人工智能道德与考量 (AI Ethics and Considerations)

分裂与内容冲击的文化 (A Culture of Division and Content Shock)

There is a dividing tension in our culture, birthed from a media that preys on a population quick to believe striking headlines, especially if those headlines fit into our idea of the world. This phenomenon is known as ‘confirmation bias.’ As a society, we berate each other behind our screens, addicted to our quick-to-accuse and quick-to-defend tendencies. We protect our false sense of justice often built on fake news and half truth headlines.

在我们的文化中，存在着一种分裂的张力，这种张力源于一种媒体，这种媒体掠夺了那些Swift相信引人注目的头条新闻的人们，特别是如果这些头条新闻符合我们对世界的观念。 这种现象称为“确认偏差”。 作为一个社会，我们在屏幕后面互相指责，沉迷于我们的快速指责和快速防御的趋势。 我们通常基于虚假新闻和虚假新闻标题来保护我们的错误正义感。

We also live in a cyber age where technological advancements are exhausting to follow and new information rapidly floods the internet. In fact, every 60 seconds, there is an average of 350 million tweets, 294 billion emails sent, 54 million WhatsApp messages sent, and 500 hours of material uploaded on YouTube. With this massive amount of data, we are paralyzed with our ability to meaningfully process information.

我们还生活在网络时代，技术进步日趋枯竭，新信息Swift泛滥到互联网上。 实际上， 每60秒，平均有3.5亿条推文，发送2940亿封电子邮件，5400万条WhatsApp消息以及500个小时的素材上传到YouTube。 拥有如此大量的数据，我们瘫痪了有意义地处理信息的能力。

This divisive tension, paired with the overwhelming surplus of data, can, and will be, exploited and further reinforced by the coming generation of fake news — deepfakes.

这种分裂性的紧张情绪，加上压倒性的大量数据，可以而且将会被下一代假新闻(深造)所利用，并将进一步得到加强。

令人生畏的深度学习假冒 (Daunting Deep Learning Fakes)

Deepfakes are synthetically generated media forged to impersonate an individual’s likeness. Essentially, they are produced clips that present someone as behaving in a way that they are not. This type of synthetic media is not new, but only recently has it evolved to be frighteningly realistic through the use of deep learning techniques and by combining technologies such as face re-enactment and voice synthesizers. Hence, the term ‘deepfake’ was coined: to be a combination of ‘deep learning’ and ‘fake’. Through the use of artificial intelligence, anyone can produce persuasive replicas by training on a large set of images and videos of a specific person where the model will learn and reproduce their behaviors.

Deepfake是合成生成的媒体，伪造成可模仿个人的肖像。 从本质上讲，它们是制作的剪辑，以某种方式呈现某人的行为，而事实并非如此。 这种类型的合成媒体并不是什么新鲜事物，但是直到最近，它才通过使用深度学习技术以及将诸如面部重现和语音合成器之类的技术相结合，发展成为令人恐惧的现实。 因此，创造了“ deepfake”一词：是“ deep learning”和“ fake”的组合。 通过使用人工智能，任何人都可以通过训练特定人的大量图像和视频来制作有说服力的复制品，模型将学习并重现他们的行为。

打造声乐签名 (Forging our Vocal Signatures)

Voice synthesizer technologies have been in production for centuries. Admittedly, there are a large volume of respectable applications for synthetic speech, such as providing an ease of communication for the deaf or blind. However, from the birth of synthetic speech, ensued a myriad of existential questions regarding the ethics of replicating speech. Our voices are sometimes referred to as our audible fingerprints. We each have a distinct vocal signature formed by our body compositions. Replicating this can feel like a violation, impeding our unique vocal birthmarks that form our sense of self.

语音合成器技术已经生产了数百年。 诚然，合成语音有大量受人尊敬的应用，例如为聋人或盲人提供方便的交流。 但是，从合成语音的诞生开始，随之而来的是关于复制语音的伦理学的众多存在性问题。 我们的声音有时被称为可听指纹。 我们每个人的身体成分都有其独特的声音特征。 复制这种感觉就像是一种侵犯，阻碍了形成我们自我感的我们独特的声音胎记。

Once we combine face re-enactment and face replacement technologies with the sophisticated synthetic speech technologies, we receive a piece of media that makes it seem as though a particular individual is saying and doing anything to the whims of another individual.

