中国ai chip初创公司
By Jared Newman
杰里德·纽曼(Jared Newman)
In late April, an investigation by The Guardian tied widespread 5G and coronavirus conspiracy theories to an an evangelical pastor in England.
4月下旬， 《卫报》的一项调查将广泛的5G和冠状病毒阴谋论与英格兰的一位福音派牧师联系在一起。
The pastor, Jonathan James, baselessly claimed in an audio recording that 5G wireless networking would allow Bill Gates to track people who’d been vaccinated for the coronavirus, and that the virus itself was a cover-up for 5G’s poisonous effects. On YouTube and private messaging apps such as WhatsApp, anti-5G groups pushed the sermon-like recording as coming from a “former Vodafone boss,” lending it an air of credibility.
牧师乔纳森·詹姆斯(Jonathan James)在音频记录中毫无根据地声称，5G无线网络将使比尔·盖茨能够追踪曾接种过冠状病毒疫苗的人，而该病毒本身就是5G毒害作用的掩盖。 在YouTube和WhatsApp之类的私人消息传递应用程序上，反5G团体将这种讲道形式的录音推向了“沃达丰前任老板”，给人一种可信的氛围。
In fact, The Guardian reported that the pastor had only worked in a sales role at Vodafone for less than a year in 2014, long before networks even started deploying 5G. But the publication wasn’t working alone in debunking the story. It had help from a startup called Logically, which combines AI, automation, and human fact-checkers to root out fake news.
实际上，据《卫报》报道，这名牧师在沃达丰仅在2014年担任销售职务不到一年，远远早于网络甚至开始部署5G。 但是，该出版物并不能独自揭穿这个故事。 它得到了一家名为Logically的初创公司的帮助，该公司结合了AI，自动化和人类事实检查员，以铲除虚假新闻。
Logically has already worked with government agencies in India and the United Kingdom to identify extremist content and disinformation campaigns, and now it’s launching in the United States, both as a fact-checking product for consumers and as a service for organizations and government agencies that need help dealing with a tide of misinformation. In doing so, it’s going a clear step further than social networks themselves, which have struggled to flag and contain bad information on their own platforms.
逻辑上已经与印度和英国的政府机构合作，确定极端主义内容和虚假宣传活动，现在它已在美国启动，既是面向消费者的事实核查产品，又是为需要帮助的组织和政府机构提供的服务帮助应对错误信息。 这样做比社交网络本身迈出了明显的一步，后者一直努力在自己的平台上标记和包含不良信息。
“What we’d like to do is bring the industrial revolution to fact-checking,” founder Lyric Jain says in an interview.
创始人Lyric Jain在接受采访时说：“我们要做的是将工业革命带入事实核查中。”
更快的事实检查 (Faster Fact-Checking)
Logically isn’t the only group that’s trying to fuse fact-checking with automation. Facebook says it uses machine learning to flag potential misinformation on its network so third-party fact-checkers can review it. Full Fact, a nonprofit based in the United Kingdom, gives human fact-checkers tools to find false claims and match them with refuting facts or reporting, and is developing a machine learning model that can debunk some claims without human involvement. The Duke University Reporters’ Lab created a tool for readers that automatically pairs public figures’ claims with the work of human fact-checking organizations.
从逻辑上讲，并不是唯一试图将事实检查与自动化相融合的组织。 Facebook说，它使用机器学习来标记其网络上的潜在错误信息，以便第三方事实检查者可以对其进行审查。 Full Fact是一家位于英国的非营利性组织， 它为人类事实检查人员提供了查找虚假主张并将其与反驳事实或报告相匹配的工具 ，并且正在开发一种无需人工干预就能对某些主张进行揭穿的机器学习模型。 杜克大学记者实验室为读者创建了一个工具，该工具可自动将公众人物的主张与人类事实检查组织的工作配对。
Jain says that Logically’s methodology is a bit different. Beyond just matching up bogus claims with evidence to the contrary, the startup treats the fact-checking process like an assembly line, with an algorithm prioritizing and doling out tasks. As a basic example, one step might involve finding the source of a rumor, while another might involve researching the claim. Logically’s system can handle some of these tasks automatically, but it can also hand out assignments to human fact-checkers based on their area of expertise.
贾恩(Jain)说，逻辑上的方法有些不同。 除了将伪造的索赔与相反的证据进行匹配之外，初创公司将事实检查过程像装配线一样对待，并通过算法对任务进行优先级分配。 作为一个基本示例，一个步骤可能涉及查找谣言的来源，而另一步骤可能涉及研究索赔。 逻辑上的系统可以自动处理其中一些任务，但也可以根据其专业领域将任务分配给人类事实检查人员。
“We break down the entire fact-checking process into piecemeal steps, so that each individual is able to carry about a step that they specialize in,” Jain says.
