Tyler Le/Insider
I get more emails than I can handle. Every day, I watch as the unread messages in my inbox pile up despite my best efforts. It's not because I spend too much time responding to the emails that matter — I spend too much time reading the ones that don't.
There are newsletters, brand-outreach spam, press releases, recruiters, social-media updates — emails that are almost never urgent but still make me feel obliged to check to make sure I'm not missing anything important. Even carefully deployed filters go only so far to keep the ever-growing volume of unwanted emails at bay. One study found that more than half of our emails were not important, and another study found that one-quarter of an average American's workday was spent answering and reading emails. This cascade of digital messages is making us miserable.
The disaster that is the modern inbox is a prime area for disruption — and companies are racing to cash in using the hottest new tech: artificial intelligence. A company called Shortwave, founded by a team of ex-Googlers, has created a tool that uses AI to sum up your emails for you. It spits out a simple summary of all your important communications in an attempt to save you time. I decided to try it out to see whether AI could really solve my email problem, and to an extent, it did.
The tool goes to show that while there's a long way to go, AI can help improve our online lives. By eliminating the need to manually take care of menial tasks like reading every email, AI can help us be more productive so that we can focus on more important and creative aspects of our lives and work.
The AI revolution comes for email
Similar to Google's now defunct Inbox product, Shortwave organizes your inbox by prioritizing conversations with people and tucking away the rest into their own bundles, such as "newsletters," "travel," and "purchases." Its AI addition, which launched last week, runs on GPT-3, the language model behind the viral chatbot ChatGPT. Since the Microsoft-backed research firm OpenAI made ChatGPT public in November, over 100 million people have used it to generate movie plots, write their college essays, and code computer programs from scratch. Now companies like Shortwave are building products on top of the tech.
When I log into Shortwave, which costs $9 a month for the full product and offers a free limited version, there is a "summarize" button at the top of each email. When I select it, it "reads" the email's content and condenses what it's about into a few sentences to give me a TL;DR version. For example, someone sent me a report on the 10 most successful crypto countries. Instead of going through the email, which included numerous figures and tables that would have taken awhile for me to read, I asked Shortwave to summarize it. In about half a minute, the app strung together a 70-word digest of the report, pulling from its highlights. In this case, the summary gave me enough information that I decided to move on without reading the entire email. The app told me that I had saved two minutes.
That may seem insignificant, but in the few weeks that I have used the tool, those minutes have added up to nearly three hours of saved time over a week — even though I sometimes used the "summarize" button just a few times a day.
Shortwave's tool has especially come in handy for lengthy email chains — instead of reading each new email, I can check the AI's summary, which sums up the entire thread, to see whether any of the new emails concern me. If I want to loop in another person, there's an option for me to share the summary with them. It also works great for newsletters I want to stay on top of but don't have the time to read. Once I have an overview of what a newsletter is about, it's easier for me to decide whether I'm interested enough to go deeper.
Shortwave also plans to eliminate the time we spend on emails that require only a few words, like ones confirming a meeting's schedule or acknowledging a receipt. The company let me test a feature that will be rolled out in the coming months: AI-generated responses. When someone sent me a message and asked me to let them know I got the message, I would type "confirm receipt" in the email body and click the "smart edit" button. Shortwave would then expand my message into a proper email with a greeting and a signature, with options to instantly make it sound more cheerful or formal as needed. In my experience, it's far more useful than Gmail's auto-generated responses, which are limited to a couple of words and don't offer customization.
Yoon Kim, a computer-science professor at MIT, told me that for quick emails, automated tools like this would boost worker productivity. This is the main vision behind Shortwave. Jacob Wenger, Shortwave's cofounder and head of product, summed up his thinking: "You should not have to fully read and interact with every email in your inbox to get to inbox zero."
'AI will free us from the mundane and repetitive tasks'
Shortwave isn't alone in its mission to automate emails. Over the past year, several firms have capitalized on ChatGPT-like language models to write people's emails for them, automatically generate meeting notes, and negotiate their internet bills on their behalf. Microsoft is set to add a summary tool to Outlook, and Google has for years offered tools for Gmail that help you compose emails. Since 2021, the productivity startup Flowrite has used GPT-3 to compose entire emails for you.
