Incident 17: Inappropriate Gmail Smart Reply Suggestions
Suggested citation format
CSET Taxonomy ClassificationsTaxonomy Details
Google's Gmail Smart Reply tool suggests replies, each of about three to six words long, to emails received on its platform. According to Google's management director Alex Gawley, the machine learning-based system is able to produce about 20,000 discrete responses to emails by combining suggestions from two machine learning programs. The first reads the email and comprehends the content and the second suggests responses to that content all taking place on the long short-term memory neural network. Gmail users in 2015 provided feedback that the reply "I love you" was occurring too often, leading Google to change their algorithm.
Google's Gmail Smart Reply tool was over-recommending the response "I love you" in situations where it was deemed innappropriate.
AI System Description
Google's Gmail Smart Reply tool, a machine-learning system is able to produce approximately 20,000 discrete responses by combining suggestions from two machine- learning programs: the first reads and comprehends the content and the second recommends responses.
Sector of Deployment
Information and communication
Relevant AI functions
machine-learning, natural language processing, long short-term memory neural network
content creation, smart reply, suggested reply, recommendation engine
Google, Gmail, Gmail Smart Reply
Ever wished your phone could automatically reply to your email messages?
Well, Google just unveiled technology that's at least moving in that direction. Using what's called "deep learning"—a form of artificial intelligence that's rapidly reinventing a wide range of online services—the company is beefing up its Inbox by Gmail app so that it can analyze the contents of an email and then suggest a few (very brief) responses. The idea is that you can rapidly respond to someone while on the go—without having to manually tap a fresh message into your smartphone keyboard.
"The network will tailor both the tone and content of the responses to the email you're reading," says Google product management director Alex Gawley. It gives you three of these responses, and you can then choose the one that best suits what you want to say.
Dubbed Smart Reply, the system learns to generate appropriate replies by analyzing scads of email conversations from across Google's Gmail service, the world's most popular internet-based email system. A deep learning service feeds information into what's called a neural network—a vast network of machines that approximates the web of neurons in the human brain—and this neural network analyzes the information in order to "learn" a particular task. By analyzing thousands of cat photos, for instance, a neural net can learn to identify a cat. By analyzing a database of spoken words, it can learn to recognize the commands you speak into your smartphone. In this case, the system learns to compose email replies by analyzing real-world email conversations.
Experts on deep learning, however, will tell you that such systems have their limitations. "With finite amounts of data, you can create a rudimentary understanding of the world," says Andrew Ng, chief scientist at Baidu, the Chinese Internet giant that also sits at the forefront of the deep learning movement, "but humans learn about the world in all sorts of ways [we can't yet duplicate]." Indeed, Gawley acknowledges that Google's Smart Reply system doesn't always get things right. But that's part of the reason the company provides three potential replies to each email—not just one. Plus, it lets you edit these replies and augment them with your own words.
The system uses what's called a "long short-term-memory," or LSTM, neural network. Essentially, this is a neural net that exhibits something akin to human memory. It can "remember" the beginning of an email as it's parsing the end—and that helps it, on some level, understand this natural language. In a research paper published earlier this year, a team of Google researchers showed how this technology could be used to build a "chatbot" that can carry on a decent conversation (in certain situations).
Actually, the Smart Reply system uses two neural networks. After the first one analyzes the email at hand—distilling what is being said—a second takes this information and works to generate the potential responses. This network builds these replies one word at a time, much as you would.
With Smart Reply, Google is rightly keeping the scope of the application as small as possible. The replies it generates are between three and six words long. But Gawley says that within this small scope, the system proves surprisingly nuanced. In some cases, for instance, it can tell when an email includes a joke and suggest the reply "Ha. Very funny." If someone asks "Do you have your vacation plans set yet? When you do, can you send them along?," the potential replies might be: "No plans yet," "I just sent them to you," and "I'm working on them."
