0:00:01.548,0:00:04.064
No matter who you are or where you live,

0:00:04.088,0:00:06.444
I'm guessing that you have[br]at least one relative

0:00:06.468,0:00:08.802
that likes to forward those emails.

0:00:09.206,0:00:11.106
You know the ones I'm talking about --

0:00:11.130,0:00:13.984
the ones with dubious claims[br]or conspiracy videos.

0:00:14.315,0:00:16.983
And you've probably[br]already muted them on Facebook

0:00:17.007,0:00:19.355
for sharing social posts like this one.

0:00:19.379,0:00:20.783
It's an image of a banana

0:00:20.807,0:00:23.474
with a strange red cross[br]running through the center.

0:00:23.498,0:00:25.637
And the text around it is warning people

0:00:25.661,0:00:27.819
not to eat fruits that look like this,

0:00:27.843,0:00:29.871
suggesting they've been[br]injected with blood

0:00:29.895,0:00:32.025
contaminated with the HIV virus.

0:00:32.049,0:00:34.652
And the social share message[br]above it simply says,

0:00:34.676,0:00:36.857
"Please forward to save lives."

0:00:37.672,0:00:40.966
Now, fact-checkers have been debunking[br]this one for years,

0:00:40.990,0:00:43.799
but it's one of those rumors[br]that just won't die.

0:00:43.823,0:00:45.094
A zombie rumor.

0:00:45.513,0:00:47.606
And, of course, it's entirely false.

0:00:48.180,0:00:51.139
It might be tempting to laugh[br]at an example like this, to say,

0:00:51.163,0:00:53.047
"Well, who would believe this, anyway?"

0:00:53.419,0:00:55.045
But the reason it's a zombie rumor

0:00:55.069,0:00:58.958
is because it taps into people's[br]deepest fears about their own safety

0:00:58.982,0:01:01.157
and that of the people they love.

0:01:01.783,0:01:05.056
And if you spend as enough time[br]as I have looking at misinformation,

0:01:05.080,0:01:07.500
you know that this is just[br]one example of many

0:01:07.524,0:01:10.571
that taps into people's deepest[br]fears and vulnerabilities.

0:01:11.214,0:01:15.583
Every day, across the world,[br]we see scores of new memes on Instagram

0:01:15.607,0:01:18.646
encouraging parents[br]not to vaccinate their children.

0:01:18.670,0:01:23.202
We see new videos on YouTube[br]explaining that climate change is a hoax.

0:01:23.226,0:01:27.528
And across all platforms, we see[br]endless posts designed to demonize others

0:01:27.552,0:01:31.053
on the basis of their race,[br]religion or sexuality.

0:01:32.314,0:01:35.344
Welcome to one of the central[br]challenges of our time.

0:01:35.647,0:01:39.672
How can we maintain an internet[br]with freedom of expression at the core,

0:01:39.696,0:01:42.999
while also ensuring that the content[br]that's being disseminated

0:01:43.023,0:01:46.909
doesn't cause irreparable harms[br]to our democracies, our communities

0:01:46.933,0:01:49.171
and to our physical and mental well-being?

0:01:49.998,0:01:52.085
Because we live in the information age,

0:01:52.109,0:01:55.656
yet the central currency[br]upon which we all depend -- information --

0:01:55.680,0:01:58.037
is no longer deemed entirely trustworthy

0:01:58.061,0:02:00.389
and, at times, can appear[br]downright dangerous.

0:02:00.811,0:02:04.748
This is thanks in part to the runaway[br]growth of social sharing platforms

0:02:04.772,0:02:06.414
that allow us to scroll through,

0:02:06.438,0:02:08.660
where lies and facts sit side by side,

0:02:08.684,0:02:11.755
but with none of the traditional[br]signals of trustworthiness.

0:02:12.268,0:02:15.887
And goodness -- our language around this[br]is horribly muddled.

0:02:15.911,0:02:19.014
People are still obsessed[br]with the phrase "fake news,"

0:02:19.038,0:02:21.569
despite the fact that[br]it's extraordinarily unhelpful

0:02:21.593,0:02:25.053
and used to describe a number of things[br]that are actually very different:

0:02:25.077,0:02:28.463
lies, rumors, hoaxes,[br]conspiracies, propaganda.

