想当总统？有AI冇问题

The Right Person For The Job

正确的人，正确的事

For George R. R. Martin, selecting a ruler is easy. All he has to do is to write it down, and anyone can be sitting on the Iron Throne. It was a little bit more complicated for Barack Obama, but he discovered a person who could do something similar. Obama hired a machine learning expert to his campaign team: Rayid Ghani. Ghani used analytical tools to gather voter data from social media, as well as other sources, in order to predict how likely individual voters were to support Obama, whether they could easily be persuaded into voting for someone else, and how likely they were to actually be at the polling stations come election day (Domingos 2017). His team ran 66,000 simulations every night, and based on the results, they knew exactly what doors to knock, who to call, and what to say.

对于乔治•马丁（George R. R. Martin，美剧《权力的游戏》原著作者）而言，选谁当国王都很容易。只要他愿意写，谁都可以坐上铁王座（王位的象征）。类似的事情对奥巴马来说有点复杂，不过，最终他还是找到了合适的人选。奥巴马为他的竞选团队雇佣了一名机器学习专家：Rayid Ghani。Ghani从社交媒体和其他的渠道收集选民的数据；使用工具对这些数据进行汇总分析后，他们可以预测：个体选民支持奥巴马的可能性有多大；个体选民是否会被轻易地说服该投给其他人；在选举日，个体选民有多大的可能性会到投票站投票。他的团队每天要进行六万多次的模拟；根据模拟的结果，他们可以获知上门拜访谁，打电话给谁，以及说些什么。

The Romney team could only watch. They could tell that the Obama team put up campaigns in very specific areas, but they didn’t know why. They could tell that the Obama team was knocking on doors in very specific neighborhoods, but they didn’t know why. And they could tell that the Obama team was calling specific voters, but again, they didn’t know why. While the Romney team was asking people how they would vote, the Obama team was collecting answers without having to bother to ask.

（对于这一切）罗姆尼的团队只能傻眼看着。他们知道奥巴马团队在一些特殊的区域举办了竞选活动，但是他们不知道为什么是这些区域。他们知道奥巴马团队在一些街区进行了上门拜访，但他们不知道为什么会在这些街区。他们知道奥巴马团队给一些选定的选民打了电话，但他们还是不知道为什么会是这些人。当罗姆尼的团队（不厌其烦地）询问人们将会如何投票时，奥巴马的团队却正在收集答案，他们根本不必费心去询问（人们的意见）。

And the Obama team won.

最终，奥巴马赢了。

Cleisthenes first suggesting democracy to the Athenian assembly (508 BC, colorized).

I Know Who You Are

我知道你是谁

At this point, you’re probably well aware of how companies are studying your data and recommending products and services based on who you are, and why wouldn’t politicians do the same? There are countless points of data that marketers can use to target you on social media or elsewhere, in order to convince you to buy a certain product. Such variables include your search history, your shopping behavior, your location, your age, your job, what you share, what you write, what you drink, if you smoke, if you are extroverted or introverted, if you have many or few friends, and so on. Roughly 300 data points are what are needed to know exactly who you are. Cambridge Analytica, the organization that secured victories for both Donald Trump and the Brexit Leave campaign, has created 220 million personality profiles for adults in the United States, using up to 5,000 data points (Anderson 2017).

你可能了解有些公司会研究你的数据，并根据你的信息推荐相应的产品和服务。所以为什么政治家不能这样做呢？为了说服你购买某种产品，营销人员可以使用社交媒体或者其他平台上的大量数据在定位你的身份（用户画像）。这些数据包括：你的搜索历史、购物行为、位置、年龄、分享内容、原创内容，你是否吸烟喝酒，你是外向型性格还是内向型性格，你的朋友是多还是少等等。一般来说，只需要有300条信息就能够准确了解你的身份。而剑桥分析（一家曾经帮助特朗普赢得美国大选及英国脱欧运动取得胜利的数据分析公司）为2.2亿美国成年人构建了个人（画像）信息，在这一过程中它使用了多达5000个数据点。

Social media is the best place to collect data on people, and likewise the best place to start targeting users with advertisements. The data collected and the advertisements presented on social media are done on an individual level, meaning the ads that two neighbors see are wildly different, despite the fact that they geographically live next to each other. Every single individual gets advertisements made just for them. The graphics of the advertisement can differ depending on what you are more likely to click on. Articles can be written with different styles: some comedic and others serious, some use logical arguments while others use emotional arguments, and some use a simple language whereas others use a complex language.

