評(píng)論
打印
Sometimes when you're browsing your social media feed, you might find an account or a post that just seems...off.
當(dāng)你瀏覽社交媒體時(shí)，你有時(shí)可能會(huì)發(fā)現(xiàn)賬戶或帖子似乎...離線
Maybe it sounds like a janky product endorsement, or maybe the wording seems kinda funny.
也許這聽(tīng)起來(lái)像是產(chǎn)品廣告，或者說(shuō)法有點(diǎn)好笑。
Quite often, these are bots — bits of computer code designed to operate an account and maybe masquerade as a real person.
通常情況下，這些都是機(jī)器人——設(shè)計(jì)用來(lái)操作賬戶的計(jì)算機(jī)代碼，可能偽裝成真人的樣子。
They're all over the place, especially on sites like Twitter,
這些機(jī)器人無(wú)處不在，尤其是在像Twitter這樣的網(wǎng)站，
where some have estimated up to 15% of active accounts are actually bots.
一些人估計(jì)，多達(dá)15%的活躍賬戶實(shí)際上是機(jī)器人。
You've probably heard they're shaping our online experience and could even have real-world consequences— but how true is that?
你可能聽(tīng)說(shuō)機(jī)器人正在塑造我們的網(wǎng)絡(luò)體驗(yàn)，甚至可能會(huì)對(duì)現(xiàn)實(shí)世界產(chǎn)生影響——但這是真的嗎?
And why do we fall for them in the first place? It turns out that for something so rooted in computer science,
為什么我們一開(kāi)始就迷戀上了機(jī)器人呢？事實(shí)證明，這有計(jì)算機(jī)科學(xué)根據(jù)的，
the success of bots has a lot to do with, well, psychology.
機(jī)器人的成功與心理有很大關(guān)系。
To understand the influence of Twitter bots, it helps to look at the world of artificial intelligence, or AI.
要了解Twitter機(jī)器人的影響，我們可以看看人工智能世界。
For decades, researchers in this field have been trying to convince people their computer programs are actually other humans.
幾十年來(lái)，這一領(lǐng)域的研究人員一直試圖讓人們相信，他們的電腦程序?qū)嶋H上是其他人。
A common benchmark of how human-like an AI is called the Turing test.
我們把與人相似的人工智能的常見(jiàn)基準(zhǔn)叫做圖靈測(cè)試。
In it, someone interacts with a computer and a person just through text, and they have to see if they can tell the difference.
在這個(gè)游戲中，有人通過(guò)文字與電腦、人互動(dòng)，看自己能否分辨出其中的區(qū)別。
So far, we're not really to the point of AIs reliably passing.
到目前為止，人工智能可靠傳遞的程度還沒(méi)有實(shí)現(xiàn)。
Over on the main SciShow channel, we talked a bit about why this is so difficult from a computer science standpoint,
在心理科學(xué)秀主頻道，我們從計(jì)算機(jī)科學(xué)的角度討論了為什么這個(gè)程度很難實(shí)現(xiàn)
and we'll link to that video at the end of this one.
本節(jié)目最后我們會(huì)給出視頻鏈接。
But even though there's no such thing as a perfect AI, some programs have done relatively well.
但即使沒(méi)有完美的人工智能，一些程序也做得相當(dāng)不錯(cuò)。
And obviously, tons of people mistake Twitter bots for humans all the time.
很明顯，許多人一直把Twitter機(jī)器人當(dāng)成人類。
In competitions and otherwise, the most successful programs seem to trick us by managing our expectations.
在競(jìng)爭(zhēng)以及其他方面，最成功的項(xiàng)目似乎通過(guò)把控人類的期望來(lái)對(duì)欺騙我們。
In other words, they're designed to look like people we wouldn't expect to be perfect linguists.
換句話說(shuō)，人們把Twitter機(jī)器人設(shè)計(jì)成我們不會(huì)期待的完美語(yǔ)言學(xué)家的樣子。
For example, one successful early bot posed as a seven-year-old child,
例如，早期成功的機(jī)器人偽裝成7歲的孩子，
so spelling and grammar mistakes seemed normal.
所以存在拼寫(xiě)和語(yǔ)法錯(cuò)誤似乎很正常。
And in 2014, one of the best AI at a competition at the University of Reading was supposed to be a Ukrainian teenager.
2014年，在雷丁大學(xué)的一場(chǎng)競(jìng)賽中，最優(yōu)秀的人工智能之一應(yīng)該是一名烏克蘭少年。
That way, it got some slack when it didn't get things like pop culture references.