0:02:28.911,0:02:31.823
And I really wish[br]we could stop using a phrase

0:02:31.847,0:02:34.709
that's been co-opted by politicians[br]right around the world,

0:02:34.733,0:02:36.204
from the left and the right,

0:02:36.228,0:02:39.450
used as a weapon to attack[br]a free and independent press.

0:02:40.307,0:02:45.009
(Applause)

0:02:45.033,0:02:48.495
Because we need our professional[br]news media now more than ever.

0:02:48.882,0:02:52.255
And besides, most of this content[br]doesn't even masquerade as news.

0:02:52.279,0:02:54.921
It's memes, videos, social posts.

0:02:54.945,0:02:58.398
And most of it is not fake;[br]it's misleading.

0:02:58.422,0:03:01.437
We tend to fixate on what's true or false.

0:03:01.461,0:03:05.493
But the biggest concern is actually[br]the weaponization of context.

0:03:06.855,0:03:08.823
Because the most effective disinformation

0:03:08.847,0:03:11.895
has always been that[br]which has a kernel of truth to it.

0:03:11.919,0:03:15.395
Let's take this example[br]from London, from March 2017,

0:03:15.419,0:03:16.959
a tweet that circulated widely

0:03:16.983,0:03:20.570
in the aftermath of a terrorist incident[br]on Westminster Bridge.

0:03:20.594,0:03:23.022
This is a genuine image, not fake.

0:03:23.046,0:03:26.215
The woman who appears in the photograph[br]was interviewed afterwards,

0:03:26.239,0:03:28.648
and she explained that[br]she was utterly traumatized.

0:03:28.672,0:03:30.410
She was on the phone to a loved one,

0:03:30.434,0:03:33.052
and she wasn't looking[br]at the victim out of respect.

0:03:33.076,0:03:37.036
But it still was circulated widely[br]with this Islamophobic framing,

0:03:37.060,0:03:40.106
with multiple hashtags,[br]including: #BanIslam.

0:03:40.425,0:03:42.823
Now, if you worked at Twitter,[br]what would you do?

0:03:42.847,0:03:45.409
Would you take that down,[br]or would you leave it up?

0:03:46.553,0:03:49.982
My gut reaction, my emotional reaction,[br]is to take this down.

0:03:50.006,0:03:52.148
I hate the framing of this image.

0:03:52.585,0:03:54.973
But freedom of expression[br]is a human right,

0:03:54.997,0:03:58.222
and if we start taking down speech[br]that makes us feel uncomfortable,

0:03:58.246,0:03:59.476
we're in trouble.

0:03:59.500,0:04:01.794
And this might look like a clear-cut case,

0:04:01.818,0:04:03.516
but, actually, most speech isn't.

0:04:03.540,0:04:05.976
These lines are incredibly[br]difficult to draw.

0:04:06.000,0:04:08.281
What's a well-meaning[br]decision by one person

0:04:08.305,0:04:10.382
is outright censorship to the next.

0:04:10.759,0:04:13.688
What we now know is that[br]this account, Texas Lone Star,

0:04:13.712,0:04:16.942
was part of a wider Russian[br]disinformation campaign,

0:04:16.966,0:04:19.117
one that has since been taken down.

0:04:19.141,0:04:20.704
Would that change your view?

0:04:21.322,0:04:22.481
It would mine,

0:04:22.505,0:04:24.806
because now it's a case[br]of a coordinated campaign

0:04:24.830,0:04:26.045
to sow discord.

0:04:26.069,0:04:28.030
And for those of you who'd like to think

0:04:28.054,0:04:30.885
that artificial intelligence[br]will solve all of our problems,

0:04:30.909,0:04:33.134
I think we can agree[br]that we're a long way away

0:04:33.158,0:04:35.745
from AI that's able to make sense[br]of posts like this.

0:04:36.856,0:04:39.363
So I'd like to explain[br]three interlocking issues

0:04:39.387,0:04:41.760
that make this so complex

0:04:41.784,0:04:44.906
and then think about some ways[br]we can consider these challenges.