贾恩说：“我们将整个事实检查过程分解为零碎的步骤，以便每个人都可以执行他们专门研究的步骤。”
The startup employs about 40 full-time fact-checkers overseas and expects to hire nearly a dozen in the United States within the next six to nine months, Jain says. And he describes an array of tools that the company has developed: There are algorithms that can analyze media for signs of doctoring, ways to recognize coordinated behavior from bots, a “super-focused version of Google on steroids” that tracks down sources of evidence, and forensic tools for finding the original source of a false claim. For The Guardian’s investigation, it even used an experimental “Rumor Tracer” tool that lets people report a text message or image they received, along with their phone numbers and the numbers of anyone who shared or received the rumor.
贾恩说，这家初创公司在海外雇用了约40名全职事实检查员，预计在未来六至九个月内将在美国雇用近十二名。 他还介绍了该公司开发的一系列工具：有些算法可以分析媒体的篡改迹象，从机器人中识别协调行为的方式，一种“谷歌针对类固醇的超级关注版本”，可以追踪证据来源，以及用于寻找虚假索赔原始来源的取证工具。 为了进行《卫报》的调查，它甚至使用了实验性的“ Rumor Tracer ”工具，该工具可让人们报告收到的短信或图像，以及他们的电话号码以及分享或接收该谣言的任何人的电话号码。
Jain hopes that the combination of all these methods will allow its fact-checkers to respond faster and more efficiently.
Jain希望所有这些方法的结合将使其事实检查者能够更快，更有效地做出响应。
“When you compare the volume of claims fact-checking organizations are able to get through to the misinformation that exists, the comparison isn’t even worth being made,” he says. “We’d like to change that balance.”
他说：“当您比较索赔的数量时，事实核查机构能够了解存在的错误信息，甚至不值得进行比较。” “我们想改变这种平衡。”
作为“举报人”的朋友 (Friends as “Whistleblowers”)
Identifying and debunking false claims is only half the battle, though. Logically is also bringing news consumers into the fold through a newly launched browser extension and mobile app, so they can check on dubious articles themselves.
但是，识别和揭穿虚假主张只是成功的一半。 逻辑上也通过新推出的浏览器扩展和移动应用程序将新闻消费者吸引到了一起，因此他们可以自己检查可疑文章。
I’ve been using a pre-release version of the app, and have indeed found that it can work pretty well.
我一直在使用该应用程序的预发布版本，并且确实发现它可以很好地运行。
When I searched on DuckDuckGo for “5G causes cancer,” for instance, one of the first results was from a website called cancer.news, offering “7 Reasons why 5G is a threat to overall health.” Logically’s extension immediately displayed a pop-up asking if I really wanted to read the article, and clicking the extension brought up a warning that “this article has a chance of being Junk Science.”
例如，当我在DuckDuckGo上搜索“ 5G致癌”时，最初的结果之一来自一个名为Cancer.news的网站，其中提供了“ 5G威胁整体健康的7个原因”。 逻辑上，该扩展名立即显示一个弹出窗口，询问我是否真的要阅读该文章，单击扩展名会发出警告，“本文可能会成为垃圾科学。”
Although Logically hadn’t fact-checked the cancer.news story specifically, its algorithms had accurately predicted low credibility for both the news source and the article itself. Those predictions can be based both on the track record of the source and what it calls “the hallmarks of misinformation” in the story itself. When Logically isn’t able to rate an article immediately, users can send off parts of the article for further fact-checking.
尽管Logical尚未具体检查“ cancer.news”新闻，但其算法已准确地预测了新闻来源和文章本身的可信度低。 这些预测既可以基于消息来源的跟踪记录，也可以基于故事本身中所谓的“错误信息的标志”。 当Logically无法立即对文章进行评分时，用户可以将文章的某些部分发送出去，以进行进一步的事实检查。
Because mobile apps are more limited than browser extensions, Logically’s app won’t be able to detect dubious claims automatically, but users will be able to hit the share button on iOS or Android to check on links that they read in other apps.
由于移动应用程序比浏览器扩展程序受到更多限制，因此，Logically的应用程序将无法自动检测可疑声明，但用户将能够点击iOS或Android上的“分享”按钮来检查他们在其他应用程序中读取的链接。
Jain acknowledges that the people who would most benefit from seeing a fact-check probably aren’t seeking out services like Logically. Still, he’s hopeful that a small group of truth-conscious readers will be enough to snuff out falsehoods before they become widespread, particularly on closed networks like WhatsApp.
贾恩(Jain)承认，从事实调查中受益最大的人可能并没有寻求像Logical这样的服务。 尽管如此，他还是希望一小部分对真理有意识的读者能够在虚假现象广泛传播之前就将其消除，特别是在诸如WhatsApp之类的封闭网络上。
“Misinformation is most damaging when it first comes out of those concentrated groups into the wider public,” he says. “If we have whistleblowers in the wider public, exposed to those forums . . . we effectively have a chance to respond very soon to what’s still an emerging threat, in the pre-viral stages.”