Other common replies include "Thanks," "Sounds good," and "How about tomorrow?" But it's important to remember that the system isn't offering a canned catalog of replies. In effect, the AI really is "reading" your email and coming up with what it judges the most appropriate original response in the context of a specific message. According Gawley, the system can generate about 20,000 discrete responses.
Sometimes, Gawley says, the neural network generates multiple replies that aren't that different from one another—such as "How about tomorrow?" and "Wanna get together tomorrow?" and "I suggest we meet tomorrow." So, the company has built a separate AI system that can remove such duplication. At times, Smart Reply still steps outside the bounds of what you want. After testing the system, Google found that the reply "I love you" turned up far too often. But as with other neural networks, Google is constantly tuning the system after seeing how it performs.
The company will start sharing the system with the general public on Wednesday, and as time goes on, it will only get better. But let's hope it doesn't get too good. Help with rapid-fire replies is one thing. But there is something to be said for, you know, actually writing your own email.
Just a few months ago, Google really started showing off its newest secret weapon: a robot brain that learns. Using machine learning algorithms that actually get smarter as they practice, Google's been able to identify your loved ones in your photo albums, and figure out what you might want to search for based on what's on your phone screen. And soon, it could be answering your emails too.
The new feature is called "Smart Reply" and it's for Inbox, the fancy new Gmail alternative Google announced a while back. When you're using it, Smart Reply is simple; you just get a handful of prewritten options to respond to emails, and you can either select one and send it off, or ignore them and type one in yourself.
Behind the scenes however, this is a massive accomplishment. A computer is reading your email (slightly creepy but nothing new) and understanding its meaning. It's not as simple as just identifying a few keywords like "plans" and then spitting out a few auto-responses about "I'll send them." Google's machine learning algorithms actually learn from scratch by reading thousands of emails and thousands of responses, slowly getting better and better at predicting what kind of responses match up with what kind of emails.
In an in-depth post about how this tech actually works, Google shared some fun anecdotes about the initial prototypes. Some of the early versions of the tech were a little too lovey, for instance:
[A] bizarre feature of our early prototype was its propensity to respond with "I love you" to seemingly anything. As adorable as this sounds, it wasn't really what we were hoping for. Some analysis revealed that the system was doing exactly what we'd trained it to do, generate likely responses -- and it turns out that responses like "Thanks", "Sounds good", and "I love you" are super common -- so the system would lean on them as a safe bet if it was unsure.
The new feature will be rolling out to Inbox users later this week, and even if it's not perfect yet, it'll only get better as time goes on.
There's actually a lot going on behind the scenes to make Smart Reply work. Inbox uses machine learning to recognize emails that need responses and to generate the natural language responses on the fly. If you're interested in how Smart Reply works, including how researchers got machine learning to work on a data set that they never saw, you can read more about it on the Google Research Blog.
And much like how Inbox gets better when you report spam, the responses you choose (or don't choose!) help improve future suggestions. For example, when Smart Reply was tested at Google, a common suggestion in the workplace was "I love you." Thanks to Googler feedback, Smart Reply is now SFW :)
Smart Reply will be rolling out later this week on both Google Play and the App Store in English. If you've got a lot of emails on your plate, now's a great time to try Inbox and get through them faster than ever.
Not every email deserves a handcrafted response. Sometimes, all it takes is a sentence to answer the one burning question that someone has dropped in your inbox, which is why Google is using the power of machine learning to make email triage a little bit faster.
The company announced a new Smart Replies feature for its Inbox email client on Tuesday, which gives users up to three quick options to send back in reply to emails based on a machine learning analysis of the message’s content. People can use the short replies as either a way to quickly respond, or a way to start a longer message. Smart Reply could be a boon to mobile users who are on the go and don’t have time to tap out a whole response.
Because of the way the feature is built, it learns to suggest better responses over time. That’s important, because it should help cut down on bad suggestions that are inappropriate. (In early testing at Google, Smart Reply suggested that co-workers send “I love you” to one another.)