一旦我们将面部重现和面部替换技术与复杂的合成语音技术结合起来，我们就会获得一种媒介，这似乎使一个特定的人好像在说和对另一个人的异想天开做任何事情。

伪造品的令人不安的后果 (Disturbing Consequences of Deepfakes)

The issue is that we are visual creatures that are quick to adjust our perspective and truth once we see any given event; the awareness that these technologies exist hardly helps. We are susceptible to immediately believing what we visually understand. Before we explore helpful actions we can take to reduce potential damage, let’s look into a few problematic scenarios that could arise.

问题是我们是视觉上的生物，一旦看到任何给定的事件，它们就会Swift调整我们的观点和真理。 这些技术的存在几乎没有帮助。 我们容易立即相信我们在视觉上理解的东西。 在探索可以采取的减少潜在损害的有益措施之前，让我们研究一下可能出现的一些有问题的情况。

There are two damaging, overarching repercussions that can transpire in the era of deepfakes:

在伪造的时代中，有两种具有破坏性的总体影响可以体现：

One of which is the nullification of video footage as reliable evidence. Let’s say violent maltreatment is captured on film and the victim pleads for justice by presenting this evidence to the court. The issue is, video is now no longer seen as honest and feasible evidence. Then, there is a lasting injustice that prolongs the pain of the victim and will not be proven due to the invalidation of video evidence.

其中之一就是取消录像作为可靠的证据 。 假设在电影中记录了暴力虐待，并且受害人通过向法院出示证据证明了正义。 问题是，视频现在不再被视为诚实可行的证据。 然后，存在持久的不公正现象，延长了受害者的痛苦，并且由于视频证据无效而无法得到证明。

The second type is unfair allegations of an individual that destroys their reputation. The scenario where there is no crime committed, yet a deepfake video is generated to couple someone to a scandal. If we follow the same world as the first scenario, we can say that the video is still not seen as feasible evidence. However, that video could severely disrupt an individual’s social reputation. Even rumors plant a seed of doubt in the public’s mind of a certain person, what more a video that plays on our emotional, visceral, what-we-see-is-truth tendencies.

第二类是对个人的不公正指控，破坏了他们的声誉。 没有犯罪的情况，但生成了一个深造视频，将某人与丑闻耦合。 如果我们遵循与第一种情况相同的世界，则可以说该视频仍不被视为可行的证据。 但是，该视频可能会严重破坏个人的社会声誉。 甚至谣言也在某个人的公众心中埋下了怀疑的种子，更何况是一部播放着我们情感，内心，我们所看到的真实趋势的视频。

假新闻会引起第三次世界大战吗？ (Will Fake News Cause World War III?)

One thought experiment that really disturbed me was when someone offered this scenario:

一个真正令我困扰的思想实验是有人提供这种情况时：

Imagine that someone generated a very timely, realistic deepfake that depicted our president’s declaration to engage in war. As the technologies begin to make reality and fake videos indistinguishable, how are we to determine if a recording is real of fake?

想象一下，有人产生了非常及时，切合实际的假象，描绘了我们总统宣战的宣言。 随着技术开始使现实与假视频无法区分，我们如何确定录音是否真实？

Shamir Allibhai, the CEO of Amber, believed that a world war may break out over a deepfake. He proposed that a nation could “fabricate evidence ‘as a pretext to strike first and go to war’ […] (or) incite mayhem by pitting different countries against each other using the technology.”

琥珀(Amber)首席执行官沙米尔·阿里拜(Shamir Allibhai)认为， 一场世界大战可能会因一场深渊而爆发。 他提出，一个国家可以“以制造证据为借口，以“先发动战争” […](或)通过使用该技术使不同国家彼此对抗来煽动混乱。

亲人的恐怖求救 (A Loved One’s Horrific Cry for Help)

And then, there’s this cruel financial scam that always leaves me with goosebumps:

然后，有这样残酷的财务骗局，总是让我感到鸡皮ose：

Your telephone rings and on the other line you hear your daughter (or any loved one) in distress. Your heart starts racing and time stops as their long, loud piercing cries resonate in your ears. Another voice begins to speak and demands that you send a large sum of money for release. The painful echoes of screams drive you to hysteria and you fail to consider the possibility that this may not be real because how could it not be? You heard their distinct, unique voice. Except for the fact that it isn’t really them. It’s a voice synthesizer. It’s a hoax.

