利用 AI 伴侣扩展心理健康支持 - Kai.ai

欢迎回到医疗创新播客。

Welcome back to the Medical Innovation Podcast.

我们仍然在拉斯维加斯健康展现场。

We're still here at Health in Las Vegas.

我们非常高兴能采访到Kai.ai的首席执行官兼联合创始人亚历克斯·弗伦克尔。

We're excited to be talking to Alex Frenkel, the CEO and co founder of Kai.ai.

亚历克斯，非常感谢你来参加我们的节目。

Alex, thank you so much for joining us.

谢谢你们邀请我。

Thank you for inviting me.

我们很想

So we'd love

先请你做个自我介绍。

to start off by asking you to introduce yourself.

跟我们聊聊你的背景，以及你是怎么走到Kai.ai的。

Tell us a little bit about your background, your journey to Kai.

所以，我的背景其实是临床方面的，这并不意外。

So, not surprisingly my background is clinical.

我主修临床心理学，后来在公立和私立诊所、医院里当治疗师，服务过成人和青少年。

I majored in clinical psychology and actually worked as a therapist in public and private clinics, hospitals, adults and young people.

我还做过两所高中的学校心理师，做了大量评估工作，这就是我的职业路径。我亲眼看到心理治疗、尤其是临床心理学对人们生活和家庭的巨大影响，但也看到了它的局限——心理医生太少，排队等预约的时间太长。

Was the school psychologist for two high schools, did a lot of diagnostics and that was like my path, right, and I could both like see the tremendous impact that therapy and especially clinical psychology can have on people's life and families, but also I could see the limitations, not enough clinicians, long wait lists.

在我工作的学校里，有些学生一年只能见我几分钟，一两次就完了，这远远不够。

There were some students at the schools where I practiced that I could see only for a few minutes once or twice a year, and it wasn't enough.

后来，我几乎纯属偶然，和一个从小一起长大的朋友一起创办了第一家创业公司，跟心理治疗完全没关系。当时我们只是把它当成一段短暂的冒险——先做个创业项目，卖了就回去继续当治疗师。但没想到，我接触到了另一个世界：和一群超棒的人、超棒的团队一起工作，用技术去触达几十万、甚至上百万的人。

And then almost by coincidence I started my first startup company not related to therapy in any way with a childhood friend and we're really looking at this as a short time adventure, like we'll build a startup, sell it and I'll go back to practicing and I discovered the parallel universe like working with amazing people, amazing teams and then leveraging technology to reach hundreds of thousands of people and then eventually millions of people.

我的科技职业生涯非常幸运，参与过几个大规模的全球平台，最终服务了数以百万计的用户。

So, I was very, very lucky in my tech career to work on large scale global platforms that eventually reached millions of people.

第一家创业公司之后，我没回去做临床，虽然我们没把公司卖到几百万美元，花了超过一年时间，但之后我就没再回去，而是继续在其他公司工作。我做过出行、社交电商，甚至区块链公司，从这些经历中学到了很多。五年前，机缘巧合，我遇到了两位特别棒的联合创始人，其中一位和我一样主修临床心理学，于是我们就一起创办了Kai。我们的愿景是：把认知行为疗法（CBT）和其他正向心理学的有效工具和方法，以更低的门槛、更少的羞耻感、更低的成本，普及给大众。

So, after the first startup, instead of going back, we didn't sell it for millions of dollars and it took more than a year, but after that instead of going back to the clinic I stayed and worked for other companies, so I worked on mobility and social commerce, even blockchain companies, learned a lot by working for others, and then five years ago the stars aligned and I had like two amazing cofounders, one of them majoring in clinical psychology with me and that's how we started to work on Kai, and the vision was like bringing those amazing tools and techniques from CBT and other modalities, positive psychology to the masses, like reducing the friction, reducing the shame and the stigma, reducing the costs and making it accessible.

这就是

That's how

我们是在新冠疫情之前不久开始的。

we started a bit before COVID.

这太了不起了。

That's incredible.

我特别欣赏那些有临床经验、真正做过患者照护的医生去创业、主动行动，因为只有你们最清楚问题出在哪里、哪里可以改进，也知道该怎么打造一个真正对患者有意义的平台，而不是单纯为了赚钱或者盲目扩张。

I think I always love when clinicians, people who have actually practiced and done the patient care, start companies, take initiative, because you're the best person to really understand what the problems are, where there's room for improvement, how to start a company, how to make a platform well in a way that's meaningful to patients, instead of just trying to make profit or just, you know, spread.

真的非常了不起。

So, it's really incredible.

我很想再多了解一些，Kai这个平台具体是怎么运作的。

I'd love to hear a little more detail about exactly how Kai works, the platform you mentioned.