0:04:45.348,0:04:49.238
First, we just don't have[br]a rational relationship to information,

0:04:49.262,0:04:50.730
we have an emotional one.

0:04:50.754,0:04:54.548
It's just not true that more facts[br]will make everything OK,

0:04:54.572,0:04:57.672
because the algorithms that determine[br]what content we see,

0:04:57.696,0:05:00.823
well, they're designed to reward[br]our emotional responses.

0:05:00.847,0:05:02.228
And when we're fearful,

0:05:02.252,0:05:05.426
oversimplified narratives,[br]conspiratorial explanations

0:05:05.450,0:05:08.868
and language that demonizes others[br]is far more effective.

0:05:09.538,0:05:11.412
And besides, many of these companies,

0:05:11.436,0:05:13.982
their business model[br]is attached to attention,

0:05:14.006,0:05:17.696
which means these algorithms[br]will always be skewed towards emotion.

0:05:18.371,0:05:22.669
Second, most of the speech[br]I'm talking about here is legal.

0:05:23.081,0:05:24.527
It would be a different matter

0:05:24.551,0:05:26.892
if I was talking about[br]child sexual abuse imagery

0:05:26.916,0:05:28.843
or content that incites violence.

0:05:28.867,0:05:32.137
It can be perfectly legal[br]to post an outright lie.

0:05:33.130,0:05:37.164
But people keep talking about taking down[br]"problematic" or "harmful" content,

0:05:37.188,0:05:39.797
but with no clear definition[br]of what they mean by that,

0:05:39.821,0:05:41.085
including Mark Zuckerberg,

0:05:41.109,0:05:44.521
who recently called for global[br]regulation to moderate speech.

0:05:44.870,0:05:47.085
And my concern is that[br]we're seeing governments

0:05:47.109,0:05:48.401
right around the world

0:05:48.425,0:05:51.101
rolling out hasty policy decisions

0:05:51.125,0:05:53.871
that might actually trigger[br]much more serious consequences

0:05:53.895,0:05:55.609
when it comes to our speech.

0:05:56.006,0:05:59.712
And even if we could decide[br]which speech to take up or take down,

0:05:59.736,0:06:01.910
we've never had so much speech.

0:06:01.934,0:06:04.065
Every second, millions[br]of pieces of content

0:06:04.089,0:06:06.196
are uploaded by people[br]right around the world

0:06:06.220,0:06:07.388
in different languages,

0:06:07.412,0:06:10.180
drawing on thousands[br]of different cultural contexts.

0:06:10.204,0:06:12.736
We've simply never had[br]effective mechanisms

0:06:12.760,0:06:14.498
to moderate speech at this scale,

0:06:14.522,0:06:17.323
whether powered by humans[br]or by technology.

0:06:18.284,0:06:22.228
And third, these companies --[br]Google, Twitter, Facebook, WhatsApp --

0:06:22.252,0:06:25.093
they're part of a wider[br]information ecosystem.

0:06:25.117,0:06:28.469
We like to lay all the blame[br]at their feet, but the truth is,

0:06:28.493,0:06:32.323
the mass media and elected officials[br]can also play an equal role

0:06:32.347,0:06:35.260
in amplifying rumors and conspiracies[br]when they want to.

0:06:35.800,0:06:40.744
As can we, when we mindlessly forward[br]divisive or misleading content

0:06:40.768,0:06:42.053
without trying.

0:06:42.077,0:06:43.877
We're adding to the pollution.

0:06:45.236,0:06:47.854
I know we're all looking for an easy fix.

0:06:47.878,0:06:49.545
But there just isn't one.

0:06:49.950,0:06:54.395
Any solution will have to be rolled out[br]at a massive scale, internet scale,

0:06:54.419,0:06:57.680
and yes, the platforms,[br]they're used to operating at that level.

0:06:57.704,0:07:01.176
But can and should we allow them[br]to fix these problems?

0:07:01.668,0:07:02.900
They're certainly trying.

0:07:02.924,0:07:07.010
But most of us would agree that, actually,[br]we don't want global corporations

0:07:07.034,0:07:09.366
to be the guardians of truth[br]and fairness online.

0:07:09.390,0:07:11.927
And I also think the platforms[br]would agree with that.

