Your guide to a better future
Companies like Snap and Google are infusing generative AI into their messaging apps.
Generative AI is starting to show up in messaging apps.
If you’re like me, you proofread texts at least three times before texting someone you’re not very close with, like an old acquaintance or a former colleague. I’ve even asked my husband to look over a draft of a text to get a second opinion on whether I’m striking the right tone. I also find myself pinging between apps to find answers to questions that come up in my text threads, such as where to get dinner on the cheap after work.
It doesn’t have to be that way. Google has a new feature that could help out. So does Snapchat. And before long, many other messaging apps probably will too.
If all that really does make texting easier, you’ll have generative AI to thank.
Tech companies and app developers are betting that generative artificial intelligence, or AI that creates content based on training data when prompted, can upgrade the texting experience. Various types of AI have long been a staple of the tech products and apps we use everyday, but the tech has had a breakout moment in 2023.
Ever since OpenAI’s AI-powered chatbot ChatGPT gained widespread popularity in late 2022, tech giants have been racing to build new products, services and features that use generative AI to boost productivity, whether that’s by enhancing your search results or drafting your next email.
While much of the discourse around generative AI has focused on work-related tasks like writing emails, conjuring cover letters and penning social media posts, Google, Meta and Snap are exploring the technology’s usefulness in improving the way we communicate more casually, too. It’s a natural progression, since new technologies tend to gain widespread popularity on the device we carry in our pockets everyday: the smartphone.
“When we see any kind of big trend emerge in the world, we generally see that it sort of dovetails really nicely onto mobile rather quickly,” Lexi Sydow, head of insights at app analytics firm Data.ai, said when asked why generative AI is starting to arrive in messaging apps. “Because mobile is so ubiquitous, we have so many smartphones, and for so many folks around the world it is their one sort of portal to the internet.”
But exploring these new uses for AI can also open up new concerns around privacy and the authenticity of our future text exchanges.
AI isn’t new to messaging; if you’ve ever used your iPhone’s autocorrect or voice dictation to send a message, you’ve probably used some form of AI. But app-makers are now experimenting with generative AI to add another layer to the messaging experience, such as rewriting your text in a different tone or introducing chatbots to your inbox.
Snapchat’s My AI chatbot can recommend a nice spot for dinner.
Earlier this year, Snapchat parent company Snap Inc. added a chatbot called My AI to its popular ephemeral messaging app that uses OpenAI’s language model to generate responses. Snap positioned My AI as a tool for assisting with social obligations like coming up with birthday gift ideas or planning a hiking trip. I tried using My AI for advice about how to approach reaching out to friends who never responded to a party invitation. It told me I should send a follow-up message to see if they’re still interested. The bot also reassured me that sending another message a week after the initial invite wasn’t too soon.
In the two months since My AI was released broadly, Snap says, more than 150 million people have sent 10 billion messages to My AI looking for party planning tips, design inspiration, food and travel recommendations and more. While you could just turn to ChatGPT or Google for such advice, perhaps the idea is to provide access to these insights within the same app you might already be using to communicate with friends about upcoming parties or travel plans. That’s not the only way Snapchat is leveraging generative AI; Snap Inc. previously announced new Lenses (that is, Snapchat’s augmented reality effects) based on the technology.
Meta, which owns Facebook, Instagram and WhatsApp, is working on AI-powered digital agents for use in Messenger and WhatsApp designed to help and entertain people, according to a Meta spokesperson. It’s also developing features that will allow users to create chat stickers and modify their own photos for sharing on Instagram Stories through text prompts. These developments were first reported by The Verge and Axios, which said the company discussed such features at a company meeting in June. The spokesperson also said these generative AI experiences are “under development in varying phases.”
This follows remarks from Meta CEO Mark Zuckerberg, who posted on Facebook in February that the company was creating a new “top-level product group” to “turbocharge” its generative AI efforts.
Google’s Magic Compose gets its bard on.
Google’s approach is a bit different. Instead of employing an AI-driven chatbot in its Messages app (although Google does have a separate conversational AI program called Bard), the company is introducing an optional tool that uses AI to suggest responses or rewrite text messages in a different tone. The tool can also suggest responses with the tap of a button.
Called Magic Compose, the feature can recast your text in Shakespearean prose or make it sound more professional. Google announced Magic Compose at its developer conference in May and has since launched it in its Messages app in beta.
While part of the inspiration for Magic Compose came from Google’s efforts to sprinkle more personalization into its Messages app, the company’s user research also found that people don’t always know how to initiate a conversation over text.
“What we’re hearing is users just appreciate maybe some additional nudge and additional creativity from their messaging app to basically make those conversations go more smoothly,” said Jan Jedrzejowicz, product management lead for Google Messages.
Magic Compose offers seven options for rewriting your texts: Remix, Shakespeare, Lyrical, Chill, Formal, Short and Excited. Google started with these options because they strike the right balance of “helpful” and “fun,” according to Jedrzejowicz. A simple text like “Hey, what are you up to?” turns into a cute little rhyme when you choose the Lyrical option.
