WinkLingo
ログイン

The Social Dilemma · 2020 — 第 7 / 14 章 · 3'36" · C1

AIs Plot Ben's Resurrection, Explained：a mind of its own

AI voices in a control room treat one user's attention as a problem to be solved: "We need something actually good for a proper resurrection, given that the typical stuff isn't working." Every usual nudge has failed, so they calculate that one option — a notification about Ana — has a 92.3 percent chance of resurrection.

あらすじ

In a control room, AI characters review a user who has gone quiet. Tagged photos, invitations, current events, even a direct message have failed. They cross-reference him against 13,694 similar users, then calculate that one option — a notification about Ana — has a 92.3 percent chance of resurrection. Cut against this, a voice explains what algorithms are and who they serve.

MeetingsExplainingSuggestingC1mediumamericanClean
IMDb ↗

注目の表現

switch toas far as the eye can seecommercial interestin order to

00:00 / 03:36
Space 再生 · ← → セリフ移動 · R リピート
いま
  6. ❝ 独白The algorithm warning31 行練習ページ ↗
セリフをタップするとそこから再生

このシーンの表現

a mind of its own

acting independently

The narrator says a person writes the algorithm, yet it behaves as if it has its own will. The tone is explanatory, not joking.

My calendar seems to have a mind of its own this week.

このシーンの他の表現

  • switch to

    change to

  • as far as the eye can see

    to the horizon

  • commercial interest

    business motive

  • in order to

    so as to

  • cross-reference

    compare across records

このシーンのメモ一覧

50 行のセリフに 29 件のメモ

表現 · B1switch to something

change to

Someone sees the numbers failing and calls for a hard switch into a win-back mode; the tone is a quick decision, not a debate.

No. Something's very wrong. Let's switch to resurrection mode.
発音switch to

Practice linking switch and to as one unit, with no pause between them.

No. Something's very wrong. Let's switch to resurrection mode.
スラング · C1resurrection mode

user-winback mode — Product-team jargon for re-engaging lost users; avoid with outsiders.

No. Something's very wrong. Let's switch to resurrection mode.
表現 · B2what people miss is that ...

the overlooked point is that ...

And what people miss is that AI already runs today's world right now.
表現 · B2run the world

controls how the world works run means to operate or drive, not to jog

And what people miss is that AI already runs today's world right now.
文法 · B2what people miss is that

subject clause: The what-clause is the subject; the real claim starts after is that.

And what people miss is that AI already runs today's world right now.
発音is that

Keep is that light and put the beat on AI already.

And what people miss is that AI already runs today's world right now.
表現 · B2as far as the eye can see

to the horizon

The narrator describes absurdly large server rooms and pushes rows of machines to the horizon; the tone is like a tour that turns overwhelming.

Tons and tons of computers, as far as the eye can see.
表現 · B1tons and tons of something

a huge amount of

Tons and tons of computers, as far as the eye can see.
表現 · B1back and forth between ...

to and fro

sending information back and forth between each other all the time.
表現 · B2be described as something

be called

some could be described as algorithms that are so complicated, you would call them intelligence.
表現 · B2be embedded in something

built into code

I like to say that algorithms are opinions embedded in code...
表現 · C1be optimized to + 目标

tuned toward

Algorithms are optimized to some definition of success.
表現 · B1someone's definition of success

what counts as success

Algorithms are optimized to some definition of success.
表現 · B2a commercial enterprise

business company

So, if you can imagine, if a... if a commercial enterprise builds an algorithm to their definition of success,
表現 · B2machine learning

machine learning

That's where the term "machine learning" comes from.
表現 · B1in the right order

in the correct sequence

And so, every day, it gets slightly better at picking the right posts in the right order so that you spend longer and longer in that product.
文法 · B1so that you spend longer and longer

purpose clause: So that links better ordering to a result: users stay longer.

And so, every day, it gets slightly better at picking the right posts in the right order so that you spend longer and longer in that product.
表現 · B1in order to do something

so as to

The speaker says nobody truly understands what the machine is doing, then adds the goal with in order to. The tone is explanatory, almost forensic.

And no one really understands what they're doing in order to achieve that goal.
表現 · B2have a mind of its own

acting independently

The narrator says a person writes the algorithm, yet it behaves as if it has its own will. The tone is explanatory, not joking.

The algorithm has a mind of its own, so even though a person writes it,
文法 · B2even though a person writes it

concession clause: Even though sets up a contrast: people write the code, yet the system can still slip away.

The algorithm has a mind of its own, so even though a person writes it,
表現 · B2a handful of people

a small number of

There's only a handful of people at these companies, at Facebook and Twitter and other companies...
文法 · B2who understand how those systems work

relative clause: The who-clause defines the few people; how those systems work is what they understand.