Smart Reply will start rolling out to Inbox users later this week. The new feature is in keeping with Inbox’s positioning as an email client that helps users sort and handle their messages intelligently.
The idea of machine learning-powered replies using the context gleaned from a message isn’t new in the communications space. Both Google and Apple have default keyboards in their mobile operating systems that suggest text based on the messages users are responding to.
Using machine learning to improve the way that people communicate over email is important for Google as it competes with other companies to be a user’s preferred provider. Microsoft uses similar techniques to categorize users’ email in Outlook, and continues to build out new capabilities in its products.
Replying to routine email requests just got easier with a new smart reply feature for Gmail users who get their email through Google Inbox. Starting this week, the app's Smart Reply feature will analyze incoming email messages and offer a selection of brief stock replies. All you have to do is pick one and hit Send.
Smart Reply will be a welcome feature for users who receive an overwhelming volume of emails every day. The feature acts like a virtual assistant that processes your incoming email messages, separating them into messages that can be handled with a short reply (such as “Hey, coffee later after the meeting?”) or long replies (such as “What can you say about my proposal at the meeting?”). It guesses what your reply might be and proposes three suggestions, sort of like the auto-suggest feature of your smart keyboard.
Using the quick-reply messages makes replying to simple requests and messages quick and convenient. Just tap one of the generated replies — no typing and very little thought required. For longer replies that require more thought, the suggestions can jump-start your reply. Just choose a reply from the list and base your response from that one.
Smart Reply does not come with smarts out of the box. It needs to be trained, just as you teach Gmail better spam detection by actually reporting spam messages. Smart Reply learns your message reply patterns based on the suggestions you pick or don’t pick from its list of replies.
While Google was testing Smart Reply, “I love you” appeared as a frequently suggested reply, but of course that could be an awkward phrase to send in the workplace. Based on tester feedback, that phrase appeared less often on the list of suggestions. But the more you use Smart Reply, the more it will get to know your replies and the better its suggestions will become.
Besides Google Inbox, you can make handling your email simpler with a number of apps for managing your mobile mail. If Gmail is your favorite, be sure to read our five tips for getting more out of Gmail.
[Image credit: Google, woman typing via Shutterstock]
Google will soon scan the content of your emails and serve up what it thinks is the perfect reply.
The new feature, called Smart Reply, identifies Gmail emails that require a response and presents three options for replies. Smart Reply will roll out this week to people using Google's Inbox for Gmail app.
For example, if you get the email, "Do you have any documentation for how to use the new software?" the Inbox for Gmail app will show these three responses at the bottom of your screen:
"I don't, sorry."
"I will have to look for it."
"I'll send it to you."
In another example, you might get an email that says, "Do you have your vacation plans set yet? When you do, can you send them along?"
Google (GOOGL) will give you three replies:
"No plans yet."
"I just sent them to you."
"I'm working on them."
If you decide to choose a Google-provided response, you can add to it or edit the reply after tapping on one of the three options.
Smartphone software makers have been trying to figure out how to save keystrokes for years. Autocorrect, swipe gestures and even emojis are helpful sometimes, but they still require a bunch of taps or swipes to type what you want to say. Quick responses have existed for years, but they're canned messages and don't change based on the content of the text or email that you received.
"For those emails that only need a quick response, it can take care of the thinking and save precious time spent typing," said Bálint Miklós, software engineer for the Gmail team, in a blog post. "And for those emails that require a bit more thought, it gives you a jump start so you can respond right away."
Miklós said that getting computers to understand how to do that was a difficult task. It takes a great deal of heavy-duty machine learning to pull it off, and it might take some time to perfect it. The responses that Gmail customers choose (or don't choose) will help the Inbox app improve its suggestions in the future, he noted.
"For example, when Smart Reply was tested at Google, a common suggestion in the workplace was 'I love you.' Miklós said. "Thanks to Googler feedback, Smart Reply is now SFW :)"
When Google launched Inbox, its most recent email app from the Gmail team, it touted the app's ability to act almost like an assistant. Now, a year later, Google is making the app more like an aide than ever.