0:07:12.257,0:07:15.138
And at the moment,[br]they're marking their own homework.

0:07:15.162,0:07:16.360
They like to tell us

0:07:16.384,0:07:18.963
that the interventions[br]they're rolling out are working,

0:07:18.987,0:07:21.527
but because they write[br]their own transparency reports,

0:07:21.551,0:07:25.169
there's no way for us to independently[br]verify what's actually happening.

0:07:26.431,0:07:29.773
(Applause)

0:07:29.797,0:07:32.749
And let's also be clear[br]that most of the changes we see

0:07:32.773,0:07:35.767
only happen after journalists[br]undertake an investigation

0:07:35.791,0:07:37.402
and find evidence of bias

0:07:37.426,0:07:40.255
or content that breaks[br]their community guidelines.

0:07:40.815,0:07:45.410
So yes, these companies have to play[br]a really important role in this process,

0:07:45.434,0:07:46.994
but they can't control it.

0:07:47.855,0:07:49.373
So what about governments?

0:07:49.863,0:07:52.959
Many people believe[br]that global regulation is our last hope

0:07:52.983,0:07:55.863
in terms of cleaning up[br]our information ecosystem.

0:07:55.887,0:07:59.053
But what I see are lawmakers[br]who are struggling to keep up to date

0:07:59.077,0:08:01.418
with the rapid changes in technology.

0:08:01.442,0:08:03.346
And worse, they're working in the dark,

0:08:03.370,0:08:05.191
because they don't have access to data

0:08:05.215,0:08:07.865
to understand what's happening[br]on these platforms.

0:08:08.260,0:08:11.331
And anyway, which governments[br]would we trust to do this?

0:08:11.355,0:08:14.125
We need a global response,[br]not a national one.

0:08:15.419,0:08:17.696
So the missing link is us.

0:08:17.720,0:08:20.843
It's those people who use[br]these technologies every day.

0:08:21.260,0:08:25.851
Can we design a new infrastructure[br]to support quality information?

0:08:26.371,0:08:27.601
Well, I believe we can,

0:08:27.625,0:08:30.982
and I've got a few ideas about[br]what we might be able to actually do.

0:08:31.006,0:08:34.109
So firstly, if we're serious[br]about bringing the public into this,

0:08:34.133,0:08:36.514
can we take some inspiration[br]from Wikipedia?

0:08:36.538,0:08:38.362
They've shown us what's possible.

0:08:38.386,0:08:39.537
Yes, it's not perfect,

0:08:39.561,0:08:42.195
but they've demonstrated[br]that with the right structures,

0:08:42.219,0:08:44.854
with a global outlook[br]and lots and lots of transparency,

0:08:44.878,0:08:47.974
you can build something[br]that will earn the trust of most people.

0:08:47.998,0:08:51.160
Because we have to find a way[br]to tap into the collective wisdom

0:08:51.184,0:08:53.493
and experience of all users.

0:08:53.517,0:08:56.163
This is particularly the case[br]for women, people of color

0:08:56.187,0:08:57.533
and underrepresented groups.

0:08:57.557,0:08:58.723
Because guess what?

0:08:58.747,0:09:01.482
They are experts when it comes[br]to hate and disinformation,

0:09:01.506,0:09:05.022
because they have been the targets[br]of these campaigns for so long.

0:09:05.046,0:09:07.396
And over the years,[br]they've been raising flags,

0:09:07.420,0:09:09.085
and they haven't been listened to.

0:09:09.109,0:09:10.389
This has got to change.

0:09:10.807,0:09:15.133
So could we build a Wikipedia for trust?

0:09:15.157,0:09:19.346
Could we find a way that users[br]can actually provide insights?

0:09:19.370,0:09:23.067
They could offer insights around[br]difficult content-moderation decisions.

0:09:23.091,0:09:24.554
They could provide feedback

0:09:24.578,0:09:27.619
when platforms decide[br]they want to roll out new changes.

0:09:28.241,0:09:32.403
Second, people's experiences[br]with the information is personalized.

0:09:32.427,0:09:35.070
My Facebook news feed[br]is very different to yours.

0:09:35.094,0:09:37.839
Your YouTube recommendations[br]are very different to mine.