您的电话铃响了，在另一条线上，您听到您的女儿(或任何亲人)遇险。 随着他们长长而刺耳的哭声在您的耳朵中产生共鸣，您的心脏开始加速，时间停止。 另一个声音开始讲话，要求您汇出一大笔钱进行释放。 尖叫的痛苦回声将您带入歇斯底里，而您却没有考虑到这可能不真实的可能性，因为怎么可能不真实？ 您听到了他们独特而独特的声音。 除了事实并非如此。 这是语音合成器。 这是个骗局。

This horrifying scenario is a form of fraud known as “virtual kidnapping”. This is exactly what happened to Tracy Holczer who lost “$3,300 during the ordeal in three separate money transfers the scammers convinced her to make.”

这种可怕的情况是一种称为“虚拟绑架”的欺诈形式。 这正是Tracy Holczer所发生的事情，她“在折磨中使诈骗者说服了她，使她损失了3,300美元，分三笔汇款。”

Despite my awareness of this scam, and typical skepticism, I cannot tell you with full certainty that hearing such an intimate cry for help from my loved one wouldn’t send me into an irrational (or perhaps, very rational) frenzy. I cannot confidently say that my first reaction would not be to just give the perpetrator whatever materialistic item they wanted in order to stop the screams. And just imagine if people are not aware of this. A deepfake can quickly cause trauma with awareness, and moreover, an escalation of financial damage without awareness.

尽管我知道这种骗局和典型的怀疑态度，但我无法完全确定地告诉您，听到亲人如此亲切的呼求，不会让我陷入非理性(或非常理性)的疯狂之中。 我不能自信地说，我的第一个React不是只是给肇事者他们想要的任何物质物品，以制止尖叫。 试想一下，如果人们不知道这一点。 一次深深的假冒会在有意识的情况下Swift造成创伤，而且，在没有意识的情况下，财务损失也会升级。

呼唤亲人的声音 (A Cry for a Loved One’s Voice)

On a morbid, tangential note, this is somewhat reminiscent on black mirrors episode, “Be Right Back”, where a grieving widow turns to a new technology that allows her to text her late boyfriend.

从病态切线的角度看，这有点让人想起黑色镜子的情节“ Be Right Back”，一个悲痛的寡妇转向一种新技术，该技术使她可以给已故的男友发短信。

This progresses quickly to calls where the technology trains on recordings of his voice and merges all of the personality quirks from his prior posts to create a realistic, communicative voice on the other line. This brings the existential issues of voice synthesizers to an entirely significant level and frankly, this doesn’t seem too far fetched from where we are now.

这很快就发展到呼叫，在呼叫中，技术会根据他的声音记录进行训练，并合并他先前职位中的所有个性怪癖，从而在另一条线路上创建逼真的，可交流的声音。 这将语音合成器的存在问题带到了一个非常重要的水平，坦率地说，这与我们现在所处的位置似乎并不遥不可及。

Deepfake检测软件呢？ (What about Deepfake Detection Software?)

The good news is that there are technologies in the making that verifies the legitimacy of a video source. These technologies will be highly demanded in use and sophistication. This may address the issue of video footage losing its credibility if we are able to confidently and consistently distinguish real from fake as deepfake software continues to progress.

好消息是，正在制作中的技术可以验证视频源的合法性。 这些技术将在使用和复杂性方面有很高的要求。 如果我们能够随着Deepfake软件的不断发展而自信地，始终如一地将真假与假冒区别开来，这可能会解决录像丧失可信度的问题。

Nonetheless, our culture is suffused with justifiable paranoia regarding our article sources since we are often faced with contradictory opinions with seemingly equal amounts of evidence. We are paralyzed with distrust. As a result of this, there might be social implications that leave damage before someone takes the initiative to check if it is real or not. And frankly, if we’re already not reading past headlines, I wouldn’t put it past us.

