in its current form,
28:38
but they might be better solutions to correct the problem.
28:42
SPEAKER_3: And so honestly the examples I gave in the paper are not that consequential.
28:46
But the key of knowing that there are intermediate stages between light on
28:50
and light off, gives us a chance to have that dialogue.
28:53
SPEAKER_3: And it really ought to be a challenge to regulators.
28:56
If they're going to try to address antisocial content or conduct online,
29:03
can they think more creatively than just ban it?
29:06
SPEAKER_3: Are there other options that might be helpful?
29:08
SPEAKER_3: But that's not what we're seeing today.
29:10
One of the one of the topics of the paper,
29:12
one of the areas of examination was financial consequences.
29:16
But that was within,
29:17
as I understand the paper,
29:18
within the context of the platform itself as opposed to any sort of cause of action
29:24
or any sort of fines imposed.
29:26
Is that a correct reading of the paper that you saw those measures in terms
29:30
of anti social content as something that the platform would of its own accord impose
29:36
as opposed to government imposing it or through some sort of private legal action.
29:41
What I was trying to do was to model out how
29:44
the user generated content services could use tools that might not be required by
29:51
the law, but might still help improve the overall community or discourse.
29:56
So in the context of the paper,
29:59
I was trying to show that one of the tools that the service in
30:03
the middle could impose financial consequences that are independent
30:07
of any consequences that might be imposed by the law.
30:10
SPEAKER_3: The law is in the background in that conversation.
30:12
It still exists and it still has a binding effect,
30:16
but the services could add their own layer of financial controls as a way of trying
30:21
to motivate people to do the right thing in their environment.
30:24
SPEAKER_0: Can you give an example of that, Eric?
30:26
SPEAKER_0: How would that work?
30:27
SPEAKER_0: Let's take YouTube for example: someone puts up torture porn.
30:31
SPEAKER_0: How can the platform impose an economic cost on the poster?
30:34
Well, the most obvious example,
30:36
and I think that it would actually apply in the circumstances you gave,
30:40
is that the services could demonetize the content.
30:43
So they're not removing the content,
30:45
they're simply saying we're not gonna let you get a financial benefit from that.
30:49
There could be scenarios where in fact not only do they demonetize the content,
30:53
but they impose an additional financial charge above and beyond that,
30:56
some kind of internal penalty, like a liquidated damages type principle.
31:00
If you're going to engage in anti social conduct,
31:03
we're gonna make you feel some financial consequence to us as well.
31:06
SPEAKER_3: But demonetization is the most obvious example and in fact it's a widely used tool.
31:10
You know,
31:10
I was gonna ask,
31:11
you have a very detailed and well reasoned and particular take on these issues.
31:18
I'm wondering if that was something that was from the start
31:20
or how your how your views have evolved over the course of your career.
31:25
Was everything kind of obvious to you in terms of first principles from the outset
31:30
or has there been some evolution?
31:32
SPEAKER_1: Are there things that you feel like you've markedly changed your mind on?
31:37
I don't think I've changed much,
31:39
and that's why I think I'm now more of an outlier,
31:42
because I think a lot of people around me have changed.
31:45
SPEAKER_3: I will mention a couple of things though, about my evolution in my thinking.
31:50
I think I'm more clear today than I might have been 30 years ago about
31:55
the critical role of the rule of law.
31:58
SPEAKER_3: That there's a certain order in which governments must function.
32:01
SPEAKER_3: And if governments don't function there, a lot of the other things don't matter.
32:05
SPEAKER_3: Like technology may not be able to fix that.
32:07
And so I don't think I was as clear about that 30 years ago,
32:11
but in the last 10 years I become very clear about how much we depend upon
32:16
a well functioning government in order for everything else to work.
32:19
The other thing I think I might have changed about is that in the 1990s,
32:25
I generally had a presumption of good faith about entrepreneurs trying
32:30
to spur user conversations.
32:32
SPEAKER_3: And so take General Magic as a fine example about that.
32:35
I had a general presumption that the people running General Magic,
32:39
even though they were in it for the money,
32:41
actually had some ethics and principles that were animating their decision making.
32:46
And I've become,
32:46
I think,
32:47
clearer that there are many companies today that don't have those ethics
32:52
and principles.