The company will soon be rolling out a new feature called "Smart Reply," which uses artificial intelligence to suggest replies for your messages. The feature will be coming to Inbox's iOS and Android apps later this week, according to the company.
See also: 5 Google Inbox features that will make email suck less
The Smart Reply feature works a bit like predictive text in keyboard apps, except that it predicts phrases you're most likely to use when replying to messages. The feature will show up to three short sentences based on the message and how people have responded to similar emails in the past.
"For those emails that only need a quick response, it can take care of the thinking and save precious time spent typing. And for those emails that require a bit more thought, it gives you a jump start so you can respond right away," software engineer Bálint Miklós writes.
Behind the scenes, the feature took quite a bit of engineering to get it working, Miklós, explains. It uses machine learning techniques to surface relevant responses in realtime. As with other tools that use natural language technology, Smart Replies should improve over time as people use the feature more. He notes that when they first began testing it, a commonly suggested reply was "I love you," which is probably not appropriate for most work-related correspondence. (He promises that reply suggestions have been tweaked to be more SFW.)
If you want a closer look at the technology behind it, Google also published a lengthy post on its research blog detailing the process.
It's still early so it's difficult to say how big of an impact the feature will have on Inbox and how people respond to email, but it offers an enticing look into how Google is using artificial intelligence to make some of the most mundane tasks we do every day — like answering emails — a little easier.
On April 1, 2009, Google unveiled Gmail Autopilot, a plug-in that promised to read and generate contextually relevant replies to the messages piling up in users’ inboxes. “As more and more everyday communication takes place over email, lots of people have complained about how hard it is to read and respond to every message,” the product page explained. “This is because they actually read and respond to all their messages.” For those who hadn’t registered the date, the terms-and-conditions page spelled out the joke: “No, we don’t plan to scan every one of your incoming messages and automatically send the perfect reply.”
On November 5, 2015, Google unveiled Smart Reply, a plug-in that reads and suggests responses to e-mails. This time the innovation actually exists, as part of the company’s Inbox app for Android and iOS. If Smart Reply thinks that it understands a message that requires an answer, it will suggest three options, alongside a cheerful invitation to “start composing your reply with one tap.” When I wrote to Matt Jones, the design director at Google Research, to say that it had been great to see him the other night and to ask whether he could put me in touch with the team behind Smart Reply, he sent back a screenshot of the options that the app had given him: “It was great seeing you too!” “It was fun!” “Will do!” He connected me with Alex Gawley, the product manager for Gmail, who said that I shouldn’t be offended. (“I mean, I’m sure it was great to see you.”)
The April Fool’s joke of 2009 began to be taken seriously in early 2015, Gawley told me, thanks to two developments. The Google Research team, which had recently acquired DeepMind, the company behind an arcade-game-winning form of artificial intelligence, was making rapid progress in language-related areas of machine learning, including translation and speech analysis. At the same time, Americans were reading more and more of their e-mails on mobile devices—fifty-three per cent of them, in fact, up from eight per cent in 2011. “It’s a little screen and a little keyboard, which means e-mail is easy to read and really hard to reply to,” Gawley said. He added that the combination of fat thumbs and autocorrect is “a real pain point for our users.”
Smart Reply uses what is known as an artificial neural network—an intimidating term for a particular kind of mathematical model—to tease out the patterns and probabilities that underlie e-mail communications. For privacy reasons, humans are not allowed to read Google’s vast corpus of e-mail messages. Machines, however, are, and by drawing on that data they can gradually sort sentences into “thought vectors,” or coördinates in linguistic space. In other words, by plotting similarities in context, word frequency, and sentence structure, the neural network can teach itself to recognize and group together the endless variety of ways that humans have developed to say much the same thing: “How does this afternoon look for a call?” “Can we talk later today?” Or “Does this P.M. work for a quick chat?” By trawling through the data again, the machine can then find and suggest the most typical responses to this particular thought vector: “Sure, what time were you thinking?” “Sure, anytime.” Or “Sure, what’s up?”