你们的目标是提升可及性、降低心理健康服务的成本，但能给我讲讲患者使用的过程吗？

It's to increase accessibility, decrease cost of mental health, but could you walk us through maybe the patient journey?

如果我有兴趣，该怎么开始使用你们的平台呢？

If I was interested, how would I go about using your platform?

是这样的，Kai是一种混合式服务，也是一个完整平台，我们并不取代临床医生。

Yeah, so Kai is a hybrid service and it's a full platform, so we do not replace clinicians.

我们使用自己训练的对话式AI和大型语言模型，确保这些工具都经过临床验证，帮助医生提升效率，让更多人获得支持。

We leverage conversational AI, large language models that we train ourselves and make sure that the tools are clinically validated to empower clinicians and make sure that every clinician can be more effective and reach more people.

用户的使用流程非常简单。

So, the user journey would be very straightforward.

我们通常与企业、医疗机构、大学和高中合作，由他们通过二维码或链接向用户推荐这项服务。

We usually partner with employers, with providers, with universities and high schools, and they will introduce the service to their audience, usually through a QR or a link.

你只需扫描二维码，就能直接在聊天中使用。

You would scan the QR and it works directly on messaging.

我们会你在苹果信息或WhatsApp上创建两个联系人。

So, we create two contacts on Apple messages or on WhatsApp.

其中一个完全是AI，你会看到Kai像普通联系人一样出现在你的通讯录里。

One is only AI, so you'll have Kai as a contact like any other contact.

Kai会主动发起对话。

Kai is proactive.

今天早上Kai就给我发消息了，说：早上好，亚历克斯，你昨晚睡得怎么样？

So, Kai wrote to me this morning, good morning, Alex, how did you sleep?

它知道我正在拉斯维加斯参加这次会议，工作特别忙，经常飞来飞去，它会支持我，帮我抽点时间放松，调整心态，保持积极和成长型思维。

It knows that I'm in Vegas for this conference and my work is hectic and I fly to different places, and it supports me and helps me with practicing, taking a minute for myself and focusing on like a positive mindset, a growth mindset.

它会时不时主动问我最近怎么样，评估一下我的状态。

It will check-in with me and will triage every once in a while to see how am I doing.

我们还设置了一个由真人临床团队组成的联系人，这个团队也能通过消息随时回应，不需要实时在线。

And then we create another contact that is a human clinical team and the team is available also on messaging asynchronically.

根本不用预约，也不用安排面谈。

So no need to schedule meetings, no need for like appointments.

临床团队还能调整AI的运作方式，比如面对青少年、老年人、护士或退伍军人时，他们会针对不同人群定制Kai的回应，有的侧重焦虑抑郁，有的专注成瘾问题。

And the team adjusts the way the AI works, so when we work with different audiences the clinicians will adjust Kai for teenagers or elderly people, for nurses or for veterans, very, very different use cases focusing on anxietydepression or focusing on addiction.

让临床专家能参与平台调整，这一点至关重要。

So, the fact that clinicians are able to work with the platform and adjust is very, very important.

作为用户，你首先会完成一个简单的初步评估，比如你目前最需要哪方面的帮助，主要困扰是什么，情绪压力的起点在哪里。

And then as a patient you will go through a basic intake of like help me like understand like what type of help do you need, what are the main challenges, what's the baseline of like emotional distress that you start with.

之后你可以选择语音或文字交流，操作特别简单，不用下载任何应用，真的管用，我很乐意再多跟你聊聊。

And then it works either on voice or on messaging, very simple to use, no need to install anything, yeah and it actually works, so I would love to tell you more about it.

是的，我觉得你这个双渠道设计特别好，一边是随时能答问题的AI，一边又知道什么时候该把问题转给专业团队。

Yeah, I really like the idea that you have the two contacts right, so then it's like we have an AI that's accessible twenty four seven that can answer questions, but knows when, okay, this is something out of my control.

让AI主动把情况标记给临床团队处理。

Let me flag the the clinical team.

这样就多了一层安全保障，系统本身也内置了这个机制。

So there's there's that there's that security layer that that is built into it as well.

对。

Yeah.

这一点真的特别关键。

That's super, super important.

目前大语言模型最重要的应用场景，就是提供情感支持。

Like, the number one use case for LLMs is for emotional support.

现实中太多人因为错误的建议、幻觉、糟糕的推荐，甚至被AI误导着去取代真实人际关系，最终做出错误的人生决定。我们坚信，唯一稳妥的做法，就是始终保留人类临床专家，让AI在察觉异常时及时发出预警，提醒人类介入。

There's so many bad cases of, like, people taking wrong life decisions and, like, hallucinations and and, like, bad recommendations and even sometimes bad motivations of like replacing your relationships that are so bad and we really believe that the only ways to make sure that we always, always have a human clinician and the AI can raise a flag and say, hey, something is happening, more attention is needed, and the human can intervene.

