Natural language processing
Hi, everyone. Another week without Donald Trump’s tweets and Facebook posts. At least we have clips of his “perfect” speeches replayed in the Senate trial. Good times.
This is a special free edition of Plaintext. To read future subscriber-only columns, subscribe to WIRED (50% off for Plaintext readers) today.
The Plain View
I first met Dan Siroker 13 years ago, when I literally went around the world with a group of young Google product managers. It was a memorable trip. But Siroker might not recall it as well as I do. A few years ago, he learned that he suffers from aphantasia, an inability to visualize images. This blindness in “the mind’s eye” makes remembering things more difficult. His discovery came not long after he was diagnosed with a separate condition that affected his hearing. The latter could be addressed by technology—the former could not.
The discovery led him to study memory, and he found that, for the most part, whether or not some things are best forgotten, we forget them anyway. Research shows that we don’t remember 90 percent of what happened the previous week. The pitiful state of our memories had led us to invest billions of dollars in pencils and notepads. But Siroker, a software engineer and entrepreneur, felt there was a better way. So he set out to build “the modern equivalent of a hearing aid for memory.”
To make that happen, he started a company called Scribe.ai, hoping to supplement our neurons to make even the most mundane occurrences into something unforgettable. “Our long-term ambition is to provide perfect recall of every memory,” he says. “We want to help you remember everything.”
Siroker won’t be drilling into skulls to rewire brains. Instead, the plan is to methodically capture and store all sorts of data—audio, video, and eventually biometric—that can be easily searched or cleverly invoked in a way that augments your actual memory with stuff you otherwise wouldn’t have possibly remembered, unless you were Marilu Henner.
That’s the long term-vision. Scribe’s first product is a more modest effort. “For practical purposes, we’re focusing on dominating a niche and growing from there,” he says. That niche is an add-on product to Zoom, transforming the audio and video from the platform into an exceptionally accessible data trove. “Meetings are a good place to start,” he says, explaining that his product will free people to concentrate on the subject and interact with others. “We’ll take care of the remembering,” he adds. When you invite Scribe to a meeting—it shows up as a faceless participant—you have a dynamic rapporteur who not only logs what people say and what they look like as they say it, but will eventually be able to dive into previous meetings or other corpuses to find relevant snippets of conversation or documents. “It’s like having a chief of staff whispering in your ear,” says Siroker.
Scribe’s impressive list of investors, including luminaries from Facebook, Google, and Y Combinator, approve of Siroker’s gradual approach to total recall. “When people talk about merging with AI or computers, they always think of the Elon Musk Neuralink approach,” says Sam Altman, cofounder of Open AI (with … Elon Musk). “But we've already somewhat merged with technology. Our phones somewhat control our behavior, and we're willing to outsource a lot of our decisionmaking and memory already,” adds Altman, who was first to contribute to Scribe’s initial $5 million funding. “I don't memorize facts anymore, because I know I can just get whatever I need quickly on the internet.”
But there are perils of offshoring one’s memory, chief among them a privacy concern. While outsiders can’t delve into our brains to access our memories, they can certainly plunder the servers that store the personal histories that Siroker hopes we will preserve via Scribe. And many of those conversations will be recorded passively, through inputs like the increasingly pervasive microphones in devices like Alexa. Or the augmented-reality devices that companies like Facebook and Apple are developing. Or biometric recording devices.
Siroker says that he’s very privacy-conscious, and all of those digitally stored memories will go into “your own personal vault.” He’s also thinking a lot about how to make sure the people you interact with don’t feel you’re stealing their words. He doesn’t want Scribe to become something that wipes out the concept of “plausible deniability.” And so he’s thinking of giving people mulligans. “You can go back and delete something you said earlier if you don’t like it,” he says. “It’s not a permanent record, but something you have control over.”
Wait a minute … if people can mess with your memories, or you can edit them yourself, doesn’t that give us the power to alter history? Siroker says he does not want people to alter records in a way that supports nefarious uses like deepfakes. But that sentiment doesn’t deal with the fact that storing one’s history means capturing someone else’s.
I doubt that these futuristic concerns will hamper Scribe’s foray into the Zoom add-on market. In fact, the initial product—now in private beta and widely available later this year—seems pretty useful without making us hyperthymesiacs. When using Scribe with Zoom, you can not only easily find everything uttered in a meeting but perform analytics. For instance, you can instantly create a pie chart exposing who is dominating the conversation and who isn’t saying much. And it’s simple to use the tool to create a highlight reel of a meeting that can be shared to those who weren't in attendance.