32:53
Maybe they never did,
32:54
maybe they have sacrificed them as part of the ordinary evolution of the company.
32:59
But I've seen more what I consider to be objectionable behavior from companies
33:05
that I would view as a lack of ethics or social awareness that I would have expected
33:12
from them.
33:12
And in particular,
33:13
I would point to the fact that the user content companies are not nearly
33:18
as committed to principles like free speech,
33:23
their free speech,
33:24
put their users' free speech aside for a moment,
33:26
but their free speech,
33:27
compared to things like the old newspaper model,
33:30
where the newspapers always understood the value of protecting their ability
33:35
to publish what they choose.
33:37
SPEAKER_3: The internet companies are not nearly as clear about that.
33:40
And so I just have seen a degradation of their willingness to stand
33:45
behind their principles or they've demonstrated they don't have them.
33:49
What have you seen change about your students in terms of their attitude
33:53
toward using tools like generative AI in their in their academic work?
33:58
Before I answer about generative AI,
34:00
let me just talk about the internet because I think it's an interesting evolution.
34:02
So I started teaching internet law in 1996,
34:05
and it was just a given when I taught the class in the 1990s that every student in
34:11
the room believed in the internet,
34:13
thought it was a good thing, and wanted it to succeed.
34:17
SPEAKER_3: And that is definitely not the case anymore.
34:20
My Gen Z students today taking internet law,
34:22
and I do surveys of them,
34:24
and I have for many years,
34:25
about half of them today,
34:26
if given the option to have the light switch to turn off the internet,
34:29
they would turn it off entirely.
34:31
That's how little they appreciate its role and how angry they are about its effects
34:36
on them or on society generally.
34:39
SPEAKER_3: So I'm no longer in sync with a majority of my class.
34:42
SPEAKER_3: I come in and I say, let's talk about all the good things.
34:44
SPEAKER_3: And my students are like, Why in the world will we do that?
34:47
Let's talk about all the problems,
34:48
how we're gonna blast all the bad guys on the internet, including the services.
34:52
SPEAKER_3: It's just like a major culture clash on that front.
34:55
SPEAKER_3: And one of the things I do is I try to explain why I'm such a supporter of the internet today.
34:59
And I think I do get some of the students to think more critically
35:03
about their negative attitudes.
35:05
SPEAKER_3: Some of them might even come along with me for the ride, but a bunch of them don't.
35:08
SPEAKER_3: Even at the end of the semester, I still haven't reached them on that.
35:11
What's interesting is I already have some students
35:13
who are already anti-generative AI before we even start talking.
35:17
SPEAKER_3: They come into the room with that.
35:19
SPEAKER_3: There's an entire segment of Gen Z that's anti-AI.
35:24
And even if they aren't anti-AI in the abstract,
35:27
I have a number of students who say,
35:29
I don't want to use it either because I think it's gonna short circuit my learning
35:33
or because of the environmental consequences.
35:36
SPEAKER_1: How do you use AI at this point?
35:39
Rarely, But if you're talking about something like the chatbots,
35:42
I use it for things like if I want to illustrate a principle
35:45
in my teaching materials or on my blog posts,
35:48
I will sometimes use generative AI to create an image that I had no artistic ability
35:53
to create on my own.
35:54
SPEAKER_3: I have used generative AI as kind of like what I consider the wisdom of the crowds.
35:58
So when I want to know what is the consensus view about a topic,
36:01
I can go to generative AI and say,
36:03
answer this question,
36:04
knowing that it's going to give me in all likelihood the Wisdom of Crowds-type
36:08
of stuff.
36:08
SPEAKER_3: I'll give you example.
36:09
I can ask Generative AI,
36:11
give me an idea of some of the tourist destinations
36:13
that meet my particular specifications for what I want on a trip.
36:17
SPEAKER_3: And I know I'm not getting the perfect answer.
36:20
I'm getting a consensus answer,
36:21
but that's actually quite helpful for me and can expose things I wouldn't think
36:24
about otherwise.
36:25
And I have tried to use generative AI for editing pieces to identify either things
36:30
that I'm missing or to help me fix any errors that I've made.
36:35
I wouldn't write something with generative AI,
36:38
but having written something,
36:39
I would use it as a basically the same way I might use an RA,
36:43
like hey, be a second set of eyes on this.