社交媒体是收集个人信息的最佳场所，它同时也是推送个性化广告的最佳场所。数据的收集和广告的展示都是在个人层面上完成的，这也就意味着两个邻居看到的广告可以完全不同，尽管他们在地理位置上是紧邻的。每个人看到的广告都是最适合自己的。广告的图像可能会有所不同，其具体内容取决于用户更可能点击哪种内容。文章也可以拥有不同的风格：一些偏喜剧，另一些则偏严肃；一些使用逻辑论证，另一些则使用情感论据；一些使用简单的语言，另一些则使用复杂的文字。

If the individual is already strongly supporting your party, you can mention positive news relating to the candidate, as an encouragement to remember to go vote. On the other hand, if the individual is strongly supporting your opponent(s), and may be found to be incredibly difficult to persuade to your side, you can instead create a divide in your opponent’s party by highlighting other, less popular candidates from the opposing party, or even attempt to persuade them to vote less, by highlighting mistakes made by their preferred leader. Your political opponents will never even know what your advertisements look like, as you are in control over who sees them and who doesn’t. That is the incredible power of modern propaganda.

如果选民已经强烈支持你的党派，你可以投放一些和该候选人的正面新闻，以鼓励他去投票。如果选民强烈支持你的对手，而且他也很难被说服改变意见，你可以（投放广告）突显对方党派中不那么受重视的候选人，从而为竞争对手制造绊子，减少其党内支持度；你甚至可以（投放广告）突显竞争对手所犯的错误，从而降低支持者参与投票的热情。你的政治对手永远不会知道你的广告长什么样子，因为你在背后控制了谁可以看到，谁又看不到。这就是现代（政治）宣传不可以思议的力量。

Use your human resources selectively. Most advertising platforms provide you with tools that allow you to examine conversion rates. Use the data to determine where to send your volunteers to the locations where they will have the biggest impact. For Obama, this was key. If you do have volunteers on your side, make the best use of them.

请有针对性的使用你的人力资源。大多数广告平台都提供了相应的工具好让你检查转化率。利用这些数据，将你的志愿者送到他们能发挥最大影响力的地方。对于奥巴马来说，这就是获胜的关键。如果你确实有志愿者资源的话，请充分利用他们。

This system of targeted advertisements also allows politicians to make inconsistent promises — or lies if you will — that are displayed depending on the individual in question. If you want to tread down this path of lies, however, you must do so very, very carefully. As if collecting data on citizens and manipulation opinions through microtargeting is not ethically bad enough, spreading untruthful information could be the tipping point.

这种个性化的广告系统还允许政治家们对个人做出各不相同的承诺。当然，如果你愿意的话，撒谎也是可以的。但是，如果你真的要沿着这条谎言之路走下去，你必须得非常小心谨慎。毕竟目前来看，收集公民数据和操纵舆论意见在道德上还没有那么糟糕，但是传播虚假信息可能是（舆论的）转折点。

A Network of Lies

网络欺骗

Bots, or fake social media profiles, are robots whose purpose is to spread political propaganda. Today, social media contains massive networks of bots. Once a person has liked a post on social media, a bot will follow them around the internet, providing them with targeted advertisements on other, seemingly unrelated, websites as well. A would-be president can use bots to quickly spread their propaganda.

机器人，或者说拥有虚假社交媒体资料的机器人，它们存在的目的就是为了政治宣传。而如今，社交媒体上存在着大量类似的机器人。如果一个人对一个帖子点了赞或者喜欢，那么这些机器人就会开始关注他，并在一些看似无关的网站上向他展示针对性的广告。这样，一位总统候选人就可以使用机器人来迅速传播他的政治理念。

Indeed, bots are an excellent way of spreading news, normally fake such, by making the news seem popular enough that they appear trustworthy to humans. When a falsified news article is first uploaded, it is bombarded with likes and shares from bots, until it appears trustworthy enough that real users also begin interacting with it. Ultimately, the bigger amount of shares and likes come from human users. The bots just need to kickstart the news.

