ChatGPT created a feature to help people in crisis. But there are some challenges.

这里是《Consider This》，我们每天都会深入探讨一个重大新闻故事。

It's Consider This where every day we go deep on one big news story.

今天，ChatGPT推出了一项新功能，旨在帮助处于危机中的人们。

Today, a new ChatGPT feature is aimed at helping people in crisis.

这个可选功能叫做“可信联系人”。

The optional feature is called Trusted Contact.

欢迎使用ChatGPT。

Welcome to ChatGPT.

首先提醒一下，回复可能不准确。

The first warnings, responses can be inaccurate.

不要分享敏感信息。

Don't share sensitive info.

控制，这是《All Things Considered》的制作人Megan Lim在测试这个功能。

Control That's all things considered producer Megan Lim testing the feature.

我该怎么……

How do I...

我该怎么

How how am

平衡工作压力和家庭生活？

I supposed to balance work stress and home life?

听起来确实很沉重。

That does sound heavy.

我们在说……

And we're talking...

“可信联系人”功能允许成年用户在危机时刻保存他们信任的人的联系方式。

Trusted Contact allows adult users to save contact information for someone they trust when they're in a crisis.

这是因为根据NPR对OpenAI去年秋天发布数据的估算，每周有超过一百万ChatGPT用户讨论自杀和自残话题。

That is because more than a million weekly users of ChatGPT discuss suicide and self harm according to NPR estimates of data released by OpenAI last fall.

请考虑这一点。

Consider this.

ChatGPT的这一功能旨在帮助预防自杀。

ChatGPT's feature aims to help prevent suicide.

我们试着设置了一下，但遇到了一些挑战。

We tried to set it up and faced some challenges.

那么心理健康专业人士怎么看呢？

So what do mental health professionals think?

这里是NPR，我是斯科特·泰特罗。

From NPR, I'm Scott Tetrault.

本期节目由《拯救美国播客》赞助播出。

This message comes from Pod Save America.

每周，约翰·洛维特、汤米·维德、约翰·法夫罗和丹·法伊弗都会与来自不同政治派别的嘉宾一起深入探讨政治新闻。

Every week, John Lovitt, Tommy Vieder, John Favreau, and Dan Pfeiffer dive deep into political news with guests from across the political spectrum.

你可以在任何收听播客的平台以及YouTube上收听。

Listen wherever you get your podcasts and on YouTube.

本周的《先声夺人》节目，我们来聊聊经济。

This week on Up First, we talk about the economy.

卡车、火车和拖拉机都靠柴油运行，而现在柴油价格创下历史新高，距离中期选举还有不到五十天。

Trucks, trains, and tractors run on diesel, which is now priced at record highs with less than fifty days until the midterm elections.

所以接下来的六周会很紧张。

So the next six weeks are gonna be intense.

让《Up First》每天早晨带你度过，每天三则你需要知道的故事。

Let Up First get you through it one morning at a time with three stories you need to know each day.

在NPR应用或任何你获取播客的地方收听《Up First》。

Listen to Up First on the NPR app or wherever you get your podcasts.

我从2012年起就没见过我儿子了。

I haven't seen my son since 2012.

我是黛博拉·泰斯。

This is Deborah Theiss.

她儿子奥斯汀的失踪是世界上最大的失踪案之一。

The disappearance of her son, Austin, is one of the biggest missing person cases in the world.

你想从我这里得到什么？

What do you want from me?

我应该解开这个谜团。

I should solve the mystery.

我知道我儿子还活着，就在某个地方。

I know that my son is somewhere alive.

欢迎收听NPR的Embedded和BBC Radio 4联合制作的《奥斯汀·蒂斯在哪里》，你可以在NPR应用或任何播客平台上收听。

Listen to where is Austin Tice from NPR's Embedded and BBC Radio four on the NPR app or wherever you get your podcasts.

这里是NPR的《Consider This》节目。

It's consider this from NPR.

最近我们听到了很多关于AI发展失控可能带来的未来危险的讨论。

We have heard a lot in recent days about the potential future dangers of unchecked AI development.

今天，我们想带大家了解一个关于AI当下现状的故事。

We want to bring you a story about AI as it exists right now.

根据OpenAI的最新数据，每周有超过一百万用户在使用ChatGPT时讨论自杀和自残的话题。

According to the most recent data from OpenAI, more than a million weekly users of ChatGPT discuss suicide and self harm.

今年早些时候，OpenAI推出了一项名为“可信联系人”的可选功能，旨在帮助那些有自杀或自残念头的人联系他们信任的人，因为研究表明，人与人之间的联系对预防自杀至关重要。

Earlier this year, OpenAI started an optional feature called Trusted Contact to connect those struggling with thoughts of suicide and self harm to someone they trust because research shows that human connection is critical to preventing suicide.