0:09:37.863,0:09:40.355
That makes it impossible for us[br]to actually examine

0:09:40.379,0:09:42.402
what information people are seeing.

0:09:42.815,0:09:44.204
So could we imagine

0:09:44.228,0:09:49.006
developing some kind of centralized[br]open repository for anonymized data,

0:09:49.030,0:09:51.894
with privacy and ethical[br]concerns built in?

0:09:52.220,0:09:53.998
Because imagine what we would learn

0:09:54.022,0:09:57.283
if we built out a global network[br]of concerned citizens

0:09:57.307,0:10:00.601
who wanted to donate[br]their social data to science.

0:10:01.141,0:10:02.863
Because we actually know very little

0:10:02.887,0:10:05.768
about the long-term consequences[br]of hate and disinformation

0:10:05.792,0:10:07.767
on people's attitudes and behaviors.

0:10:08.236,0:10:09.403
And what we do know,

0:10:09.427,0:10:11.620
most of that has been[br]carried out in the US,

0:10:11.644,0:10:14.025
despite the fact that[br]this is a global problem.

0:10:14.049,0:10:15.684
We need to work on that, too.

0:10:16.192,0:10:17.342
And third,

0:10:17.366,0:10:19.676
can we find a way to connect the dots?

0:10:19.700,0:10:23.138
No one sector, let alone nonprofit,[br]start-up or government,

0:10:23.162,0:10:24.584
is going to solve this.

0:10:24.608,0:10:27.172
But there are very smart people[br]right around the world

0:10:27.196,0:10:28.577
working on these challenges,

0:10:28.601,0:10:32.177
from newsrooms, civil society,[br]academia, activist groups.

0:10:32.201,0:10:34.099
And you can see some of them here.

0:10:34.123,0:10:37.050
Some are building out indicators[br]of content credibility.

0:10:37.074,0:10:38.320
Others are fact-checking,

0:10:38.344,0:10:41.935
so that false claims, videos and images[br]can be down-ranked by the platforms.

0:10:41.959,0:10:44.172
A nonprofit I helped[br]to found, First Draft,

0:10:44.196,0:10:47.164
is working with normally competitive[br]newsrooms around the world

0:10:47.188,0:10:50.691
to help them build out investigative,[br]collaborative programs.

0:10:51.231,0:10:53.540
And Danny Hillis, a software architect,

0:10:53.564,0:10:55.945
is designing a new system[br]called The Underlay,

0:10:55.969,0:10:58.744
which will be a record[br]of all public statements of fact

0:10:58.768,0:11:00.097
connected to their sources,

0:11:00.121,0:11:03.775
so that people and algorithms[br]can better judge what is credible.

0:11:04.800,0:11:08.156
And educators around the world[br]are testing different techniques

0:11:08.180,0:11:11.664
for finding ways to make people[br]critical of the content they consume.

0:11:12.633,0:11:15.774
All of these efforts are wonderful,[br]but they're working in silos,

0:11:15.798,0:11:18.478
and many of them are woefully underfunded.

0:11:18.502,0:11:20.555
There are also hundreds[br]of very smart people

0:11:20.579,0:11:22.231
working inside these companies,

0:11:22.255,0:11:24.580
but again, these efforts[br]can feel disjointed,

0:11:24.604,0:11:28.541
because they're actually developing[br]different solutions to the same problems.

0:11:29.205,0:11:31.474
How can we find a way[br]to bring people together

0:11:31.498,0:11:34.776
in one physical location[br]for days or weeks at a time,

0:11:34.800,0:11:37.196
so they can actually tackle[br]these problems together

0:11:37.220,0:11:39.038
but from their different perspectives?

0:11:39.062,0:11:40.402
So can we do this?

0:11:40.426,0:11:43.665
Can we build out a coordinated,[br]ambitious response,

0:11:43.689,0:11:47.398
one that matches the scale[br]and the complexity of the problem?

0:11:47.819,0:11:49.192
I really think we can.

0:11:49.216,0:11:52.175
Together, let's rebuild[br]our information commons.

0:11:52.819,0:11:54.009
Thank you.

0:11:54.033,0:11:57.761
(Applause)