AI 安慰得更到位，
你为什么还是想找朋友？
If AI Sounds More Caring, Why Do We Still Choose People?
深夜想倾诉，AI 秒回，朋友却可能明天才看到。哪一种安慰更好？一项心理学研究发现：人们对回复内容的评分，和他们想找谁倾诉，未必指向同一个答案。
研究团队访谈发表于2026年10月 · CEFR参考难度：B2
AI · EMPATHY · CHOICE
逐段精读
Close Reading
01 / 06
A bad day ends, and you want to tell someone about it. A chatbot may answer at once; a friend may reply much later. Which would you choose? Psychologists studied a related question: when people can choose an empathy source, do they pick a human or AI? The answer revealed a choice paradox. What people chose was not always what they rated most highly.
译文
糟糕的一天结束，你想找人说说。聊天机器人可能立刻回复，朋友却未必在线。你会选谁？心理学家研究了一个相关问题：当人们能选择安慰的来源时，会选真人还是 AI？选择的对象，与事后评分最高的对象，并不总是一致。研究者把这种反差称为一种“选择悖论”。
释义与用法
empathy source · 共情回应的来源
source of support 支持来源；这里比较回应来自真人还是 AI。
choice paradox · 选择悖论
paradox 指看似矛盾的现象；这里是“选择”与“评价”方向不同。
一句话拆解
What people chose was not always what they rated most highly.
what 引导两个名词性从句；not always 表示“不总是”，避免说成“从不”。
02 / 06
Across four studies, the team compared people's choices with their ratings of caring replies. Some situations were imagined; one study also asked people to describe events from their own lives. Participants could choose a human responder or an AI responder. Later, they judged how empathetic the replies seemed and whether the words made them feel heard. These were controlled study settings, not ordinary chats with close friends.
译文
研究团队在四项研究中，把参与者的选择与他们对安慰回复的评分放在一起比较。有些情境是想象的；其中一项还请参与者写下自己的真实经历。参与者可以选真人或 AI 来回应，之后再评价回复是否有同理心、是否让自己感到被倾听。实验比较的是研究情境中的回应，并不是与亲密朋友日常聊天的直接测试。
释义与用法
human responder · 真人回应者
respond to someone 回应某人；responder 是作出回应的人。
AI responder · AI 回应者
这里指生成安慰文字的 AI；不是有真实情感的人。
一句话拆解
The team compared people's choices with their ratings of caring replies.
compare A with B 是“把 A 与 B 比较”；choice 和 rating 是两个不同的测量。
03 / 06
In one study with 196 university students, participants chose a human reply in an average of 57.8 percent of trials. Yet they generally rated AI replies as more empathetic and more likely to make them feel heard. The number is an average across choices, not the percentage of students who always chose humans. This is the striking result: people often wanted a human source, even when they liked the AI-written message more.
译文
在一项有 196 名大学生参与的研究中，真人回复平均在 57.8% 的选择轮次中被选中。然而，参与者总体上给 AI 回复更高的共情评分，也更倾向于认为它让自己感到被倾听。57.8% 指选择轮次的平均比例，不是“57.8% 的人永远只选真人”。这正是研究中最有意思的反差。
释义与用法
57.8 percent of trials · 57.8% 的选择轮次
trial 是一次实验轮次；不要误读成参与者人数占比。
feel heard · 感到被倾听、被理解
feel + 过去分词描述感受；例如 feel understood 感到被理解。
一句话拆解
People often wanted a human source, even when they liked the AI-written message more.
even when 引出反差：更喜欢某条回复，不等于更愿意选它的来源。
DATA SNAPSHOT
57.8%：选真人的轮次占比
在第二项研究的 196 名大学生中，参与者平均有 57.8% 的选择轮次选了真人；这不是“57.8% 的参与者选择真人”。同一研究里，AI 回复在共情与“被倾听”评分上更高。
蓝色横条仅表示选真人的平均轮次比例；两类评分使用不同指标，不能画在同一标尺上。
04 / 06
Why might a polished AI reply score well? The researchers asked how much perceived effort each responder seemed to put into a message. Replies judged as more effortful were also judged as more empathetic. Surprisingly, participants often perceived AI replies as more effortful than human ones. But this does not settle why people preferred humans; the study did not identify one simple cause for that preference.
译文
为什么写得周到的 AI 回复容易得高分？研究者还让参与者判断：每位回应者看起来花了多少心思。被认为更用心的回复，通常也被认为更有同理心。出人意料的是，参与者往往觉得 AI 回复显得更用心。但这不能单独解释，人们为什么仍更常选择真人。
释义与用法
perceived effort · 感知到的用心程度
perceived 强调“人们觉得如此”，不是对实际付出的直接测量。
perceived · 被认为；被感知到
perceive A as B 表示“把 A 看作 B”。
一句话拆解
Replies judged as more effortful were also judged as more empathetic.