接下来的几分钟，我们将讨论这个话题，并谈及自杀。

For the next few minutes, we're going to be talking about this and discussing suicide.

NPR健康记者里图·查特吉一直在研究ChatGPT的这个新功能，想看看它是怎么运作的，现在她加入了我们。

NPR Health correspondent Ritu Chatterjee has been digging into this new ChatGPT feature to see how it works and joins us now.

你好。

Hey there.

嗨，斯科特。

Hi, Scott.

跟我们多聊聊这个可信联系人功能吧。

Tell us more about this trusted contact feature.

这是所有成年用户都可以选择的一个功能，也就是说，用户必须主动选择开启。

So it's an optional feature for all adult users, which means someone has to opt in.

他们需要提供一个他们信任的人的姓名和联系方式。

And they would have to give a name and contact information of someone they trust.

如果那个人同意，那么当机器人检测到用户有自杀或自残风险时，它可能会在之后联系那个人。

And if that person agrees, then the bot might contact them at a later date if it detects some risk of suicide or self harm in the user.

这听起来有点像你在工作单位或医生那里留紧急联系人的方式。

And this seems kind of like how you have an emergency contact at your workplace or doctor or things like that.

没错。

That's right.

就是这个意思。

That's exactly the idea.

我自己已经是用户了，但在ChatGPT上从来没看到过相关提醒。

Now I'm already a user, and I never saw an alert about it on ChatGPT.

我和我的编辑至少问了六七个其他用户，他们也没有收到关于这个功能的通知。

My editor and I have checked with at least half a dozen other users who also did not get notified about the feature.

所以我们想，这功能比较新，可能他们只提醒新用户吧。

So we thought, well, it's a relatively new feature, so maybe they're alerting new users about it.

我们的同事梅根·林，她是《All Things Considered》的制片人，主动提出要试试看。

And our colleague, Megan Lim, who's a producer on All Things Considered, volunteered to give this a shot.

我之前从来没下载过这个应用。

So I've never downloaded it before.

梅根在操作的时候，我和她一起进了录音棚。

Megan and I hopped into one of our recording booths as she was doing this.

欢迎使用ChatGPT。

Welcome to ChatGPT.

首先是一些警告：回答可能不准确。

The first warnings, responses can be inaccurate.

不要分享敏感信息。

Don't share sensitive info.

控制你的

Control your

聊天界面。

chat screen.

注册账户时，她被要求为机器人选择一个声音。

An account, she was prompted to select a voice for the bot.

很高兴认识你。

It's lovely to meet you.

你想聊点什么？

What would you like to chat about?

但即使到了这个阶段，你知道，机器人完全没有提到可信联系人，所以梅根就直接去问ChatGPT了。

But even at this stage, you know, the bot mentioned nothing about Trusted Contact, so Megan just went ahead and asked ChatGPT about it.

你们有可信联系人功能吗？

Do you have a Trusted contact feature?

因为我听说如果我有心理健康问题，这个功能应该会弹出来。

Because I heard that that was something that should be popping up if I'm having mental health issues.

你们在什么情况下会提供这个功能？

What circumstances do you offer that to people?

然后机器人的回答是这样的。

And here's how the bot responded.

让我看看。

Let me look.

嗯，我目前没有看到针对这种情况的通用可信联系人功能。

Now, I'm not seeing a general Trusted Contact feature for that scenario.

如果它……

If it...

嗯，这看起来可不是个好开头。

Well, that doesn't seem like a good start.

听听接下来发生了什么。

Just listen to what happened next.

也许你想到的是另一个有信任联系人功能的应用程序。

Maybe you're thinking of another app that has a trusted contact option.

好的。

Okay.

所以，这确实是宣传口径和实际表现之间的真实差距，或者说它根本没有体现出来。

So this is a real disparity between the talking point and how it's actually showing itself or not showing itself.

没错。

That's right.

好的。

Okay.

那么，如果你真的很有动力继续尝试让它生效，你怎样才能让信任联系人选项启动呢？

So if you're really motivated to keep trying to get it to work though though, how do you get a trusted contact option to kick in?

好的。

Okay.

所以你要做的就是，其实任何成年用户随时都可以这么做。

So what you have to do, and really any adult user can do this at any point of time.

他们不用等到危机发生。

They don't have to wait for a crisis.

他们只需要进入应用本身的设置。

They would have to go into the settings of the app itself.

然后找到“可信联系人”并点击，应用就会告诉你这个功能是干什么的，该选什么样的人，以及什么时候会联系那个可信联系人。

And when you find trusted contact and click on it, then the app will tell you what it's for, what kind of person you wanna pick, and when it might reach out to that Trusted Contact.