Earlier this year, Google researchers used the network to develop an intelligent chatbot with which they could discuss the purpose of life. (“To serve the greater good,” according to the machine.) Still, when the company’s engineers applied their neural net to the problem of e-mail, it did not work perfectly right away. “Most machine-learning work is actually about tuning,” Gawley said. The A.I. has a tendency to suggest replies that say the same thing in slightly different ways, which is less useful than offering users responses that represent a range of different likely replies. (For instance, “No, sorry, I’m busy” would have been a more useful alternative than “Sure, anytime” in the example above.) The team has corrected for this, to some extent, by adding a parameter that encourages the machine to choose disparate responses—ones that have sufficient distance between them when plotted as vectors in semantic space.
The early iterations of Smart Reply were overly affectionate. “I love you” was the machine’s most common suggested response. This was a touch awkward: because the model has no knowledge of the relationship between an e-mail’s sender and its receiver, it provides the same suggested responses whether you are corresponding with your boss or a long-lost sibling. “The team was really puzzled about this,” Gawley said. “It turns out that our internal testers are very affectionate and that ‘I love you’ is a very common thing for people at Google to say.” When the engineers inspected their model, they discovered that whenever an e-mail did not give a particularly strong signal as to the appropriate response, the machine hedged its bets with a declaration of affection. This fact may yet become the subj
A system Google designed for replying to emails proved a little friendlier than anyone expected.
There are quite a few humans out there who are absolutely convinced robots will kill us the moment they become self-aware. But what if the exact opposite was true, and they just wanted to be friends? At least, that's the scenario Google encountered when it designed a program that assists uers with automatic email replies. While the system worked a lot better than expected, it had one strange quirk: It kept trying to reply with "I love you" when that wasn't an appropriate response.
The information arrives in relation to Smart Reply, a system Google unveiled this month. One of Google's research scientists provided a technical breakdown for how it works, but in short, Smart Reply reads your email, determines if it only needs a quick reply, and generates some suitable responses you can pick from. Let's suppose a friend sends you an email asking if you'd like to grab lunch on Wednesday. Smart Reply would pick out the key words (specifically "lunch" and "Wednesday") and suggests "I'll be there!" or "Sorry, can't make it" as quick responses.
The functionality ended up working very well, and Google implemented the feature into Inbox for Android and iOS. But Smart Reply's prototype was odd - it got a little too friendly with its replies. The program constantly recommended "I love you" as a reply for pretty much anything, even when it wasn't appropriate. According to Google's Greg Corrado, "I love you" is just such a common human phrase that Smart Reply kept leaning on it when it wasn't sure of a better response. Which is kind of adorable, but not always helpful when replying to an email from your boss.
Corrado explains that Smart Reply now normalizes the likelihood of "I love you" appearing based on how often you use it with specific individuals, so it shouldn't come up as often. Here's hoping we haven't cut the emotional center out of an emergent AI in the process, because if movies have taught us anything, that might come back to haunt us.
If you use Google products for email, you may have noticed those suggested replies at the bottom of your mobile inbox. These give users options to choose short and snappy responses to emails, such as, “Will do,” “I am working on that now,” and “Thanks, you too!” This feature is called Smart Reply, and it uses a type of AI called an artificial neural network. It basically “learns” how individuals communicate with each other and then mirrors those interactions in its suggested replies. The feature has become a hit among those with busy lives who need to respond to their emails but don’t always have the time to do so.
Google originally rolled out this feature in 2015 with its app Inbox, but the company added a newer version of this same technology to the Gmail app in the spring of 2017. It immediately garnered attention from techies and average users alike. It’s so fast! It’s so convenient! It’s so accurate! It didn’t seem like anyone had much criticism to lodge about this feature, unlike some of Google’s other ideas, such as Project Fi, the company’s attempt at a cellular service.