36:45
What I find is that generative AI doesn't really love my writing style
36:48
because I have all my little writing quirks and it wants to smooth all those out
36:53
and make it more homogenized.
36:55
SPEAKER_0: Yeah.
36:55
I have to give a lecture this afternoon at Stanford and I created a Word document
37:00
and I asked Copilot to turn it into a set of PowerPoint slides.
37:02
SPEAKER_0: And it did.
37:03
I said you can't change any of the words though,
37:06
because first I just said make slides and it did exactly what you said.
37:09
SPEAKER_0: It took all of the vigor and strength out of what I had written and homogenized it.
37:14
SPEAKER_3: It just automated a task.
37:16
SPEAKER_3: You could have done that yourself, but it sped things up.
37:19
SPEAKER_3: And if it speeds things up, that's a win.
37:21
SPEAKER_3: I'm gonna call that a win.
37:23
Before we get to the end of time,
37:25
which we're approaching,
37:27
in addition to thanking you for such thorough thinking and challenging a lot of
37:31
the notions that Mike and I arrived with,
37:34
I think there are at least two pieces of business.
37:37
SPEAKER_1: First of all, damn it, and I think that checks the box.
37:40
Eric,
37:41
we have a rule that I have to use profanity once and only once in every interview,
37:44
so fuck it.
37:45
I'm sorry,
37:45
if you had told me profanity was on the table,
37:46
we could have dropped a lot of F bombs along the way here.
37:47
SPEAKER_3: So sorry for the missed opportunity.
37:49
SPEAKER_1: Well, fuck, if I had known that.
37:51
SPEAKER_1: There we go.
37:52
In all seriousness,
37:53
not that I was kidding,
37:54
we also ask every guest what their bumper sticker is and
37:58
whether that's what would be on the back of your car or the through line
38:01
of your career,
38:02
or just a phrase that always tickles your fancy or resonates with you.
38:06
SPEAKER_1: So I put that question to you.
38:08
SPEAKER_1: What is Eric Goldman's bumper sticker?
38:11
SPEAKER_3: Literally gonna answer it with a bumper sticker because it's right here.
38:14
SPEAKER_3: It's on my desk.
38:15
SPEAKER_3: I look at it every day and I do this for my wife because I love her so much.
38:21
SPEAKER_3: It says, I love my wife.
38:22
SPEAKER_3: That's the bumper sticker that I have on my desk right now.
38:25
SPEAKER_3: I look at it every day.
38:25
SPEAKER_3: And it's true.
38:26
SPEAKER_1: Well, there you go.
38:28
SPEAKER_1: Well, fair enough.
38:29
SPEAKER_1: Mike, do you have concluding questions or thoughts?
38:31
SPEAKER_0: Are you doing any policy work trying to affect the outcome of potential regulations?
38:36
SPEAKER_0: Or is your writing the way you enter the fray?
38:39
SPEAKER_3: The short answer is that I don't do much policy work anymore.
38:42
SPEAKER_3: And the main reason why is because nobody wants to hear what I have to say.
38:46
When I come in and I'm talking about,
38:48
hey, there might be good aspects of the internet, like I just get blank stares.
38:51
SPEAKER_3: Like, okay, but that's not the conversation we want.
38:54
SPEAKER_3: We want to know how to shut down the entire internet.
38:57
SPEAKER_3: And you're not helping.
38:58
SPEAKER_3: My writing certainly is a part of the work that I do.
39:01
That's what it means in my mind to be an academic is to help shape the conversations
39:04
that I think are important.
39:06
SPEAKER_3: So I use my writing deliberately for that purpose.
39:11
One of my major milestones of life was helping create an organization called
39:15
the Trust and Safety Professional Association and a sister organization,
39:18
the Trust and Safety Foundation,
39:20
where we tried to put together the people who are actually doing the work so
39:23
that they can learn from each other,
39:25
talk about best practices,
39:27
and speak about the issues that they are uniquely positioned to opine about.
39:30
So, from my perspective,
39:31
rather than trying to directly engage the policymakers,
39:34
I have been working with this organization to help organize the community so
39:39
that we can bring more facts to the table that I hope will help inform
39:42
the conversation better.
39:44
SPEAKER_1: Fantastic work, Eric.
39:46
I guess I do have one more question,
39:47
which is my sense is your fears are not so much about AI,
39:50
but how society and government will respond to it.