但是，由于我们经常面临看似平等的证据，因此我们的文化在文章来源方面充斥着合理的偏执狂。 我们因不信任而瘫痪。 结果，在有人主动检查是否真实之前，可能存在社会隐患。 坦率地说，如果我们已经不读过去的头条新闻，那我就不会过去。

生存鸿沟文化的实用技巧 (Practical Tips to Survive the Culture of Divide)

I was compelled to write this article, partially to provide potential consequences of deepfakes; however, my main intention was to emphasize that we need to be more conscientious and critical consumers of information for all news and media. Therefore, for this last portion of the article, I want further explain how we are vulnerable to fake news, highlight potential biases in research, and share pointers on how we can approach information to be better equipped.

我被迫写这篇文章，部分是为了提供深造的潜在后果。 但是，我的主要目的是强调我们需要对所有新闻和媒体的信息更加认真和批判。 因此，对于本文的最后一部分，我想进一步解释我们如何容易受到假新闻的侵害，强调研究中的潜在偏见，并分享有关如何利用信息来更好地装备信息的指示。

To grasp the degree in which we rely on our flash judgement processing, we can be directed to Annie Reneau’s article, “ I wrote a news headline that didn’t even link to a story. Over 2,000 people commented on it anyway.” The article’s title captures the essence of her experience. She wrote a full article, but what circulated was a post with her article’s headline that linked to a 404 error page instead of her article. Still, there were over 2,000 reactions and thousands of shares containing comments such as, “this article is full of race-baiting b.s.” and “this article is fake news”. This is a clear display of our culture of high reactivity and flash judgements lacking critical and objective analysis.

要了解我们依赖于快速判断处理的程度，可以直接转到安妮·雷诺(Annie Reneau)的文章，“ 我写了一个新闻头条，甚至没有链接到一个故事。 无论如何，有2000多人对此发表了评论。 ”文章的标题抓住了她经验的精髓。 她写了一篇完整的文章，但所散发的却是一篇文章标题链接到404错误页面而非她的文章。 尽管如此，仍然有2000多个React和数千股包含评论的评论，例如“这篇文章充满了种种诱人的bs”和“这篇文章是假新闻”。 这清楚地显示了我们的高React性和缺乏批判性和客观性分析的快速判断文化。

And I don’t blame us. In the same manner that stories have helped us to survive in a world with limitless possibilities (to reenact a certain scenario without having to live through it), we create shortcuts to keep up with the limitless explosion of information. We are hardwired to take shortcuts for survival. As a whole, we are plummeted with excessive information which affects our productivity in decision making and, ironically, understanding something holistically and in-depth.

而且我不怪我们。 故事以相同的方式帮助我们在无限可能的世界中生存(无需经历现实即可重现特定场景)，我们创建了捷径以跟上信息的无限爆炸。 我们很难走捷径以求生存。 总体而言，我们过多地吸收了过多信息，这些信息会影响我们的决策制定效率，并且具有讽刺意味的是，要全面，深入地理解某些内容。

“数字不撒谎，但骗子图” (“Figures Don’t Lie but Liars Figure”)

The quote above encompasses the idea that while pure data is objective, the person that interprets that data could sometimes pervert the truth to fit their own agenda. In ‘Math for Life’, Jeffrey Bennett explains that we “might think that it would be straightforward to draw conclusions once (we) have the results, but […] if someone wants to support a predetermined view, it is easy to “spin” the results of a report by selectively deciding which ones to emphasize.”

上面的引言包含这样一个思想，即虽然纯数据是客观的，但解释该数据的人有时可能会歪曲事实以适合他们自己的议程。 杰弗里·本内特(Jeffrey Bennett)在“生命的数学”一书中解释说，“我们可能认为，一旦(我们)获得了结果，就可以得出结论很简单，但是[…]如果有人想支持预定的观点，就很容易“旋转”。 “通过有选择地确定要强调的内容来得出报告的结果。”

相关=！ 因果关系 (Correlation =! Causation)

A widely circulated ‘mantra’ in the field of statistics is, “correlation does not equal causation.” This serves to warn us against the naive assumption that a change in a variable causes a change in another variable in the presence of a correlation. For example, let’s say you collect data and find a correlation between the sales of ice cream and the sales of sunglasses. A correlation is present, but it would be naive to assume that an ice cream sale causes a sunglasses sale or vice versa.