两个 judged as 并列；说的是评分之间的关联，不能直接写成因果。
05 / 06
Here the title needs a careful pause. A friend knows your history, voice and shared jokes. Most human replies in these experiments came from people without that relationship context. The paper therefore cannot tell us that AI comforts people better than their own friends. It also cannot show what happens after months of talking to the same chatbot. A reply's quality and a personal connection are related questions, but they are not identical.
译文
读到这里，标题需要停一下。朋友知道你的经历、语气和只有你们懂的玩笑。实验里的多数真人回复并没有这样的关系背景。因此，研究不能证明 AI 比你自己的朋友更会安慰你，也没有检验长期与同一个聊天机器人交谈会怎样。一条回复写得好，与一段关系带来的连接感，是两个不同的问题。
释义与用法
relationship context · 关系背景
context 是理解一段话所需的背景；此处包括共同经历与熟悉程度。
personal connection · 人与人之间的个人连接
feel a connection with someone 表示“与某人有连接感”。
一句话拆解
A reply's quality and a personal connection are related questions, but they are not identical.
but 连接相关与不同两层意思；identical 比 similar 更强，指“完全相同”。
06 / 06
So what can we take from the paradox? AI may offer useful words when nobody is available right away, and it may supplement human support. At the same time, the study shows that many people still value choosing a human listener. If a friend says only, 'I'm here,' that may mean something a perfectly written message cannot capture. The study invites us to ask not only whether a reply sounds caring, but also whom we want beside it.
译文
这个悖论能带走什么？当身边暂时没有人可聊时，AI 也许能提供有帮助的话语，成为真人支持的补充。同时，研究显示，许多人依然看重自己能否选择一位真人来听。朋友只说一句“我在”，也可能带着一条漂亮回复无法替代的关系意义。判断安慰是否有用，可以同时问：话说得怎样？我希望是谁在说？
释义与用法
supplement · 补充
supplement human support 指补充真人支持，而不是证明可以替代它。
listener · 倾听者
a good listener 善于倾听的人；这里强调回应者的身份。
一句话拆解
The study invites us to ask not only whether a reply sounds caring, but also whom we want beside it.
not only ... but also ... 连接两个问题：回复内容与回应者身份。
KEEP THESE WORDS
本期重点表达
词组若没有单独的 CEFR 等级，不把组成单词的等级硬套给整个词组。
empathy · 心理学名词｜CEFR C2
同理心；共情
常用搭配：show empathy / an empathetic reply
例句：A caring reply can show empathy.
用法：研究测量的是人们对回应的共情评分，不是 AI 真的拥有情感。
prefer · 动词｜CEFR A2
更愿意选择
常用搭配：prefer A to B / prefer to talk to someone
例句：Some people prefer a human listener.
用法：prefer 描述选择偏好；与对文字质量的评分可以不同。
feel heard · 词组｜CEFR 未单列
感到被倾听
常用搭配：make someone feel heard
例句：That reply made me feel heard.
用法：heard 是 hear 的过去分词，表示自己的感受被认真接住。
responder · 名词｜CEFR 未标级
回应者
常用搭配：a human responder / an AI responder
例句：The responder wrote a caring message.
用法：既可以指真人，也可以指系统；文中核心是来源差异。
perceived effort · 研究词组｜CEFR 未单列
感知到的用心程度
常用搭配：rate perceived effort
例句：Readers rated how much effort each reply seemed to take.
用法：perceived 指受试者的主观印象，不是实际成本。
paradox · 名词｜CEFR C2
悖论；看似矛盾的现象
常用搭配：a choice paradox / an apparent paradox
例句：The choice and rating results formed a paradox.
用法：这里不是逻辑错误，而是两个指标呈现不同方向。
relationship context · 词组｜CEFR 未单列
关系背景
常用搭配：lack relationship context
例句：A close friend knows the relationship context.
用法：实验中的真人回应不能等同熟识的朋友。
supplement · 动词｜CEFR 未核定
补充
常用搭配：supplement human support
例句：AI may supplement support when friends are unavailable.
用法：supplement 不等于 replace；避免把研究读成“AI 可以替代朋友”。
TWO QUESTIONS
一条回复好，和想找谁聊，是两回事
A better-rated reply does not automatically become the reply we choose.
这项研究测量了选择偏好与回复评分；它没有直接比较 AI 与你自己的亲密朋友，也不能证明 AI 已经取代真实关系。
TRY IT YOURSELF
读后想一想
1. 为什么“回复得分更高”不等于“人们更愿意选择它”？
Why does a higher rating not always mean a more popular choice?
2. 57.8% 指的是参与者比例，还是选择轮次的平均比例？
What exactly does 57.8 percent describe in the study?
3. 如果朋友暂时不在线，你会如何使用 AI，同时保留真实的人际联系？
How might you use AI when a friend is unavailable while keeping human connections?