There's only a few people who understand how those systems work, and even they don't necessarily fully understand what's gonna happen with a particular piece of content.
表現 · B2lose control over something

no longer control

So, as humans, we've almost lost control over these systems.
スラング · C1psychometric doppelgangers

lookalike psychological profiles — Data-team jargon for users with matching behavior profiles; not for daily talk.

His psychometric doppelgangers.
表現 · C1long-term engagement

lasting user involvement engagement means user involvement here, not a marriage promise

My analysis shows that going political with Extreme Center content has a 62.3 percent chance of long-term engagement.
発音62.3 percent

Say the number and percent in one breath, stressing 62.3.

My analysis shows that going political with Extreme Center content has a 62.3 percent chance of long-term engagement.
スラング · C1Extreme Center

radically centrist — A half-satirical political label; use with care in formal settings.

My analysis shows that going political with Extreme Center content has a 62.3 percent chance of long-term engagement.
表現 · B1send someone a direct message

private message

Okay, okay, so we've tried notifying him about tagged photos, invitations, current events, even a direct message from Rebecca.

「AIs Plot Ben's Resurrection」を読み解く

What's really going on

The joke of the control room is that the AIs sound like product managers, not villains: "My analysis shows that going political with Extreme Center content has a 62.3 percent chance of long-term engagement." Engagement is the number being optimized, so a person becomes a probability — "a 92.3 percent chance of resurrection." The lines that cut in make the argument underneath: "algorithms are opinions embedded in code" and "algorithms are not objective." Someone chose the target, and the target is time spent in the product. What makes it unsettling is that even the people running the system don't fully understand it: "no one really understands what they're doing in order to achieve that goal."

Theme

PowerA handful of people understand how these systems work, while the systems optimize for a commercial definition of success and control the information users see.
SatireThe AIs speak flat product-manager jargon while engineering one person's return, which is the joke.

発音のポイント

  • switch to — Practice linking switch and to as one unit, with no pause between them.
  • is that — Keep is that light and put the beat on AI already.
  • 62.3 percent — Say the number and percent in one breath, stressing 62.3.
And what people miss is that AI already runs today's world right now.

The danger is not a robot in the future; these systems already decide what we see today.

I like to say that algorithms are opinions embedded in code...

Ranking is not neutral: someone's judgment about what counts as a good outcome is baked into the code.

Algorithms are optimized to some definition of success.

Every algorithm chases a target a person picked; for a company that target is usually profit or time spent.

So, as humans, we've almost lost control over these systems.

Even the builders can't fully predict what a system will do with one post, so control is largely gone.

このシーンについての質問

Why do the AIs switch to "resurrection mode"?

Because the usual prompts stopped working — tagged photos, invitations, current events and a direct message have all failed, so they try something stronger.

Why does a notification about Ana matter?

After the other nudges failed, the system calculates a 92.3 percent chance of resurrection with a notification about Ana, so that is the option it puts forward.

Who benefits when a company defines what its algorithm should achieve?

The company. One speaker says a commercial enterprise builds an algorithm to its own definition of success, so it serves a commercial interest, and "It's usually profit."

Why isn't the 62.3 percent plan used first?

One line calls it "not good enough to lead with" — a decent probability isn't strong enough to open with, so they keep looking.

How many people understand how these systems work?

Only a handful of people at these companies, and even they don't necessarily fully understand what will happen with a particular piece of content.

監視資本主義: デジタル社会がもたらす光と影 · 全 14 章

すべて見る →
次のチャプター · 第 8 章reckon withThe Wikipedia Analogy
第 1 章rakes in moneyYou Are the Product第 2 章human futuresSurveillance Capitalism Explained第 3 章dial up the adsTech Insiders Explain the Growth Dials第 4 章fall prey toPinterest President Can't Put Phone Down第 5 章let the record showBen Takes the One-Week Phone Bet第 6 章conflateWhy Social Media Harms Teens再生中 · 第 7 章a mind of its ownAIs Plot Ben's Resurrection第 8 章reckon withThe Wikipedia Analogy第 9 章tilt the floorFake News Spreads Faster Than Truth第 10 章weaponizing social mediaFacebook Weaponized in Myanmar第 11 章digital FrankensteinsSpeaker Reframes the Tech Threat第 12 章willful ignoranceThe Civil War Interview第 13 章put the genie back in the bottleInterviewees Call for Regulation第 14 章set in stoneTech Insiders Explain What Went Wrong
第 7 / 14 章 · 残り約 19 分