And while some have concerns about privacy, there isn’t an actual person reading your messages. Google already scans the contents of your emails to provide you with tailored advertisements anyway. Gmail has never been truly private, but Smart Reply itself isn’t really to blame for that shortcoming.
So, what’s the problem? It’s not really that Smart Reply presents new challenges we must overcome in the digital age, but it does seem to be a symptom of a much wider-ranging issue—one that’s going to be a lot harder to solve than by simply turning off a function in your settings.
The celebration of Smart Reply and other suggestion-response features like it reveal the ever-quickening pace of our lives and what we choose to prioritize with the time we have available. Communication has, largely, become more of a chore than a conscious connection with another person, making our relationships shallower and less honest but leading us to believe we’re actually connecting more deeply with others than ever before. For example, Facebook users are more likely to receive an onslaught of well wishes on their birthdays than those who don’t use the site. But how many of your friends would know it was your birthday if not for the notification they received on their mobile device? How many of them know your address and would have sent you a card? Probably not nearly as many. Therefore, we’re tricked into thinking that we are building strong human connections when we’re just fulfilling what have become basic social obligations.
In our current cultural climate, it has become too time-consuming to write a quick “thank you” email to a client; instead, we opt for a one-click automated response. And while, yes, the responses are based off of how you as an individual use language, the author of the email isn’t writing these responses—a computer is.
Some may roll their eyes at this criticism. So what? It’s the fact that we’re responding that counts. Why does it matter if we type the message ourselves? But there are some indications that Google users do realize that there is a difference between choosing a suggested response and typing one out themselves.
Image: Google: Photo: Konstantin Sergeyev
The most recognizable feature of Gmail’s newly rolled-out redesign is the so-called smart reply, wherein bots offer three one-click responses to each mail message. Say your email contains the words “you free for lunch?” The autoreplies Gmail presents will be something like “Sure!” and “Yes!” and “Looking forward to it!” The idea, especially on a small, one-hand phone screen, is that you can tap and send using one thumb, without typing. It’s not clear just how many of these prewritten options there are, or how sophisticated the machine learning behind them is. The AI is not yet sharp enough to offer genuinely useful responses like “Please, for the love of Christ, stop sending me these offers to buy those sandals whose ad I clicked on last month” or emotionally honest ones like “Hey, it would be wonderful if someone in our group cancels our drinks tonight because I would rather stay home and order dan dan noodles while watching Succession.” Until then, we’re stuck with the few dozen simple responses that appear regularly. Some are better than others. Shall we rank? Ok, sounds good!
- “No worries, thanks for the heads up!”
A disaster. First of all, it’s a seven-word phrase that contains two grammatical errors (a comma splice and the missing hyphen in “heads-up”). And then the phrase “no worries!” injects a note of concern where there was none. Should I in fact be worried? Should I have been worried? The only possible use of this response is when a frantic employee sends an email to the boss about a potential problem, and the boss wants to knock down his or her anxiety a little. I’m not projecting here.
- “You are so welcome!”
That “so” in the middle is what dooms this one. It’s unusably chirpy mean, really underminer stuff. You use this when someone thanks you for getting the T-shirts in time for rush week. Comes with an unspoken “[asshole!]” at the end. In the South, the equivalent would be “Bless your heart.”
- “Thank you, you too!”
What does this phrase even mean? Stare at it more, and you’ll understand it less with every passing minute. Plus it’s another goddamn comma-splice run-on sentence. Get these bots a copy of Words Into Type already.
- “It was great seeing you too!”
The hell with this. Either you didn’t actually think it was great (in which case, sure, send this, why not), or you did, in fact, think it was great (in which case you should probably write something a little more heartfelt and less anodyne).
- “Love it!”
No editor will ever use this.
- “Have a great weekend too!”
Isn’t “Have a great weekend” perfectly and explicitly the way you end an email exchange? It all but defines the you-don’t-have-to-respond-to-this genre of messages. Why the hell would you write back?