这一点非常重要。

That's very, very important.

我们还发现，把AI的声音和真人声音混在一起并不好，会让人感觉特别怪异。

And also we discovered that it's not helpful to mix between the AI and the human voice because then it creates this uncanny valley.

我到底是在跟谁说话？

Who am I messaging with?

谁能看到我的对话内容？

Who can see my conversation?

AI部分是加密且私密的，绝不会向你索要个人信息，但它会记住对话内容，并与临床医生协同工作；而临床医生是独立的通道，绝不与AI混用。

So, the AI is only the encrypted, private, will never ask you for personal information but will create memory and will work with the clinicians, and the clinicians are a separate channel, never mixed.

AI是全天候在线的，比如我可以说：‘我待会儿要录一个重要播客，现在有点紧张’，它马上就能回应。两个通道都支持语音，体验非常自然。

And the AI is 20 fourseven and I can say, hey, I have an important podcast and I'm feeling anxious and I will get a reaction right now and both channels work with voice as well so it's very, very natural.

比如你可以说：‘嗨，Kai，在吗？’

Say, hey Kai, are you here?

比如现在是晚上十一点。

Like it's eleven p.

嗯。

M.

我有时候睡不着，或者睡眠有问题，又或者刚和伴侣大吵一架，这时候团队会在后台默默调整计划，随时待命。

And I can't fall asleep, right, and I have challenges with sleep or I just had a terrible fight with my partner and the team is there in the background to customize the plan and be there when needed and

AI可以将情况转交给真人处理。

the AI can escalate to a real human.

这一点我很认同，临床团队能随时介入、灵活调整，真的特别关键。

Yeah, I love that like that downstream like tweaking of the clinical team, having them available is so important.

再往上游看，谈到AI产品本身，我们通常都会想，怎么和这类工具——比如ChatGPT——日常互动才自然。

When you think about it more upstream when you're talking about the AI product itself, we always think about how do we interact with these tools on a general level with like ChatGPT and things like that.

但这类工具常被批评太讨好用户，不管你说什么，它都附和你、让你感觉被认可，可有些场合、有些时候，这种反应反而有安全隐患。

And typically they've been critiqued for being like sycophantic, very like people pleasing, where like no matter what you say, it kind of validates you and makes you feel like it when certain places, certain situations, you know, that it could be a safety issue.

以你临床心理学家的背景来看，有哪些原则是你一直坚持的？

You know, as your background as a clinical psychologist, what are some of those principles that you held onto?

比如在训练Kai这类模型时，你具体遵循了哪些准则？怎么确定这个模型具备合格精神科医生的基本素质？

Like, when you're thinking about training these models for the uses of Kai, like, what were those, like, principles that you followed and like, okay, we need to make a model that meets these criteria in order for it to be a good psychiatrist?

我们明确不是在打造一个AI精神科医生，但伦理和核心价值观绝对至关重要。

So definitely, like we're not building an AI psychiatrist, but the ethical, like core values are very, very important.

我们是用真实的认知行为疗法、接纳与承诺疗法和积极心理学方案来训练的，Kai在需要时能进行激励式访谈。

So we trained on actual like CBT, ACT positive psychology protocols, Kai is able to do like a motivational interview when needed.

这一点非常、非常重要。

So, that's like very, very important.

我们生成了大量合成数据，用来训练不同性格和使用场景，因为每个人都不一样，这样能让临床团队更早发现特殊情况，并及时调整策略。

We generate a lot of synthetic data to actually train for different personas and different use cases because everyone is so different and this allows us to see edge cases much more, the clinical teams can see edge cases and adjust when needed.

另一个核心原则是，Kai会鼓励用户与他人建立联系、分享感受、主动寻求支持。

Another like very important core values that Kai would recommend connecting with other people, sharing with other people, reaching out to other people.

Kai还会建议用户参与真实的生活活动，比如锻炼、规律作息、多接触大自然。

Kai would recommend actual like life activities, working out, taking care of your sleep, spending time in nature.

所以，我们的核心关注点跟其他产品很不一样。

So, the core focus areas are very, very different.

我们从不追求延长互动时间，也没有人能访问你的数据，所有信息都是私密且匿名的。我们始终明确区分：你是在和一个工具对话，而不是在交朋友，更不是在寻找人类替代品——这一点至关重要，因为我们都清楚，有些糟糕的情况就是人们把AI当成了朋友甚至家人。

We don't have an aim to extend the interaction and like no one has access to the data, everything is private and anonymous, never confusing like who are you interacting with, not becoming your friend, not becoming like a human replacement is very, very important because we all know, right, that's like some bad cases would be like people acting as if the AI is a friend or a family member.

