Del Harvey: The strangeness of scale at Twitter

Summary

The speaker from Twitter discusses the immense scale of user activity on the platform, emphasizing the challenge of maintaining user trust and safety amidst a sea of tweets. With the exponential growth in tweets, even rare edge cases become commonplace, requiring constant vigilance to prevent harm. The talk highlights the complexity of identifying abusive content, spam, and phishing without context, and the importance of designing protections for the unexpected. The speaker illustrates this with examples, including the humorous use of 'yo, bitch' and the serious implications of photo geodata, advocating for proactive risk assessment in technology development.

Takeaways

• 🛡️ The speaker's job at Twitter is to ensure user trust, protect user rights, and keep users safe from various threats, including themselves.
• 📈 Twitter has experienced exponential growth in tweets, from 2 million per day in 2009 to 500 million tweets in less than six minutes by 2014, representing a 24,900-percent increase.
• 🔎 The vast majority of Twitter activity is harmless, but the sheer scale means even rare events become common, emphasizing the need for vigilance.
• 🤔 The role involves identifying and preventing harmful activities, which can be challenging due to the scale and the innocuous nature of most tweets.
• 🐶 Context is crucial in determining the intent behind tweets, as seemingly abusive language can have benign uses, such as role-playing or referencing popular culture.
• 🚫 Combating spam is complicated by the fact that what appears to be spam could be a legitimate notification, like an alert for the International Space Station.
• 🔗 Phishing attempts, which try to steal account information, must be distinguished from important messages, like those sharing evidence of police brutality.
• 🤔 The speaker must constantly evaluate and reassess account behaviors, considering hundreds of parameters to avoid wrongful suspensions.
• 💡 The importance of designing protections for the unexpected is highlighted, as even the most innocuous features, like photo sharing, can have unforeseen risks.
• 🐱 The example of a cat photo illustrates the potential dangers of metadata, such as geolocation, which Twitter proactively removes to protect users.
• 🌐 The speaker emphasizes the need to visualize potential catastrophes and build safeguards against them, even when dealing with positive and harmless interactions.

Q & A

What is the primary role of the speaker at Twitter?

The speaker's primary role at Twitter is to ensure user trust, protect user rights, and keep users safe from each other and sometimes from themselves.

How has the volume of tweets on Twitter changed from January 2009 to January 2014?

The volume of tweets on Twitter increased from more than two million new tweets each day in January 2009 to more than 500 million in January 2014, which is a 24,900-percent increase.

Why does the speaker emphasize the importance of considering edge cases at Twitter's scale?

The speaker emphasizes the importance of considering edge cases because at Twitter's scale, a one-in-a-million chance happens 500 times a day, making these rare situations more like norms.

What percentage of tweets are estimated to pose no risk to anyone according to the speaker?

According to the speaker, 99.999 percent of tweets pose no risk to anyone.

How does the speaker describe the challenge of dealing with the scale of tweets at Twitter?

The speaker describes the challenge as dealing with a tiny percentage of tweets that, after removing the vast majority that pose no risk, still amounts to roughly 150,000 tweets per month.

What is one of the unexpected uses of the phrase 'Yo, bitch' on Twitter mentioned by the speaker?

One of the unexpected uses mentioned by the speaker is when people are role-playing as dogs on Twitter, where 'Yo, bitch' is used as an accurate greeting.

Why does the speaker mention the difficulty in identifying spam accounts on Twitter?

The speaker mentions the difficulty because what seems like spam could also be a legitimate notification, such as one about the International Space Station passing overhead.

What is the potential risk of automatically suspending accounts that send the same message and link to many people?

The potential risk is that it could be a bystander sharing crucial information, like a video of police brutality, and suspending the account would silence that important speech.

How does the speaker approach the task of designing protections for the unexpected on Twitter?

The speaker approaches this task by visualizing catastrophe and thinking through how even benign actions, like adding a photo to a tweet, could lead to harmful outcomes.

What measure did Twitter take when they launched the ability to add photos to tweets to protect users?

When Twitter launched the ability to add photos to tweets, they decided to strip out the geodata from the images to protect users from potential harm based on location information.

How does the speaker's worldview remain positive despite focusing on the worst-case scenarios?

The speaker's worldview remains positive because the vast majority of interactions they observe on Twitter are positive, with people reaching out to help, connect, or share information.