- “Cool, thanks!”
“Thanks!” is meaningful. “Cool, thanks!” means … what, exactly? I like what you said, and thank you for saying it? Literally a phrase I have never spoken or typed. But Google will type it for me!
- & 12. “Yes, he is,” and “No, he’s not.”
Two observations here: The internal commas make the vibe curiously old-fashioned, stilted, almost Victorian. And the flat period, replacing the chirpy bang, inadvertently comes off as relatively dead-eyed, almost angry. “Yes, he is. The man you just made fun of is indeed the same one whose blood I wear in a vial around my neck. Care to continue?”
- “It was my pleasure!”
Amid all the “Cool!” and “Very cool!” replies appears this oddly formal one. Useful for the 49-year-old mid-level manager who’s said yes to his or her third “informational interview” in a week. I’m not projecting here.
- “I’ll ask.”
This one has potential. No exclamation point of fake enthusiasm here; what it means is “if it were up to me I’d tell you to go to hell, but since it is not up to me, I’m happy to deflect responsibility for your coming rejection onto someone else. Either way, don’t get your hopes up.”
- “I’ll look into it.”
See “I’ll ask,” above. This is the version for graduate-level players only. (The Ph.D. version reads “I’ll look into it …” with the hanging ellipsis.)
- “Looking forward to it!”
Can’t quite decide about this one because I am so rarely looking forward to anything that comes by email. Maybe someone else is. I will reserve judgment for now.
- & 6. “So true!” and “I agree!”
Not as useful as they might have been a few years ago, when heartwarming Upworthy videos were more popular. Maybe they’ll come back into relevance around the election, when you start getting more of those lefty chain emails from your great-uncle in Vermont, the ones that end with the “Coexist” logo.
- “Got it!”
Could be something to send after you get a sensitive document and you want to acknowledge it. “Got it — thanks,” sans exclamation point, with em-dash, would be vastly better. Either way, it’s the text equivalent of the return receipt. Allowable under some circumstances.
- “I don’t think so.”
Every editor can use this. Also, if they sold it on a T
“Thanks for letting me know.”
Have you noticed an uptick in this phrase appearing in emails, on social media, and IRL? Do you find yourself saying it even though maybe a year ago you would never use the phrase? Is there a new wave of semi-formality sweeping our communications and infecting our brains? What’s going on here?
The answer isn’t some insidious conspiracy by the Illuminati to make the world more polite. “Thanks for letting me know” just happens to be one of the more common answers offered by Google Smart Reply, and as a result, it has become embedded in our daily lives, like so many other products the company offers.
In November 2015, Google unveiled Smart Reply, a plug-in that reads and suggests responses to emails. The plug-in does its best to take in the content of the message you’ve received and suggest three potential appropriate responses.
The benefits in terms of efficiency are obvious. Many of us wake up to cluttered inboxes and spend the rest of our days grasping toward the zen of “inbox zero.” Like all types of nirvana, inbox zero is an elusive state. Try as we might, we never quite get that last piece of mail off of our digital desk before another deluge of correspondence crashes onto our servers.
It would be considered callous, even cruel, to simply respond, “Got it. Thanks.” to the bulk of your emails. Smart Reply, with its personally tailored yet concise responses, offers a way to maintain a modicum of human decency while preserving your e-sanity.
A perfect solution to a uniquely modern problem? Not quite.
Almost three years into the Smart Reply experience, we are starting to see the cultural cost of automatic responses. With our personally tailored series of pre-fabricated lines drilled into our brains, we are all starting to sound like brands.
The new gmail makes for great, life-affirming conversations.
Ok, thanks for letting me know.
You’re the best!
My pleasure! — Lynley Stace (@lynleystace) June 26, 2018
This makes sense in terms of the goals of Smart Reply. For most people, the majority of emails are work-related, and while friends and family will forgive a touch of formality, being too loose at work has ended a lot of careers. The most important thing Smart Reply can do is not get you fired.