甚至有人还会向AI求婚。

They will even propose sometimes to the AI.

Kai跟别的东西完全不同。

Kai is very, very different.

你跟GPT互动，通常是问答式的，但跟Kai互动就完全不一样了——Kai会主动问什么、怎么回应，这些全都是由专业临床人员训练出来的。

So, you take an interaction with LGPT, usually a question answer type of interaction, an interaction with Kai, very, very different with what Kai would ask, how it will react, like all of that is trained by actual clinicians.

到目前为止，我们已经服务了超过二十万人，大多数都住在美国。

And we worked with over 200,000 people so far, most of them are here in The U.

不过，用户遍布全球各地。

S, but all over the world.

所以我们看到的多样性特别多，也有大量证据表明它确实有效。

So, again, we see a lot of variety and so many indications that it's effective.

我们已经发表了两篇研究论文，显示使用四周后，用户状况平均改善了百分之三十。

We published two research papers so far showing an actual improvement, 30% after four weeks.

我们刚完成了迄今为止规模最大的临床研究，设置了两个实验组，一组用Kai，另一组做面对面咨询，结果再次证实，焦虑、抑郁和物质滥用情况都显著减轻了。

We just completed the largest ever clinical study with two test groups leveraging Kai and comparing it to face to face interactions, and again, we see significant reduction in anxiety and depression, substance use.

连需要面对面看医生的次数也减少了，这一点非常重要。

In the need to see clinicians face to face, so that's very important.

这确实很有道理。

That makes a

我觉得就像你刚才说的，在定期关怀和提供心理支持之间，确实有个微妙的界限。

lot of sense.

如果处理不好，很容易让人误以为对面是个真人。

I think, like you said, there's a real fine line between checking in and sort of being that support structure versus being interpreted as a human.

正如阿文德和你提到的，现在太多人用聊天机器人或其他大模型用得太顺手了，结果反而聊了些不合适的内容，或者没有按照设计初衷来使用。

I think, as Arvind said and you said, too many people have become sort of too comfortable using tools like ChatGPT or other LLMs in that they're telling inappropriate things or they're sort of not using it in the intended way.

这可能会导致自杀念头，或者做出一些非常危险的行为。

It can lead to, you know, suicidal thoughts or really harmful actions.

所以你能考虑到这些，并且仔细调整模型，小心翼翼地把握分寸，这真的很棒。

So, it's really great that you consider all that and are sort of fine tuning your model to sort of walk that line very carefully.

我们后台始终有人工介入，所以我们有

We also always have a human in the background, so we have

一个全天候的安全团队。一旦AI识别到敏感词汇、情绪变化或负面情绪及心理困扰，系统会立即标记，临床医生就能及时介入。

a twenty fourseven safety team and it's like the AI will recognize like any like sensitive keyword or like changing sentiment or like negative emotions and emotional distress, it would be flagged immediately and the clinician can intervene.

这一点非常重要，不能只是同情或评判。

And that's very, very important, not judge a pity.

我很好奇，你提到过有个升级机制，但同时

I'm curious, like, you mentioned that, you know, there's an escalation feature, but at

聊天内容是加密的，临床医生也无法看到你和机器人对话的具体内容。

the same time, the chats are encrypted and sort of the clinicians can't see what you're talking to the bot about.

所以，假如有人提到一些需要升级的情况，比如有自杀念头或其他危险信号，背后的临床医生能看到什么？他们会收到哪些信息？

So, let's say, you know, someone mentions something that needs to be escalated, whether it's suicidal thoughts or anything, what does the clinician behind the scenes see and what information are

他们会被提供哪些信息？

they given?

我会把两种情况分开说：一种是AI在一般情况下检测到中等程度的焦虑或情绪困扰并发出提醒，这时临床医生是看不到你和Kai的聊天内容的，但Kai会为医生生成一份主要话题的摘要。

Yeah, so I would separate between the AI raising flags in a normal situation with like moderate levels of like distress and needs, so the clinician cannot read chat with Kai, but Kai would work for the clinician and will create a summary of the main topics.

还有最后的分诊评分——AI在执行PHQ抑郁量表、GAD焦虑量表，或者其他经过临床验证的评估工具方面非常出色，临床医生可以查看这些数据，并据此调整方案，比如告诉医生：Alex需要更多关注如何处理这段关系，或者采用某种特定技术，比如针对恐惧症的暴露疗法。

The last triage score, so the AI is very good in administrating PHQ assessment or a GAD assessment for anxiety or depression or any other clinically validated assessment and the clinician has access to those stats and is able to adjust and say Alex needs to focus more on managing this relationship or on a specific technique, right, like phobia and exposure around phobia.