If Dan Siroker’s dream of total recall gets put aside while he builds an empire of similar business tools exploiting documentation, there will be no need to apologize about not taking on the larger mission. We’ll probably forget he brought it up.
Before starting Scribe and an earlier company called Optimizely, Siroker worked on the 2008 Obama campaign, bringing Google-style web analytics to its fund-raising efforts. I wrote about this in my history-of-Google book, In the Plex—which just happens to be updated in paperback edition this month!
At campaign headquarters in Chicago, Siroker began looking at the web efforts to recruit volunteers and solicit donations. His experience at Google gave him a huge advantage. “I’d worked on Google ads, a huge system, which probably only three people in the world—even at Google—truly, fully understand,” he says. “It’s the mentality of taking data and trying to figure out how to optimize something.” The Obama web operation was run by smart people who’d picked up tech skills along the way but were not hardcore engineers. “I was probably the only computer science degree in the whole campaign,” he says.
Siroker became the chief analytics officer of the Obama campaign. He saw his mission as applying Google principles to the campaign. Just as Google ran endless experiments to find happy users, Siroker and his team used Google’s Website Optimizer to run experiments to find happy contributors. The conventional wisdom had been to cadge donations by artful or emotional pitches, to engage people’s idealism or politics. Siroker ran a lot of A/B tests and found that by far the most success came when you offered some swag—a T-shirt or a coffee mug. Some of his more surprising tests came in figuring out what to put on the splash page, the one that greeted visitors when they went to Obama2008.com. Of four alternatives tested, the picture of Obama’s family drew the most clicks. Even the text on the buttons where people could click to get to the next page was subject to test. Should they say “Sign up,” “Learn more,” “Join us now,” or “Sign up now”? (Answer: “Learn more,” by a significant margin.) Siroker refined things further by sending messages to people who had already donated. If they’d never signed up before, he’d offer them swag to donate. If they had gone through the process, there was no need for swag—it was more effective to have a button that said “Please donate.” There were a lot of reasons why Barack Obama raised $500 million online to McCain’s $210 million, but analytics undoubtedly played a part.
Someone posted a picture of Siroker on his Facebook wall on election night. Everyone else at campaign headquarters was cheering or crying with joy. Siroker was sitting at his computer with his back to the TV, making sure that the new splash page that would welcome website visitors was the one celebrating the victory, not the one they’d prepared saying he’d lost. After that, he was going to push the start button on yet another test, to see which one of four victory T-shirts would be the most effective in garnering donations for the Democratic National Committee. Just as Google ad campaigns never ended, neither did online political campaigns.
Ask Me One Thing
Shumon asks, “Is social media moderation alone a sustainable solution to the growing disinformation/misinformation problem? Isn’t the real problem that people can’t spot obvious disinformation?”
Thanks for the question, Shumon. Of course, eliminating disinformation on social media—or even eliminating social media itself—wouldn’t solve the larger problem of people believing crazy or destructive stuff. There are plenty of ways that intentional lies can be spread in order to mislead people. (Have you heard talk radio in, oh, the last 20 years?) Your second question—which, like the first, seems to be one where you seem to think you know the answer already—also makes an obvious point: A more skeptical, evidence-based populace might not embrace disinformation so readily. But I’m not willing to let social media off the hook. When a social network’s algorithmic system recommends groups and “news” sources that promote lies and conspiracy theories, there’s a danger that even media-literate people will be lured to accept them—not only because it’s kind of exciting, but also for the comfort of connecting with new friends who believe the same thing. After all, it’s social media, even if the content is antisocial. So Facebook and its brethren have to do better.
You can submit questions to email@example.com. Write ASK LEVY in the subject line.
End Times Chronicle
Lawyer of the week is hands-down the cat-filter attorney. Trump should have hired him for his Senate defense!
Last but Not Least
We’re wearing two masks now? OK.
Hard to imagine anything more stressful than dealing with a severely premature baby—during Covid.
The Cyberpunk 2077 maker isn’t giving in to a ransomware threat. What were the hackers going to do, make it buggier?
Do you miss offices? Read about their “secret maps.”
Don't miss future subscriber-only editions of this column. Subscribe to WIRED (50% off for Plaintext readers) today.