這樣一來(lái)，當(dāng)沒(méi)有流行文化參考時(shí)，步伐就會(huì)有些緩慢。
Managing expectations works in competitions, and it can explain why Twitter bots fool us, too.
管理期望在競(jìng)爭(zhēng)中有效，這可以解釋為什么Twitter機(jī)器人也愚弄了我們。
Many bots, especially the ones that don't self-identify,
很多機(jī)器人，尤其是那些不會(huì)自我識(shí)別的機(jī)器人，
rely on the fact that you probably aren't looking for them when you go online.
會(huì)依靠這樣的一種事實(shí)，當(dāng)你上網(wǎng)的時(shí)候，你可能不會(huì)去尋找機(jī)器人。
After all, if you notice tweets with bad grammar or weird hashtags — well, that's just Twitter.
畢竟，如果你注意到微博的語(yǔ)法很糟糕或者標(biāo)簽很奇怪——好吧，那就是Twitter。
So the bots slip under your radar. But that's not the only psychological trick in their toolbox.
所以這些機(jī)器人在你的雷達(dá)下逃過(guò)了一劫，但這并不是他們唯一的心理技巧
To pick up human followers, many accounts— especially advertising bots — take advantage
為了吸引粉絲，許多賬戶——尤其是廣告機(jī)器人——利用了
of another well-studied phenomenon in social psychology: the norm of reciprocity.
社會(huì)心理學(xué)中另一個(gè)廣泛研究的現(xiàn)象：互惠標(biāo)準(zhǔn)。
This says that if someone does something nice to you,
就是說(shuō)如果有人對(duì)你做了好事，
you tend to be a little motivated to do something nice in return.
作為回報(bào)，你往往會(huì)做一些好事。
So, if you get a new follower,
如果你有了新追隨者，
you might follow them back without thinking too hard about whether they're a person or a program.
你可能會(huì)跟隨它們，而不去想它是人還是程序。
Then, the bot can just wait until it has a few thousand followers
然后，機(jī)器人會(huì)等待，直至幾千名追隨者進(jìn)入
— and you've forgotten about it — to start spamming your feed.
——你已經(jīng)忘記了——開(kāi)始發(fā)垃圾信息。
And it seems to work. Researchers have found that, although people don't tend to reply to these tweets directly,
這似乎很有效。研究人員發(fā)現(xiàn)，盡管人們不會(huì)直接回復(fù)這些推文，
they do regularly retweet them and spread their messages.
他們會(huì)定期轉(zhuǎn)發(fā)和傳播信息。
This could be because they're posting something noteworthy, or just because they targeted potentially interested users.
這可能是因?yàn)樗麄儼l(fā)布了一些值得關(guān)注的東西，或者僅僅是因?yàn)樗麄兠闇?zhǔn)了潛在的感興趣用戶。
Generally, this isn't a big issue. Whether or not you recognize them for what they are,
一般來(lái)說(shuō)，這個(gè)問(wèn)題不大。不論你是否認(rèn)識(shí)到問(wèn)題的本質(zhì)，
a lot of bots just exist to serve you ads or share animal pictures.
很多機(jī)器人只是為了給你提供廣告或者分享動(dòng)物圖片。
But there are also more worrisome programs you might've heard about,
但你可能聽(tīng)說(shuō)過(guò)一些更加令人擔(dān)憂的項(xiàng)目，
which seem to be trying to spread misinformation and influence politics.
他們似乎在試圖傳播錯(cuò)誤信息并影響政治。
And those are trickier to understand. Despite concerns, it's really hard to measure if these bots have offline consequences —
這些問(wèn)題更難理解。盡管存在擔(dān)憂，但很難衡量這些機(jī)器人是否會(huì)產(chǎn)生離線后果——
like if they actually sway anyone's vote. It's a topic of active research and debate.
就像他們真的影響了任何人的投票一樣。這是一個(gè)研究和辯論的熱門(mén)話題。
Either way, all that political spam probably isn't a good thing,
不管怎樣，所有這些政治垃圾郵件可能都不是好事，
and many social networks like Twitter are working hard to keep bots from skewing the political climate in one direction.
Twitter等許多社交網(wǎng)絡(luò)都在努力阻止機(jī)器人扭曲政治方向。
Still, like we said earlier, bots aren't all bad news.
不過(guò)，就像之前說(shuō)的，機(jī)器人并不都是壞事。
Some share self-care tips, others help with customer service, and some share legitimately useful information.