这是常规的操作流程，临床团队每天早上会来复核、调整，优化干预方案。

So that's the normal use case and the clinical team will come in the morning and will adjust and review and will be able to tweak the interventions.

但当出现紧急情况时，情况就完全不同了。

And then something very, very different happens when there is an immediate emergency.

比如，有自杀念头就是一个很好的例子。

So, suicidal ideation is a good example.

临床医生会立即收到警报。

Clinician will be alerted 20 fourseven.

他们可以查看最后一次互动记录，但不会看到任何个人身份信息。

They will be able to review the last interaction without any personal information.

Kai 会问我希望怎么称呼我，这就像我的昵称，但我们不会问任何其他个人信息。

So Kai would ask me how would you like me to call you, that would be like my nickname, my name, but we don't ask for any other like personal information.

但在紧急情况下，如果临床医生判断需要立即干预，我们可以根据设备ID发起警报，而无需知道用户真实身份，这样相关部门就能在必要时实时介入。

But in those emergency cases when needed and the clinician thinks that an immediate intervention is needed, we will be able to escalate and ask for like an intervention based on the device ID without us ever knowing who is the person, but the authorities are able to intervene in real time when needed.

这一点非常重要：在紧急情况下需要立即干预时，系统会触发警报机制，允许采取行动。

So, that's very important, this balance when there is an emergency and immediate intervention is needed will raise the screen for like allow an intervention.

而在日常使用中，隐私始终是我们的首要考虑。

In day to day cases, like privacy is a top priority for us.

我们发现，当人们与Kai互动时，羞耻感和担心被评判的程度会降低。

What we discovered is that shame, worry of being judged is lower when people interact with Kai.

人们不会等到情况严重恶化才寻求帮助，因为Kai非常容易使用且随时可及。

People don't wait until things go terribly wrong to ask for help because it's so easy and accessible.

所以他们真的很信任它，愿意分享，而保护这种隐私和数据对我们来说至关重要。

So, they really trust it and share and it's super important for us to protect that privacy and the data.

我们获得了SOC 2认证，这是由外部机构进行的隐私与安全评估。

So, we have a SOC two certification which is an external privacy security assessment.

我们当然也遵循HIPAA和GDPR的相关规定。

We obviously work with HIPAA and GDPR.

这对我们的工作来说非常重要。

So, that's very, very important for us.

除了紧急情况，那是唯一我们会介入并确保当事人立即获得支持的时刻。

Besides like emergency cases where that would be the only instance where we're able to escalate and make sure that the person receives support immediately.

你之前已经提到过这一点。

You've already touched on this

一点点。

a little bit.

很想多听听。

Would love to hear more.

很多人对人工智能的顾虑是，不愿意分享太多个人信息，担心数据隐私问题，怕发生数据泄露。

I think one concern that a lot of people have with AI is they don't want to share too much personal information or they don't want to, like, they worry about data privacy or that there's going be a data leak.

你提到了你们符合SOC 2和GDPR标准，但你们在和患者、雇主以及医疗系统合作时，有没有遇到过他们对这种安全性和隐私保护的担忧？

You mentioned, obviously, your SOC two compliant, GDPR compliant, etcetera, but have you had concerns from patients or employers and health systems you're working with in terms of that level of security, the privacy that patients would be potentially risking?

这一直都是，一直都是

So, that's always, always

在开展这些合作时大家都会问的问题，不过我们一直非常重视，确保所有隐私和安全协议都到位，还做过渗透测试。

a question when we do those partnerships, but again, like we're very, good in making sure that all the privacy and security like protocols are in place and penetration testing.

我们和几家大型医疗管理组织合作过，他们对我们进行了严格的安全与隐私测试。

We partnered with large HMOs and we went through their like safety and privacy security testing.

至于用户，有意思的是，信任感是天然存在的，几次互动之后，人们就非常愿意分享和使用帮助，因为只有他们自己能接触到这些内容。

With users, the interesting thing is that the trust is there, and even after a few interactions, people feel very, very comfortable in, like, sharing and using the help because, like, no one besides them has access.

所以我们在这方面从来没遇到过问题。

So, we didn't have any issues with that.

恰恰相反。

It's the other way around.

我们发现，很多人每天都会和Kai互动。

Like what we see is that many people will interact with Kai every single day.

每天可能只有两到三分钟，但这种使用频率跟普通支持服务完全不一样。

It can be only for like two, three minutes a day but even that time, the frequency is very, very different than regular like support.

这种需求确实是存在的。

So, is definitely there.