在统计领域广为流传的“口头禅”是，“相关不等于因果关系”。 这可以警告我们，不要天真地假设变量的变化会在存在相关性的情况下导致另一个变量的变化。 例如，假设您收集数据并找到冰淇淋销售量与太阳镜销售量之间的相关性。 存在相关性，但是假定冰淇淋销售会导致太阳镜销售，反之亦然，这是天真的想法。

Some of my favorite demonstrations of this are on Tyler Vigen’s page of “Spurious Correlations.” Here are a few examples:

我最喜欢的一些演示在Tyler Vigen的“虚假相关性” 页面上。 这里有一些例子：

To accept correlation as causation would be to conclude that more money spent on science, space and technology from the US will lead to more suicides by hanging, strangulation and suffocation. It would also lead us to petition for Nicolas Cage to stop acting in order to spare the poor souls drowning by falling into a pool because of his acting career.

接受关联作为因果关系将得出以下结论：来自美国在科学，太空和技术上花费的更多资金将导致更多人因吊死，勒死和窒息而自杀。 这也将导致我们向尼古拉斯·凯奇(Nicolas Cage)提出请愿，要求其停止表演，以免因其演艺事业陷入困境而淹死的可怜人。

This is all for laughs, of course, but accurately demonstrates how data can easily be misconstrued to fit a narrative.

当然，这全是开怀大笑，但准确地说明了如何轻松地误解数据以适合叙述。

来源分析问题 (Questions For Source Analysis)

Whenever we read anything scandalous that immediately plays on our emotions, let’s take a step back and examine the source of information. Here are the questions we can ask ourselves in order to be keen, skeptical and critical consumers of information:

每当我们读到任何可立即影响我们情绪的丑闻时，就让我们退后一步，检查信息来源。 为了成为敏锐的，持怀疑态度的和至关重要的信息使用者，我们可以问自己以下问题：

1. Selection Bias: How many participants were there? Does the group of participants accurately represent the population or does it cause the research to favor certain outcomes?

   选择偏见：有多少参与者？ 参与者组是否准确地代表了人群，还是使研究偏向于某些结果？

2. Measurement Bias: Is the environment of the survey or experiment fair? Does it foster a certain response?

   测量偏差：调查或实验的环境是否公平？ 它会引起一定的React吗？

3. Poor validity: Does the research measure what it claims to measure? Are the tests relevant to what the experiment is testing? Are the findings generalized to the population it claims to represent?

   有效性差：研究是否衡量了它声称要衡量的内容？ 测试与实验所测试的内容相关吗？ 研究结果是否推广到它声称代表的人群？

4. Precision: Is the sample size and response rate sufficient?

   精度：样本量和响应率是否足够？

Another wonderful resource is ‘Ten Questions for Fake News Detection’ shared by the News Literacy Project. The mission of this educational nonprofit is to ‘teach students the skills they need to be smart, active consumers of news and other information and engaged, informed participants in civic life.’

另一个很棒的资源是新闻素养项目共享的“假新闻检测十个问题” 。 这个教育性非营利组织的使命是“向学生传授他们所需的技能，使其成为聪明的，活跃的新闻和其他信息的消费者以及积极，有见地的公民生活参与者”。

This type of mass initiative is exactly what we need to push back the current era of fake news and prepare for this upcoming evolution of digital news fraud. Awareness breeds cautious consumers and an urgency to create technology that will detect the legitimacy of any given media. It is my hope that you leave this article with the desire to take a meaningful look at your sources, approach each piece of information with a composed objectivity and share information with the intent of community growth.

正是这种大规模的主动行动正是我们需要推倒当前的假新闻时代，并为即将到来的数字新闻欺诈发展做准备。 意识孕育了谨慎的消费者，并且迫切需要开发一种技术来检测任何给定媒体的合法性。 我希望您希望这篇文章充满希望地对您的资源进行有意义的研究，以客观的方式处理每条信息，并出于社区发展的目的共享信息。