一些人分享自我照顧的技巧，一些人協(xié)助客戶服務(wù)，一些人分享合法的有用信息。
Some new ones can even help — at least, to a degree — with mental illness.
一些新的治療方法甚至可以幫助——至少在一定程度上——治療精神疾病。
In 2017, some psychologists followed in the footsteps of one of the very first chatbots: ELIZA.
2017年，一些心理學(xué)家追隨了最早的聊天機(jī)器人之一——伊麗莎（ELIZA）的腳步。
This bot was designed to be a bit of a parody of the therapists who would just say
伊麗莎（ELIZA）被設(shè)計(jì)成治療師的樣子，他們只會(huì)說(shuō)
That's interesting... tell me more.
這很有趣…告訴我更多！
Armed with new research about cognitive behavioral therapy, or CBT, this team took ELIZA a step further.
有了認(rèn)知行為療法的新研究，這個(gè)研究小組進(jìn)一步研究了伊麗莎（ELIZA）。
They designed a new chatbot named Woebot designed to walk users through some of CBT's tools and strategies.
他們?cè)O(shè)計(jì)了一款名為Woebot的新型聊天機(jī)器人，旨在引導(dǎo)用戶了解認(rèn)知行為療法的一些工具和策略。
Generally, CBT is a kind of therapy that
一般來(lái)說(shuō)，認(rèn)知行為療法是一種治療方法，
focuses on changing your emotions and feelings through challenging your patterns of thoughts and behavior.
它通過(guò)挑戰(zhàn)人的思維和行為模式來(lái)改變?nèi)说那榫w和感受。
But you also need, like, a person to be the therapist. Or apparently… a chatbot.
但你也需要人來(lái)做治療師，或者顯然…聊天機(jī)器人。
In a randomized control trial of 58 people,
在58人的隨機(jī)對(duì)照試驗(yàn)中，
the team found that those who interacted with Woebot for two weeks
研究小組發(fā)現(xiàn)，與對(duì)照組相比，那些與Woebot互動(dòng)兩周的參與者
reported a significant reduction in depression symptoms on a standardized self-report scale compared to a control group.
在標(biāo)準(zhǔn)化自我報(bào)告量表上的抑郁癥狀明顯減輕。
Even though they all knew it was... just a robot.
盡管他們都知道那是…只是一個(gè)機(jī)器人。
The conversation probably didn't go exactly like it would with a human therapist,
此次談話可能和人類治療師的談話不太一樣，
but it looks like it was close enough to be effective. So even though some can cause problems,
但看起來(lái)足夠接近于有效。所以即使有些會(huì)出現(xiàn)問(wèn)題，
bots aren't all bad, and we're moving in some interesting and helpful new directions with them.
機(jī)器人也不全是壞事，我們正朝著一些有趣而有益的新方向前進(jìn)。
If you really don't want to be fooled by a bot,
如果你真的不想被機(jī)器人愚弄，
you can look out for accounts that use repetitive wording and post too many links or hashtags.
你可以尋找重復(fù)措辭，發(fā)布許多鏈接或標(biāo)簽的賬戶。
But a lot of more sophisticated ones do go undetected,
但是更多更復(fù)雜的還沒(méi)有發(fā)現(xiàn)，
and even researchers have a hard time telling the difference.
甚至研究人員也很難分辨出其中的區(qū)別。
So chances are… you're probably interacting with some bots from time to time online, whether you want to or not.
所以，不管你是否愿意，你都有可能時(shí)不時(shí)地與一些機(jī)器人進(jìn)行在線互動(dòng)。
But as long as you keep in mind some of the ways they can trick you into engaging with them,
但只要你記住了它們的一些招數(shù)，你就能和機(jī)器人交往，
you can pick the ones you want to interact with. And that's definitely a start.
你可以選擇你想要與之互動(dòng)的對(duì)象，這當(dāng)然只是一個(gè)開(kāi)始。
Thanks for watching this episode of SciShow Psych!
感謝收看本期心理科學(xué)秀！
If you'd like to learn more about how Twitter bots work and why it's so hard to program them to sound like humans
如果你想了解更多關(guān)于Twitter機(jī)器人如何工作以及為什么把機(jī)器人設(shè)計(jì)成人的思維模式如此艱難
you can watch the first part of this episode over on the main SciShow channel!
你可以在心理科學(xué)秀主頻道觀看本集的第一部分!
You can find it at youtube.com/scishow.
可以訪問(wèn)youtube.com/scishow。