他说：“错误信息最先从那些集中的人群中渗入到广大公众中时，最具有破坏性。” “如果我们在广大公众中有举报人，请访问这些论坛。 。 。 在病毒传播前阶段，我们实际上有机会很快对仍在出现的威胁做出响应。”
Beyond just putting the burden on users, Logically is also interested in working with government partners and other groups to identify false claims. Last year, it worked with an unnamed “social platform” to report on the spread of fake news before India’s general election. It analyzed 1 million articles, finding 14% to be unreliable and 25% to be false. Logically also teamed up with Maharashtra’s cyber police to monitor for misinformation on social media ahead of the state’s Legislative Assembly elections, and in the United Kingdom, it worked with the Metropolitan Police to develop a tool that identifies extremist content.
除了仅给用户带来负担外，Logically还希望与政府合作伙伴和其他组织合作以识别虚假声明。 去年，它与一个未命名的“社交平台”合作，报道了印度大选前假新闻的传播情况。 它分析了100万篇文章，发现14％的文章不可靠，而25％的文章不真实。 逻辑上也与马哈拉施特拉邦的网络警察合作，在州议会选举之前监视社交媒体上的错误信息。在英国，它与大都会警察合作开发了一种识别极端主义内容的工具。
Jain says the systems it’s launching the United States are built on the work it’s already done overseas.
贾恩(Jain)说，它在美国推出的系统是建立在已经在海外完成的工作之上的。
“This is the same platform that we’ve been building for three years,” Jain says. “We’ve been battle-tested and battle-hardened.”
“这是我们已经建立了三年的平台，” Jain说。 “我们已经经受了战斗考验，并经过了艰苦的战斗。”
超越真理 (Beyond Truthiness)
Logically isn’t just interested in swatting down falsehoods. The startup’s browser extension also analyzes the sentiment of comments on sites like Twitter and YouTube, and can block out toxic, profane, or obscene content unless users click through on the contents.
从逻辑上讲，不仅仅是减少虚假信息。 这家初创公司的浏览器扩展程序还可以分析Twitter和YouTube等网站上的评论情绪，并且可以阻止有毒，亵渎或淫秽的内容，除非用户单击该内容。
“We’re kind of branching the discussion away from just false content into this broader category of harmful content,” Jain says.
贾恩说：“我们正在将讨论范围从虚假内容扩展到更广泛的有害内容类别。”
Compared to Logically’s fact-checking, the results here are a bit shakier. In my experience, it would sometimes flag tongue-in-cheek content as toxic, or innocuous content as obscene. Still, Jain believes its experimentation in this area can become a model for social networks like Twitter and Facebook over time.
与Logical的事实检查相比，这里的结果有点不稳定。 以我的经验，它有时会将嘲笑的内容标记为有毒，而将无害的内容标记为淫秽。 贾恩仍然相信，随着时间的推移，它在这一领域的试验可以成为Twitter和Facebook等社交网络的典范。
“If we’re able to return some of this power back to end users, the end user can pick and choose what type of online harm they’d like to [be exposed to], and over time there can become industry standards,” he says.
“如果我们能够将这些力量中的一部分归还给最终用户，那么最终用户可以选择他们想要[暴露于]哪种类型的在线危害，并且随着时间的流逝，它们会成为行业标准，”他说。
All of this raises the question of why social networks don’t offer these kinds of tools themselves. Imagine if Facebook, for instance, provided its users with the kinds of instant fact-checks that Logically makes available through its browser extension and app, or if Twitter gave users even more granular screening options for toxic or offensive content. Such tools would be a lot more powerful and useful if they were built into the actual products, rather than just tacked on by a third party.
所有这些都提出了一个问题，为什么社交网络本身不提供这些工具。 想象一下，例如，Facebook是否为用户提供了Logistic通过其浏览器扩展程序和应用程序提供的即时即时检查，或者Twitter是否为用户提供了更细粒度的筛选有毒或令人反感内容的选项。 如果将这些工具内置到实际的产品中，而不是仅仅由第三方使用，那么它们将更加强大和有用。
Jain believes these kinds of controls are too fraught with political peril for companies like Facebook and Twitter to offer directly. But that might change if Logically can prove itself over time.
贾恩(Jain)认为，这类控制措施充满了政治风险，因此Facebook和Twitter等公司无法直接提供这些控制措施。 但是，如果逻辑上能够随着时间的推移证明自己，那可能会改变。
“We can afford to take this rather large, innovative leap, but for a platform like Twitter or Facebook with hundreds of millions or billions of users, it’s probably a leap too far without sufficient evidence,” he says. “Hopefully we can help accelerate or provide some of that evidence.”
他说：“我们可以承担这一相当大的创新性飞跃，但对于Twitter或Facebook这样拥有数亿或数十亿用户的平台来说，如果没有足够的证据，这可能是一个太大的飞跃。” “希望我们可以帮助加快或提供一些证据。”
中国ai chip初创公司
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