是的，你之前提到过这一点，虽然只是简短带过。我觉得这背后其实涉及更深层的技术复杂性。比如今天早上Kai主动问你关于会议的事，还有你睡得怎么样，这让我想到，它能在没有人为触发的情况下，自动在合适的时间主动关心用户。

Yeah, and you mentioned this earlier I think in like very briefly that I think it actually represents a deeper technical complexity, as you mentioned that Kai this morning asked you about, you know, the conference and how you slept, and what that makes me think is like, without prompting, it's being able to know and interact and give you basically a check-in at some time.

尤其是在人们很难主动求助的场景里，一旦有人在恰当的时机主动联系，他们反而很愿意回应。我很想听听你对这一点的看法，因为从技术角度看，这要复杂得多，但却是这个领域里非常关键的一项功能。

That makes me think, especially in the space where it's often hard for people to reach out, but when they are reached out to in the right time, they're happy to interact, so I'd love to hear a little bit about that because that's a much more technical, like on the technical side it's a lot more complex, but it's a feature that is very powerful in this space.

没错，你得知道，我们在ChatGPT和类似模型出现之前就起步了。Kai从一开始就是结合了早晚、周末这些固定时间的规律互动，再加上当用户愿意深入交流时的自由对话。

Yeah, so you have to remember that we started before L and M's, before ChatGPT and Kai was always a combination between like morning, evening, weekends, like routines like scheduled like interactions and obviously the open conversation when someone shares more deeply.

这确实是我们觉得非常有帮助的一点。

And this is definitely something that we saw as very, very helpful.

人们愿意花一分钟做个简单的心理状态评估，因为他们也想知道自己最近怎么样，和上次比有什么变化。

People would take a minute and will answer a basic like well-being assessment because they also want to know how they're doing and what changed since the last time.

当你有这样一个提醒，哪怕只是停几秒钟，把注意力转向积极的事情，就能平衡我们日常生活中普遍存在的负面偏见——我们通常总在担心过去出错的事或未来可能出错的事，很少能自然地保持平和。能主动暂停、关注正面的东西，对建立心理韧性特别重要。同样，花一分钟做深呼吸，或者进行其他类似的干预，效果也是一样的。

And definitely when you have this prompt and you can pause for even a few seconds and tell your attention to something positive that's balancing the negativity bias that all of us like operate under in regular circumstances, like most of the time we're worried about like things that went wrong in the past or things that can go wrong in the future and we're not in a natural level and being able to proactively pause, pay attention to the good things is very, very important in like building psychological resilience and the same is true about like taking a minute and doing a breathing exercise or doing some other like act type of intervention.

如果没有提醒，人们就会忘记，不会主动关心自己。

Without the prompting, people forget, they don't invest in themselves.

我们早就知道，保持身体的柔韧性非常重要。

Many times we are aware that taking care of our physical flexibility is very, very important.

但直到最近，大家才开始意识到，心理的弹性也同样关键。

And I think only recently this understanding that our psyche and psychological flexibility is critical as well.

正如

As

我们之前提到过，心理健康服务的需求非常大，提高可及性尤为重要。

we mentioned earlier, there's a huge need for mental health care and so more accessibility of mental health care.

你一开始做临床心理学的时候，主要是在学校里接触孩子，但这个应用却是面向所有人的。

You started off doing clinical psychology with mostly, you said, children in schools, but this app is across everyone.

我想知道，哪里的使用效果最明显？

I'm curious, where have you seen the most traction?

有没有哪些人群、职业，或者特定群体用得更多，或者觉得更有帮助？

Are there certain demographics or certain professions or anyone that have used it more than others or have found more value?

是的，

Yeah, so

我觉得根本原因在于，全球有一百万名心理从业者缺口，很多人排着长队，心理咨询费用又高，根本得不到服务。

I think that everything starts with the fact that one million clinicians are missing all over the world and many people are on wait lists or like therapy sessions are very, very expensive and people don't get access.

所以我们很清楚地看到，俄乌战争一爆发，我们立刻把Kai翻译成俄语和乌克兰语，对接了会说这两种语言的治疗师，几天内就为数百个家庭提供了支持——这些家庭原本根本得不到任何帮助，那种日常的不安全感也减轻了不少。

So, thing that we saw very clearly is that when the war in Ukraine started, we were able to activate Kai in Russian and Ukrainian with therapists that are Ukrainian and Russian speaking and provide support for hundreds of families who couldn't receive support in any other way because we deployed in a few days basically and that's a very, very different audience and like put distress obviously that like something in the daily instability went away.

我们在年轻人中使用Kai也有丰富的经验。

We have a lot of experience with young people, with Kai as well.

我觉得这技术对他们来说特别直观好懂。

I think that the way the technology works is very intuitive for them.

我觉得Z世代应该是使用人数最多的群体。

So, I would say that Gen Zers would be like the largest group.

