人工智能的真相（2023）（3）

So this was like a simple, a toy example.
虽然圣诞毛衣是一个无关紧要的问题
Christmas jumpers aren't really so much of a problem,
但在现实中可能出现
but there can be real life instances
人工智能无意间造成偏见的情况
of AI causing unintended bias.
还有个例子 有个用于简历筛选的
There's one example of AI software being trained
人工智能软件
to sift through job applications
它只会给特定的岗位挑选男性应聘者
and picking only men for certain roles.
因为该人工智能的训练数据是
And that's because the AI had been trained on successful
过去应聘成功的人
historical applicants in the past,
其中大部分都是男性
the majority of whom had all been men.
也有例子显示 一些人脸识别软件
And there are also cases of facial recognition software
在面对肤色较深的人时容易出错
incorrectly identifying people with darker skin tones.
这可能是因为该软件的训练数据是
And this could be maybe because the software was trained
肤色较浅的人群
on people with lighter skin tones.
大家都应该知道这个问题
Well, this is an issue that you all need to know about.
我们绝不能假设人工智能一定是对的
We must never assume that the AI is just right.
如果我们觉得电脑系统或人工智能系统的判断
We all need to be prepared to speak up
有失公正或存在偏见
if we feel that a computer system
我们都要准备好指出它的问题
in general or an AI system is being unfair or is being biased.
谢谢你 苏 让我们理解了这些风险 谢谢
Thank you, Sue, for helping us understand these risks. Thank you.
谢谢 谢谢大家
Thank you. Thanks everyone.
如今 定期举♥行♥的选举常常让人感到乏味
Now, elections come around with monotonous regularity,
每到大选期间 许多人会担心人工智能会被用于
and every time they do, there is a concern that AI might be used
向特定选民发送错误和虚假的信息
to target voters with misinformation and disinformation.
为了认识复杂的虚实网络
To help us make sense of the complex web of fact and fiction,
有请来自班戈大学的新闻学教授
please welcome, from the University of Bangor,
薇安·巴克尔
Professor of Journalism Vian Bakir.
欢迎 薇安
Welcome, Vian.
政♥治♥宣传 试图说服某人给某党派投票
So, political advertising, trying to persuade somebody to vote for this party or this party,
自从有了选举 政♥治♥宣传也一直存在
surely that's been around since elections were invented.
现在有什么变化吗 有什么新问题
So what's new now? Why... What's the new thing?
新问题就是 精准推送
Well, what's new now is micro-targeting.
在过去 如果你接收到一份信息
In the past, if you were shown a message,
那么这份信息同时也发送给了其他人
it was the same message to all of you at the same time.
而现在 有了人工智能的帮助 我们能将信息定制化
Now, with the help of AI, we can have a personalised message
在人们最容易被说服的时候
which is shown to you or groups like you at a time
发送给你或你这类的人
where you're perhaps most persuadable.
这就是不同 -好的
That's the difference. - Ok.
为什么这种推送对大选而言很重要呢
So why is targeting important in the run-up to an election?
对政党来说 这种精准投送能让他们
Well, targeting for a political party, it allows them
更好地动员目标选民
to mobilise their target voters,
动员通过常规媒体也许难以接触的人群
people that perhaps are hard to reach through conventional media.
年轻人其实就难以接触
Young people are actually very hard to reach.
但能通过社交媒体
But you can be targeted through social media
特别能与你们共情的信息
and with messages that really resonate with you as well,
与个人利益相关的信息 来实现精准推送
that speak to your personal interests.
这对于政党来说很有用
This is useful for a political party.
那么现在我们要进行一个小型选举活动
Now, we're going to have a little campaign now
更详细地调查一下这个问题
to investigate this issue in a little bit more detail.
我们要解决一个
We are going to address, folks,
重大议题 一个举世无双的
the big issue, the one and only issue,
终极议题
the eternal issue,
一个人类所面临的最重大的议题
the biggest issue that humanity has ever faced.
即...选猫 还是选狗
That is... Dogs or cats?
更喜欢猫的人请举手
OK, hands up if you prefer cats to dogs.
好的 很多人举了手
OK, quite a lot of you.
更喜欢狗的人请举手
Hands up if you prefer dogs to cats.
举手的人也很多
OK, fair number. OK.
都喜欢 或者不确定喜欢哪个的请举手
And hands up if you love them both equally or you're just unsure.
也有人举手
OK, we've got a few of you.
那么你们便是本次活动的摇摆选民
Now, you are going to be our floating voters
我们要在竞选时对你们精准推送
and we are going to target you in our campaign,
我们的虚假信息
our disinformation campaign,
来看看我们是否能够改变你的偏好
and we're going to see if we can change your preferences.
请观看屏幕上人工智能生成的视频
So let's have a look at our AI-generated campaign.