Magic Compose can give you some blank verse, too.
Google isn’t alone. Apps like FireTexts and TextAI that use AI to help craft texts have also arrived in the App Store and Google Play Store. While lesser-known texting apps like these may be slow to gain traction, Sydow said there’s been a notable uptick in app downloads for general AI-powered chatbots and apps. Microsoft’s Bing mobile app, which the company updated with generative AI features earlier this year, has been downloaded 20.4 million times globally so far in 2023, says Sydow, citing Data.ai’s research. That’s compared to roughly 3.6 million downloads in 2022.
Still, since tools like My AI and Magic Compose are new to messaging apps, it’s difficult to discern how popular and impactful they may be. The 150 million Snapchat members that have used My AI so far equates to roughly 20% of the 750 million active users Snap announced in February. Jedrzejowicz also couldn’t say much about the feedback Google has received on the Magic Compose beta since it’s still early in the process.
Kiran Raj, the practice head of GlobalData’s disruptive-tech research, can see Magic Compose being useful for professionals who want to be “precise with the appropriate tone,” he said in a response to emailed questions. But Raj also pointed out that privacy may be a concern when using tools like these. Google, for example, must access some of your older texts in some cases to do its job.
Up to 20 previous messages may be sent to Google’s servers to generate suggestions for Magic Compose, according to a Google support page. The company says the messages are not stored or used for training data, and images, voice messages and messages with attachments are not shared. However, image captions and voice transcriptions may be sent to Google.
Jedrzejowicz says the number of messages shared with Google is usually smaller, since only the last several messages are typically relevant for generating a response. That “20” number is meant to account for scenarios when the last several messages are photos and videos rather than text.
“Twenty messages back, it’s most likely the conversation has moved on to something different and so it’s just not that relevant,” he said.
Text messaging has existed for more than 30 years, and in that time it’s become an important part of how we communicate. By 2011, almost 20 years since the first SMS text message was sent in 1992, text messaging users in the US were sending or receiving an average of 41.5 messages per day, according to Pew Research data from that year.
Jed Brubaker, an associate professor of information science at the University of Colorado Boulder who studies digital identity and social media, says we’ve developed our own digital dialect in that time, known academically as “netspeak.” Whether you find yourself using the same emojis frequently or regularly punctuate your text with an exclamation point, we all have our signature voice when communicating over text.
“These are the text-based versions of our intonation and how we communicate with each other,” said Brubaker.
Google’s Magic Compose can suggest responses.
Would AI agents that can rewrite texts and suggest responses eventually strip that away? It’s a concern that’s crossed the mind of Maggie Mulqueen, a psychologist who has written about the impact of digital communication on personal relationships. If we’re relying on AI tools to polish the voice behind our texts, does that mean our messages lose authenticity? That all depends on how people are using these features.
“If people are going to use it to embellish in some way or compensate in some way for who they are, then what happens?” she said.
That’s part of the reason why Google is taking a cautious approach with Magic Compose. Although Jedrzejowicz sees usefulness in customizing the feature to incorporate those personal elements mentioned earlier, like a user’s favorite emojis, Google is considering feedback from users before making such changes.
“But there’s also a very real scenario where somebody could say, ‘You know what, that’s not necessarily what I want. I don’t necessarily want my messaging app to learn my habits and learn how I communicate,’” he said.
Google also debated automatically sending a message after a user taps an option in Magic Compose, similar to the way its Smart Reply feature works. But the team decided the right approach was to populate the text box instead so that the user could review and edit the text message before hitting the send button.
“It’s great that these AI features can boost your creativity and give you some extra kind of fun flair, but you are the person sending the text,” said Jedrzejowicz. “You are the person who’s going to be associated with that text.”
More than what we say over text, AI could impact when we text, which also says a lot about our personal relationships, says Joel Schneier, a faculty member of the University of Central Florida’s Texts and Technology program. While we might respond quickly to close friends and family members, there are times when we might take an extra couple of minutes to review a message before hitting the send button. But Schneier is curious about how AI-based tools that make it easier to craft messages could change that.
“I imagine that might seem a little odd to people because the message might come too quick,” he said.
Texting has undergone many transformations in recent years, mostly attributable to other advances in technology like the advent of smartphones and social media — both of which shaped the way we communicate. AI could just be the next step.
It’ll all come down to how useful these AI-based features end up becoming. Will they become a staple when it comes to the way we text, or just a fun trick?
“If people are really finding that they are saving a lot of time or are getting more creative because they’re able to leverage some of these existing prompts from the AI, then I could see this really taking off,” said Sydow.
In the meantime, I’ll be texting away as always, and occasionally looking for a little help in getting the message right.
Editors’ note: CNET is using an AI engine to help create some stories. For more, see this post.