Kai的使用年龄从14岁到84岁都可以，你想想，我和我十几岁的女儿们用WhatsApp聊天，和我父母也用WhatsApp聊天，操作起来特别简单。

Kai works from the age of 14 to the age of 84, and when you think about it, I communicate with my teenage daughters on WhatsApp and messaging, and I communicate with my parents on WhatsApp and messaging, and that's very, very easy to use.

我们最早发表的一篇论文，研究了上万名14到18岁的青少年，他们使用Kai后情绪状态明显改善。

One of our first papers that we published, we published on 10,000 teenagers aged 14 to 18 working with Akai and improving their emotional well-being.

所以，大学生、Z世代，还有年轻科技从业者，应该是使用最广的群体。

So, would say that university students, college students, Gen Zers, like young tech employees would be the biggest court.

就连护士也一样，他们轮班工作，压力很大。

Even nurses and right, they're working in shifts.

他们常常经历失去和压力，但很少有时间去接受情感支持，更别说去接受心理咨询了。

They are experiencing loss and stress many times.

可要是随身带着一个随时在线的‘心理助手’，对她们来说简直是天大的改变。

And like many times they don't take the time to receive emotional support and go to therapy themselves, but having this almost psychologist in your pockets that is available 20 is a game changer even for them.

我们看到的使用场景和年龄层非常广泛，老年人更多是面对孤独、悲伤和失落，尤其是从忙碌的工作状态过渡到退休生活的时候。

So, we see a wide range of use cases and age groups, elderly people more around loneliness, grief, loss many times, switching from like active work mode to retirement.

有太多不同的使用场景了。

So, so many different use cases.

年轻人面临社交媒体成瘾，还有

Young people around social media addiction, around

有时候是人际关系问题，新手父母应对育儿挑战，应用场景太多了。

sometimes relationships, young parents with like parenting challenges, so many use cases.

当像你之前提到的那样，心理咨询不仅预约难，经济负担也重，有些保险能覆盖，可以报销，这确实能减轻不少治疗费用。

When you have the situation that you alluded to earlier with therapy being not only hard to get with appointments, but also financially restrictive in many ways, you know, and certain insurances can cover it and they can, you know, be reimbursed, And I think that can oftentimes alleviate the costs of therapy.

现在像Kai这样的技术，通过AI平台推出，但仍由临床专家主导，你们有没有和保险公司合作，争取对这种经过验证的治疗方式提供报销？

Now that we're having technologies like Kai that are being rolled out on an AI integrated platform that is still clinician led, are you able to work with insurance companies or get these kind of like reimbursements of validated ways of applying therapy?

目前还没有，不过我们已经开始和一些机构合作，其中一个

So, not yet, like we started to work with providers and one of

最早的应用场景，显然是缓解排队等候的问题。

the first use cases was obviously supporting the waitlists.

很多诊所排队时间很长，与其让人等三到六个月，有些地方甚至超过一年，现在能立刻提供支持。

Many clinics will have long waitlists instead of telling someone to come back in like three months, six months, in some places it's even more than a year, they can provide immediate support.

对于部分患者来说，轮到他们时，其实已经不需要面对面就诊了，这样能大幅降低成本。

For some of those patients, when their turn comes, they don't need to come for face to face meetings and it reduces costs dramatically.

但我们并不是直接提供治疗，也不只是靠AI，所以不像那种CBT应用那样能给患者直接使用，目前还没有报销机制。但我相信这一定是未来，长远来看，保险公司最终会覆盖这项服务，因为成本实在太低了。

But because we're not providing therapy per se and we're not working only with AI, there's no like CBT app you give the patient, we don't have reimbursement at this point, but I believe that this is the future, this is the future and eventually our long term vision is like the insurance companies will cover it because the cost is so much lower.

比如Kai，当我们自己配备临床团队时，每月收费大约20美元。

Like Kai, when we provide our own clinical team, we charge around $20 a month.

当我们与医疗机构合作、使用他们的团队时，Kai的授权费每月只要5美元。

When we work with a provider with their team, a Kai license is $5 a month.

所以，当我在考虑不同规模的模式时，

So, it's a different scale when I So think about

你们是直接向治疗机构、临床心理团队这些客户销售，然后由他们来提供访问权限吗？

are you selling to like therapy or therapy groups, clinical psychology groups and then they will allow access?

是的，我们会提供培训，和他们紧密合作，给他们整套操作系统，他们可以

Yeah, we'll provide the training, we will work with them, we provide them the operating system and they have like

使用完整的管理后台，邀请自己的临床医生和患者加入，而我们则在后台提供支持，帮他们顺利运行。

the full dashboard, they can invite the clinicians and they can invite their patients and we stay behind the scenes and help them operate it.

还有一些机构希望我们同时提供临床团队，我们也会这么做，

And there's some organizations that need us to provide the clinical team as well, and we'll do that, we'll

我们会需要这样做，

do that, we'll need that.