你以为你在遛狗 狗却在毁你人生
不养狗 不产便
我一看到狗 就觉得我的三明治有危险
When I see a dog, my sandwich is never really safe.
猫在各方面都更好
Cats are better in every way.
对狗说不
对狗说不
Just say no to dogs.
有请这位观众来说一下
OK, now I'm going to pick you.
你之前不确定选什么
You were unsure.
那我们是否成功改变了你的偏好
Did we manage to change your vote?
没有 没改变
No. No. Not really.
这毕竟是个人工智能生成的宣传
Well, it was an AI-generated campaign.
可能效果不是那么好
It wasn't really that great.
然而 但是
But, nevertheless,
我们是通过询问而知道各位的喜好
we found your preferences by asking you.
事实上 人工智能是从社媒上发现你的喜好
In reality, AI will find out your preferences from social media.
人工智能会查看你的社交媒体动态
It will look at your social media feed
还会查看你的数字足迹
and it will look at your digital footprint,
比如你网购了什么物品
what you buy online, for example.
这类资料收集并不罕见
This kind of profiling isn't unusual.
这类收集一直在进行 也并非总是恶意的
It's done constantly and it's not always malicious.
由此你才会看到类似的推荐 比如
It's how you see, for example, recommendations about what movies
观看什么影视剧 在音乐应用上听什么歌♥曲
or TV shows to watch and what to listen to on listening apps.
但针对政♥治♥方面的问题而言 这类精准推送
But when it comes to political questions, this kind of targeting
可能会造成巨大的影响
can have very, very big consequences.
下面请薇安详细介绍一下
So Vian, talk us through that.
我认为最严重的影响 最危险的事物
Well, I think the biggest consequence, the biggest danger
便是虚假信息 它是精准推送至你个人的
is disinformation, basically, that's been targeted to you.
虚假信息是带有故意欺骗性质的信息
So, disinformation is deliberately deceptive information.
如果它针对的是你的恐惧 希望 和利益
And if it's targeted to your fears, your hopes, your interests,
因为是精准推送 所以只有你见过
and only you see it because it's micro-targeted,
那么你如何知道
how do you know
它是真是假 社会又要怎样纠正你
whether it's false or true, and how can society correct you?
记者不知道你看过这些内容
Journalists don't know you've seen it,
因为这是精准推送至你个人的
cos it's micro targeted to you,
记者无法为你核实
so they can't fact-check it for you.
政客也不知道你看过这些内容
Politicians don't know you've seen it,
至少另一派的政客不知道
at least the opposition politician doesn't,
他们也无法证伪
and so they can't rebut it.
因此 你个人接收到的虚假信息
So there's no correction, basically, for your disinformation
基本上是无法进行纠正的
that you personally have been exposed to,
也没人真正能了解到
and no-one can really know how much disinformation
你到底接触了多少虚假信息 这也是个问题
you've been exposed to, basically, and that's a problem.
所以我想 这其中是否很多都涉及人工智能
So I'm guessing, is AI in a lot of this,
人工智能是怎么参与到竞选之中的
where does AI come into it?
人工智能随处可见
AI is all over the place here.
首先 它存在于数据中 它分♥析♥海量数据
AI is in the data, first of all, making sense of that huge amount
比如社交媒体平台上你的数据
of data that social media platforms have on you, for instance,
这是一批庞大的数据
which is a huge amount.
随后从上述海量的数据中了解你自身
It then learns about you from that data.
了解你的兴趣爱好 你的希望和恐惧
It learns about your interests, your hopes, your fears.
人工智能也被用作对你进行精准推送
And then it's also used to target you
推送的信息与你的希望
with the message that resonates
恐惧和利益相契合
with your hopes, fears and interests.
人工智能知道向你展示什么 无论是政♥治♥广♥告♥
You know, it knows what to show you, whether it's a political ad
还是你社交媒体动态中置顶展示的内容
or some content at the top of your social media feed.
人工智能甚至还能创造信息内容
And then finally, it could even create the message as well.
它能打磨信息 或者进行深度伪造
It can hone the message or it can create deepfakes.
所以大家成为选民时
OK, so as you become voters,
应当注意这些问题
you need to be aware of all of these issues.
感谢薇安帮助我们理解以上问题
Thank you, Vian, for helping us understand them.
我们能相信我们在网上看到的人像
So... Can we trust that the images of the people that we see online
是真实的吗
are actually real?
研究表明人们并不擅长识别深度伪造的内容
Research shows that we aren't very good at recognising deepfakes.
我们通常是仅看表面 因此不知道
We take things at face value, literally, and that leaves us
什么是真的 什么又是假的
confused as to what's true and what's not.
我们来看看这个吧
So, have a look at this.
对狗说不
Just say no to dogs.