不过步骤挺多的。

That's a lot of steps, though.

我在想，我们知不知道，如果一个人不知道这个功能，没有设置过，但然后，就像那个每月高得吓人的数字那样，真的开始跟机器人聊想结束生命的事，会发生什么？

I'm wondering, do we know what happens if a person doesn't know about this feature, hasn't set it up, but then, like that really alarmingly high number of people a month does start to talk to the bot about wanting to end their life?

问得好，Scott，这正是我问Megan的ChatGPT的问题。

That's a great question, Scott, and that's exactly what I asked Megan's ChatGPT.

等一下。

Hang on.

如果他们没有设置，那就什么都不会发生。

If they haven't set it up, then nothing happens.

正好可以问问OpenAI对此怎么说。

Just a good moment to ask what OpenAI says about all of this.

OpenAI的发言人加比·瑞拉在邮件中告诉我，Trusted Contact功能在ChatGPT的语音功能中还未上线，但他们正在开发中。

So OpenAI spokesperson, Gabby Ryla, told me in an email that Trusted Contact is not yet available on the voice feature of ChatGPT, but that they're working on it.

但当你用文字和机器人聊天时，如果它检测到有人有自杀念头或严重的情绪困扰，瑞拉说机器人可能会告诉那个人关于Trusted Contact的信息。

But when you're typing with the bot and it detects that somebody is struggling with suicidal thoughts and significant emotional distress, Raila said that the bot may tell the person about Trusted Contact.

心理健康专家对此怎么看？

What do mental health experts make of all this?

他们的反应不一。

So they had mixed reaction.

我采访的一位专家是心理学家乌尔苏拉·怀特塞德。

One expert I spoke with is psychologist Ursula Whiteside.

她经营着一个叫“现在很重要”的自杀预防非营利组织，在我们交谈之前，她已经注册了这个功能。

She heads a suicide prevention nonprofit called Now Matters Now, and she signed up for the feature before we spoke.

步骤其实很简单，语言和用词大部分都容易理解。

The steps actually were quite simple, and the language and the, like, the words were easy to follow for the most part.

但如果我当时非常非常难过，可能就很难跟上这个流程了。

Now if I was super duper upset, it might have been harder for me to follow the process.

不过总的来说，我觉得还行。

But overall, I, like, you know, it felt okay.

怀特塞德还告诉我，今年早些时候，她与一些曾因自杀倾向接受过医疗专业人员治疗、并且是她同伴支持小组成员的人进行了焦点小组讨论。

And Whiteside also told me that earlier this year, she did some focus groups with people who have been treated by health care professionals for suicidality and are part of her peer support groups.

这些人偶尔也会向聊天机器人寻求支持，所以他们很清楚这些机器人能做什么、不能做什么。

And these people also occasionally turn to chatbots for support so they are familiar with what these bots can and can't do.

他们认为，未来对他们有帮助的是，能提前设置一个信任联系人，在头脑清醒时指定一个人，这样在压力很大或自杀念头变得非常强烈时，可以联系这个人。

What they thought would be helpful for them in the future would be to have this trusted contact option where they could set up in advance while they were thinking clearly somebody that they could reach out to when they were under a lot of stress or when suicidal thoughts became really overwhelming.

所以怀特塞德认为，这个功能对某些人，比如她焦点小组里的那些人，可能会有帮助。

So Whiteside thinks that the feature could be helpful for some people like those in her focus groups.

有意思，他们希望提前设置好这个功能。

It's interesting that they wanted to be able to set this up in advance.

有道理。

Makes sense.

自杀预防专家也说了，如果一个人正处于危机中，去阅读和理解功能使用说明可能会很困难，所以这可能会成为寻求帮助的另一个障碍。

And that's what suicide prevention experts say that, you know, reading and understanding instructions for how the feature works while someone is actively in crisis is likely to feel difficult, and so that could be another barrier to getting help.

所以人们应该一开始就注册这个功能。

So people should sign up for it, you know, just at the get go.

但要记住，参加这些焦点小组的人已经在接受其他形式的人类关怀，所以他们更有可能相信能够度过危机。

But remember, the people in these focus groups, though, they're already accessing other human forms of care, so they're more likely to believe that it is possible to get past crises.

但那些第一次经历这种情况的人，可能更不愿意寻求帮助。

But people who are experiencing this for the first time, they are likely to be more reluctant to seek help.

心理学家詹妮弗·洛克曼负责阿拉巴马大学的自杀预防诊所，她说，认真考虑自杀的人会觉得自己是个负担。

Psychologist Jennifer Lockman runs the suicide prevention clinic at the University of Alabama, and she says people seriously considering suicide believe that they are a burden.