39:54
SPEAKER_1: Is there a leading concern that you have as kind of this would be the nightmare outcome?
39:59
I already mentioned what I consider to be nightmare outcome
40:01
of governments dictating how AI outputs information in a way that is designed
40:07
to entrench their propaganda.
40:09
But I will also point out that I've been horrified to see
40:12
the government using generative AI to produce content that is just
40:18
the government lying to us directly.
40:20
They don't have to hack the chatbots,
40:21
they'll just lie to us to our face using AI generated material.
40:26
SPEAKER_3: I've been horrified by that.
40:27
And then all the stuff that DOGE was doing,
40:29
trying to use generative AI or other AI tools as a way
40:33
of basically doing government decision making.
40:35
SPEAKER_3: There might be a role for that, but in the hands of idiots, it is a weapon.
40:40
SPEAKER_3: And we saw how that looked in the DOGE example.
40:44
Well, this is before we started our three-way conversation on the record,
40:49
I used the phrase "surveillance capitalism," which made you bristle.
40:52
But to me one of the dystopian potentials of AI is its ability to refine targeting,
40:58
tracking, and shaping behavior.
41:00
SPEAKER_0: Why does the phrase surveillance capitalism more than annoy you?
41:03
SPEAKER_0: Because it seems to me the accurate description of what the platforms have enabled.
41:08
I start with the premise that advertisers and retailers have always wanted
41:12
to know more about their customers.
41:13
SPEAKER_3: This is not a new thing.
41:15
SPEAKER_3: This is decades or centuries old.
41:18
The only question is if we've seen a qualitative leap in the ability of
41:23
the advertisers or retailers to do so.
41:26
SPEAKER_3: But it's not a new issue at all.
41:28
And I go back to the example,
41:30
you know, you pay the "tourist tax" if you go shopping in a foreign country.
41:34
If you go one of the markets,
41:36
they instantly tell that you're American and they deal with you differently
41:39
and quote you differently.
41:40
SPEAKER_3: We could call that surveillance capitalism, or we could just call it capitalism.
41:44
SPEAKER_3: Like the adding the word surveillance there adds very little, if anything, to the equation.
41:48
SPEAKER_3: It actually just reinforces we're talking about capitalism.
41:53
SPEAKER_3: We might have problems with capitalism.
41:55
SPEAKER_3: I don't, I know many people who do.
41:57
SPEAKER_3: But adding the word surveillance to the front is just I think unenlightening.
42:02
SPEAKER_3: But when it comes to knowing more about customers, there's I think a couple of concerns.
42:08
One is that the seller,
42:10
whether it's the advertiser or the retailer,
42:12
can better manipulate the customer to reorder their preferences.
42:17
So their preferences that they brought to the table might change through
42:22
the engagement with the advertiser or retailer in a way
42:24
that actually subverts their true free will.
42:28
And the other is that the retailer or advertiser might be able to better determine
42:33
the reservation price of the customer in a way that would allow them to capture more
42:37
of that and put more money in their pocket to the exclusion of the customer.
42:40
SPEAKER_3: And I don't like either of those outcomes by the way.
42:42
SPEAKER_3: To me, both of those are defects in capitalism.
42:45
SPEAKER_3: They're not new to technology, just technology might make them more powerful or stronger.
42:51
SPEAKER_3: And in both cases, I ask the question, can that be addressed through additional competition?
42:57
So for example,
42:59
if we have tools that are trying to extract our very last dollar
43:03
by determining our reservation price and setting the price that way,
43:06
that sounds like an opportunity for others to come into the market and try
43:10
to undermine that,
43:11
to take a little less of the reservation price and still make a profit.
1:14:25
SPEAKER_0: Yeah, okay.
2:39:43
SPEAKER_0: I have far less faith in markets than you do.
5:54:55
SPEAKER_0: Most of our major social problems are market failures, like healthcare, for example.
12:53:13
SPEAKER_0: But that makes it clear.
14:51:19
SPEAKER_0: Appreciate that.
16:33:01
You know,
16:59:16
we were colleagues at Cooley together and it's great to be back in touch
22:24:03
with your brilliance.
24:23:48
SPEAKER_0: I really enjoy it.
25:54:01
SPEAKER_0: Thanks very much for being with us.