事实上，机器人是传播新闻——通常都是假新闻——的绝佳方式。它们可以增加新闻的流行度，以至于这些新闻在人看来似乎是值得信赖的。一篇假新闻被伪造上传之后，机器人会对其点赞分享；当假新闻的热度达到一定程度之后，它看起来就像是真的一样，此时真实的用户开始参与到互动中。最终，该新闻大部分的点赞和分享都将来自于真实用户。机器人只需要启动新闻（的传播）就够了。

Give The People What They Want

授之以欲

The techniques described above microtargets individuals and very specific groups of people, but naturally, machine learning-tools can also be used to gather a consensus. In order to make popular decisions and establish yourself as a benign politician, you can use AI to learn what your votes want, without having to bother them with asking questions. Provided the machine learning-models (the algorithms) are good enough, you can learn what the majority wants at any point in time. Calling for long and expensive polls for every single issue in the country is impossible. AI enables you to simply run a query, as many times as you want, whenever you want, and gain the consensus on any given issue.

尽管上述技术可以构建个人和群体的画像，但本质上来说，机器学习工具也可以用于收集人群共识。为了做出受欢迎的决策，并将自己的形象设定为好好政治家，你不必使用问题来打扰支持者，你可以使用AI来了解他们的需求。如果你使用的机器学习模型（算法）足够好，那么你可以随时了解大多数人想要的东西。在这个国家（美国），想要对每一个问题都进行耗时长且昂贵的民意调查是不可能的。但AI却可以让你随时随地根据需求进行简单的查询，这样你就可以在任何给定问题上获得共识。

I Don’t Need You

没你没问题

At this point, you may be thinking to yourself: Jacob, this is interesting and all, but does AI really determine political victories, or have I been clickbaited? There are countless of other factors that go into voter decisions, and besides, not every election is that close anyway. Plus, I would never fall for fake news or political advertisements — I know what I’m voting for.

看到这里，你可能会问自己：这听起来是很有有趣，可是人工智能真得能够决定最后的胜利吗？我是不是只是被标题党骗进来的？（在现实中）有无数的因素会影响选民的决定，而且并非每次选举的态势都会如此胶着。此外，我永远也不会被假新闻和政治广告所欺骗，我知道我在为什么投票。

Good point, but actually, the margin of victory for many elections is but a few percent (Planet Money 2019). And while you may be immune to targeted political agendas — instead opting to find some sort of truth by exploring a wide-range of sources and examining all sides of any given issue — you are not necessary. In a room of a hundred people, only three need to be influenced in order to change the outcome of an election, and surely you can think of three people that could be influenced?

问得好，但实际上，很多选举最终获胜时也只赢了几个百分点而已。虽然你对个性化的政治广告是免疫的，你可以对多种信源进行比较，你可以审视某一议题的方方面面，你可以找到某种“真理”，但你的存在无足轻重。在一个一百人的房间里，只需要有三个人受到影响就能够改变选举结果，在你看来能有三个人受到影响吗？

Furthermore, the machine learning-technologies and methodologies being used today are still relatively young, and we are only starting to discover the true potential of these algorithms. As machine learning-algorithms become increasingly sophisticated, so too will its influential power increase.

此外，目前使用的机器学习（模型）在方法和技术上都还很幼稚，我们才刚刚开始发现这些算法的真正潜力。随着机器学习算法的日趋复杂，其影响力也将越来越大。

The Data Scientists Are In Charge

核心因素——数据科学家

The future winners are not necessarily those with the strongest ideas, the most money, or the largest amounts of data (data is free and everywhere!), but rather those with the most precise data models. It’s scary, hard to fully understand, easy to underestimate, and perhaps hard to believe, but the importance of AI in politics is only going to grow. AI is the biggest threat to democracy today, and the strongest tool for anyone aiming to win an election.

未来的赢家不一定拥有最强大的想法、最多的资金和最多的数据（数据是免费的，而且无处不在。），但他们一定拥有最精确的数据模型。想要完全理解（人工智能）是困难的，它很容易被低估。尽管听起来很可怕，而且也很难相信，但人工智能在政治中的重要性将会越来越大。人工智能是对民主最大的威胁，也是任何人赢得选举最强大的工具。

You can become the president.

你真的可以成为总统。

All you need is the right algorithm.

你需要的只不过是正确的算法罢了。