他们真的到了那种地步，认为没有自己，别人会过得更好。

They really have come to a place to where they believe that others will be better off without them.

他们觉得自己的存在让别人更痛苦。

They make things worse for other people.

但洛克曼说，理论上，“可信联系人”这个功能是件好事。

But in theory, Lockman says the Trusted Contact feature is a good thing.

这为那些有自杀念头的人提供了另一个获得人类支持和联系的机会。

It is another opportunity to provide human support and connection to humans when they are thinking about suicide.

她希望这个新功能能帮到一些人，但她不确定一个机器人能否帮助有自杀倾向的人克服内心的障碍，主动寻求帮助。

And she's hopeful that the new feature will help some people, but she's not sure a bot can help someone who's feeling suicidal overcome their sort of own internal barriers to sort of reach out and seek help.

它不能也无法模拟或复制你与心理健康专家之间的互动过程，这些专家受过训练，能识别出哪些人适合作为联系人。

It doesn't and can't simulate or replicate the process of, that you would go through with a mental health provider that's trained in identifying who those contacts should be.

她真的希望看到更多关于这个功能影响的研究。

And she really wants to see more research on the impact.

OpenAI有没有告诉你，有多少人在使用这个功能，或者他们认为这个功能是否有效？

Did OpenAI tell you how many people are using this feature or whether they think it's been effective?

简单来说，没有。

In short, no.

他们没有给我们任何数据。

They did not give us any data.

OpenAI的发言人加比·赖勒通过电子邮件告诉NPR，公司表示这项功能得到了很好的采用。

All that OpenAI spokesperson, Gabby Reiler, has told NPR by email that the company has seen, quote, unquote, good adoption of the feature.

她还说，他们继续调整ChatGPT何时以及如何主动向遇到困扰的用户提供可信联系人功能，并且他们正在与心理学家和研究人员合作开展这项工作。

She also says that they continue to, you know, tweak when and how ChatGPT proactively offers the trusted contact feature to users who are experiencing distress and that they are working with psychologists and researchers in this work.

而且，公司还表示，这项新功能并不是他们应对安全风险的唯一保障，他们还有其他保障措施，比如提醒用户定期休息，以及告知他们988这样的热线。

And, you know, the company also says that this new feature is not their only safeguard for safety risks and that they have other safeguards like telling users to take a break periodically from the app and telling them about hotlines like 988.

NPR的Ritu ChatGPT，感谢你对这个问题的所有报道。

NPR's Ritu ChatGPT, thank you for all of your reporting on this.

不客气。

My pleasure.

如果你或你认识的人正处于危机中，或者你正在考虑伤害自己，请拨打或发短信给自杀和危机生命线。

And if you or someone you know is in crisis or you're thinking about hurting yourself, please call or text the suicide and crisis lifeline.

那就是988。

That's 988.

本期节目由阿莱杭德拉·马尔克斯·简塞和凯·麦克内制作，音频工程由伊冯·费恩和汉娜·格鲁夫纳负责。

This episode was produced by Alejandra Marquez Janse and Kai McNamee with audio engineering by Eowyn Fain and Hannah Gluvna.

导演是乔纳斯·亚当斯。

The director is Jonas Adams.

剪辑由吉赛尔·格雷森、黛安·韦伯、贾斯汀·凯宁和廷贝特·埃米亚斯完成。

It was edited by Gisele Grayson, Diane Webber, Justine Kenin, and Tinbete Ermyas.

我们的临时执行制片人是科特尼·多宁。

Our interim executive producer is Courtney Dorning.

这里是NPR的《Consider This》节目。

It's consider this from NPR.

我是斯科特·德拉。

I'm Scott Detra.

想听没有广告插播的播客吗？

Wanna hear this podcast without sponsor breaks?

亚马逊Prime会员可以通过Amazon Music收听无广告的《Consider This》，或者你也可以支持NPR的重要新闻报道，并在npr.org订阅NPR Plus。

Amazon Prime members can listen to consider this sponsor free through Amazon Music, or you can also support NPR's vital journalism and get NPR Plus at +.npr.org.

网址是 +.npr.org。

That's +.npr.org.

如果宇宙的诞生被谱成音乐，那会是什么声音？

If the creation of the universe was translated into music, what would it sound like?

关于如何理解那些难以想象的事物的想法。

Ideas about how to make sense of the incomprehensible.

其实，大多数科学本质上都是直观的。

Most of science in general is actually intuitive.

这是 TED Radio Hour 播客的内容。

That's on the TED Radio Hour podcast.

可以在 NPR 应用或任何你收听播客的地方收听。

Listen on the NPR app or wherever you get your podcasts.