HyperWrite, an email-writing tool by the startup OthersideAI, has taken AI-generated text even further. It not only generates email text but also learns how a user writes and which kinds of topics they generally talk about so that they can spend less time editing the AI. Matt Shumer, OthersideAI's CEO and cofounder, expects personalized AI models to generate more accurate and relevant information than a generic system.
Beyond email, AI's ability to consume loads of information and break it down will address some of the web's fundamental issues. Anton Borg, a computer-science researcher at the Blekinge Institute of Technology in Sweden, told me that AI would automate routine tasks and lower the skill level required to use tools like photo-editing software. Ultimately, Shumer told me, "AI will free us from the mundane and repetitive tasks that take up a lot of our time and energy."
No longer will you have to navigate a list of links on a search engine to pinpoint what you're looking for; a chatbot will fetch a final answer for you. Don't want to read a lengthy document? Let AI do it for you. Need to design a social-media post? Punch in a couple of keywords, and Canva's tools can produce art and text out of thin air. Tools like Heyday will sort your online research for you and help you better remember what you read. Similarly, GitHub's Copilot enables programmers to write code quicker and eliminate the need for manually developing basic and mundane functions.
That list will continue to grow this year. On March 1, OpenAI released a new interface that will allow businesses to add ChatGPT to their services, making it easier for more companies to utilize AI. Snap is one of its early adopters and rolled out a chatbot for Snapchat+ subscribers called "My AI," which answers questions and offers recommendations like birthday gift ideas for a friend.
Limitations of AI
Many people are rightly skeptical about an AI revolution, citing issues like the tendency of ChatGPT and Bing's chatbot to "hallucinate" or incorporate the biases of the text they're trained on in answers. Text generation is tricky. As soon as an AI begins to communicate under the guise of a human's name, it enters an ethical gray area, Vincent Conitzer, a computer-science professor at Carnegie Mellon University, told me.
A well-written email signals many things, Conitzer said. It signals the person took the time to compose it and that they care about the topic. We might lose these important signals if AI takes over. To address this, Conitzer told me it'd be "helpful to have proof that the message was written by a human being."
Like any tech, AI tools can run into bugs. On many occasions, Shortwave's summary tool refused to produce more than a sentence or two, leaving it no more useful than the subject line. A couple of times, it didn't produce a summary at all because of an "unexpected error."
A more-pressing concern for me, however, is what the presence of AI in email would mean for privacy. This has been an issue with ChatGPT, since the company can use what people say to the chatbot to train AI models or put information in its future conversations, as it did when it gave away a journalist's phone number. I was, therefore, worried that the AI model would leak my emails and that they would show up in someone else's ChatGPT conversation. Companies across the world, including OpenAI's largest investor, Microsoft, have asked their employees to not share any sensitive data with ChatGPT. Shortwave, though, has been trying to get in front of those issues: According to Wenger, the email text you summarize goes to OpenAI, but it's never stored. More importantly, Shortwave's "summarize" tool is optional and doesn't activate by default. Only the emails I choose to summarize are processed, which is why, even though the risk of compromising my identity is little, I never deployed it for emails that contained sensitive information.
Because tech like ChatGPT collects and processes far more data than previous AI tools, Conitzer expects it to be regulated for not only privacy but also systemic biases. When someone asked ChatGPT to write code that would filter people from genders and races who would make good scientists, it set "male" and "white" as the two prerequisites. Experts have raised alarm over such biases and factual inaccuracies augmenting the spread of conspiracy theories and disinformation online. For AI to have more widespread use, it's important that the companies building these models are cognizant of these problems and that regulators make sure there's an independent check on these platforms.
Despite these difficulties, Kim, the MIT professor, agreed that "these technologies could profoundly change our internet activities." For now, I'm looking forward to not spending hours sifting through a mountain of unwanted emails. The increasing use of AI models, such as GPT-3, is allowing the tech to make a real impact on our online world. And if Shortwave's AI tool is any indication, I'm convinced it's our future.
Shubham Agarwal is a freelance technology journalist from Ahmedabad, India whose work has appeared in Wired, The Verge, Fast Company, and more.