显然，Kai 已经产生了巨大影响，不仅仅是在美国，

Clearly Kai is already making a huge impact, it's not just in The U.

在全球各地，都有大量患者在使用它。

S, around the world, you have so many patients using it.

你对未来一年、未来五年最期待的是什么？你希望实现哪些目标？

What are you most excited about in the next, you know, next year, next five years, and what do you hope to achieve?

我们看到，随着 AI 和大语言模型的发展，许多专业人士从担心被 AI 取代，逐渐转变为开放接纳，意识到 AI 可以成为增强效率的工具——作为临床医生，我可以运用自己的专业知识和经验，服务更多人，提供更广泛的支持。

So, we really see this change with AI and LLMs from many professionals really worrying that AI is going to replace them to more like openness and this understanding that AI can be a force multiplier and as a clinician I can use my knowledge and experience and my skill and reach more people and provide support for more people.

所以我特别期待更多临床工作者能接受这项技术，因为患者们其实已经在用了。

So, I'm really, really looking forward to like more clinicians adopting the technology because the patients are already using it.

2025 年，大语言模型最主要的用途是情感支持和陪伴，但如果缺乏专业监督，这种做法可能带来危险和风险。

The number one use case in 2025 for LLNs is emotional support, is emotional companionship and when it's done without like clinical supervision, it can be dangerous, it can be risky.

我们都看过那些关于糟糕案例的报道。

We all read like the stories about like bad cases.

所以我最期待的是，专业人士、医生和保险公司能意识到这项技术在患者护理中的潜力，让像Kai这样的服务能惠及更多人。

So, I'm mostly looking forward to the professionals, clinicians, insurers recognizing the potential in caring for the patients and making sure that services like Kai are accessible to more people.

没错。

Right.

现在主流越来越认可，使用的人也越来越多，毕竟效果是实实在在的，只是还需要让更多人用得更安心，而且

It's getting more mainstream approval and more usage because, yeah, it's clearly working, just there needs to be more comfortable and it

这种趋势正在慢慢推进。

to increase and that's happening slowly over time.

我认为关键不在于用机器取代人类，而在于保持高度透明，遵循明确的伦理规范。

Yeah, I think that the key is not replacing the humans in working with a lot of transparency and working with clear ethical guidelines.

我们现在正和美国心理学会合作启动一项质量认证，确保Kai的运作方式安全可靠，并获得APA的正式认可，这一点我觉得非常重要。

We're now starting this certification process at badge of quality with the APA to make sure that the way Kai works is safe and can be approved by the APA as well, and this is something that I think is very very important.

在结束之前，非常感谢你带来这么精彩的对话。我们总是喜欢问嘉宾，你们是怎么给公司起名Kai.ai的？很想听听背后的故事。

Well as we wrap up, thanks so much for this amazing conversation, we love asking our guests too about how you came about with the name of your company, we'd love to hear the story behind Kai.ai.

我们真是太幸运了，虽然我多希望是我取的名字，但其实不是。

We were super, super lucky, wish I could say that I named the company but that's not the case.

我们一开始用的名字就是‘Kai’。

We started with a different name, with the name Kai.

这个名字源自希腊神话，讲的是一个 healer 和教练，还跟一片星群有关，是个特别积极向上的名字。我跟一个朋友说了，他是 Fiverr 和 Lemonade 的联合创始人，也是个特别棒的艺术家和产品负责人，他就跟我说：‘亚历克斯，这个关于 Kai 的故事真好，你干脆就用 Kai 做公司名字吧。’

And like that's a Greek mythology like a story about like a healer and a coach and constellation of stars, a very, very positive name and I told it to a friend that is one of the founders of like Fiverr and Lemonade and like is an amazing artist and product leader and he told me, listen, Alex, like this story about Kai.

Ron 特别棒，你应该讲讲这个故事，但更重要的是，用 Kai 做名字再合适不过了——短、直接，还暗含了 AI。我们特别幸运，域名刚好没被注册，就拿下了。后来我们发现，这个名字让每个人都能拥有自己的‘Kai’：有人觉得它像智慧的老尤达，有人觉得像年轻的妹妹，有人当它是女性，有人当它是男性。简单、好记、有亲近感，对我们来说，简直是完美的名字。

Ron is wonderful, you should tell it, but you should name the company Kai because Kai is very, very short, it has the AI and we're so lucky that the domain was available and we were able to like get it and then what we discovered is that this allows every person to have their own Kai and it's so simple and easy for some like Kai users, like Kai is like an old Yoda wise person, for others it's their young sister, for some of them it's female, for others it's male and so short and easy to remember and so relatable that it was the perfect name for us.