This bears out in the history of Smart Reply’s development. When the technology was first introduced, the New Yorker’s Nicola Twilley wrote a piece on the rollout of the plug-in. Early testing showed that Smart Reply’s natural tendency was to be “overly affectionate.” “When the engineers inspected their model,” Twilley relays, “they discovered that whenever an e-mail did not give a particularly strong signal as to the appropriate response, the machine hedged its bets with a declaration of affection.” This is a very human response: How many men after too many drinks have come up with only “I love you, man!” to explain their feelings to a buddy? Google saw this as a “bug, not a feature,” and added an affection filter to the technology. Your email would never again say, “I love you.”
Google knows that our No. 1 email priority is to avoid embarrassment. After all, “Undo Send” remains one of Gmail’s most popular features and as soon as a big job offer lands on your desk, you would do well to purge any offensive tweets from your Twitter history.
It is in Google’s best interest to create a product that minimizes any outpouring of emotion that might be inappropriate. We can assume that the algorithm that produces these auto-responses isn’t just controlling for affection, but also humor, sadness, anger, wistfulness, and any other feelings that would prevent you from being a gold-star employee.
Smart Reply tries to strike a balance between office-appropriate responses and maintaining some level of personality. Andrew Chamings writing at the Bold Italic noted that while he was offered effusive responses like “Sure thing!” friends of his were given more muted suggestions like “I’ll be there.” and “I’ll try.”
The end result is a set of responses that are in “your voice,” but never transgress the acceptable limits of that voice in a professional setting. After several years of Smart Reply, we all sound a little bit like a corporatized version of ourselves.
We are also in an era where every big brand has a readily identifiable voice on social media. Everyone on Twitter knows what Denny’s and Wendy’s “sound like.” The now-defunct account @brandsayingbae chronicled the early years of company Twitters trying to sound like cool teenagers, and despite constant mockery from the snarkier corners of the peanut gallery, the brands are talking to us, releasing music, and even dissing customers.
As human beings have attempted to sanitize their public presence for fear of being James-Gunned out of their livelihood, and brands have sought ever more “authenticity,” the distance between the way your friend Wendy talks and the way the corporation Wendy’s sounds grows smaller by the day.
Three years into
When Jess Klein emailed her mother to make plans, the response startled her.
“Cool, see you there.”
Ms. Klein, a 38-year-old from Brooklyn, was suspicious. “Why is my mom talking like a Valley Girl?” she wondered.
Around 10 percent of Gmail responses are now generated with the Smart Reply feature, according to the Wall Street Journal, and it has been available on Gmail apps since last year. Ajit Varma, Google's director of product management, told the publication that the phrases offered in the email are drawn from a library that a Google bot compiled based on billions of analyzed Gmail messages. Early versions of Smart Reply included responses like "Sent from my iPhone" and "I love you" since those are used frequently in Gmail replies. Those were removed as reply options, Varma said. Over time users should find that their suggested replies become more personalized.
You can already turn off Smart Reply in your Gmail app. Just head to Settings and you'll find the Smart Reply feature under the General section. From there, you can toggle it on or off.
After users complained that Gmail’s Smart Reply wasn’t so smart, Google announced that it will soon give desktop users the ability to disable the A.I.-based feature. Although the time-saving feature relies on artificial intelligence to come up with short responses based on the context of the email message, users complained that early iterations of Smart Reply frequently offered suggestions that were either not very useful or not appropriate. Early iterations of Smart Reply often suggested “I love you” and “Sent from my iPhone” as possible responses to email messages.
Despite complaints, Smart Reply appears to be catching on. The Wall Street Journal reported that Smart Reply was used on more than 10 percent of all sent Gmail messages, which is a huge number considering that Gmail counts more than 1 billion users. The Journal also noted that in the coming weeks, Google will provide desktop Gmail users with the option of opting out of Smart Reply altogether. Google already offers users of the Gmail app the ability to turn it off.
Although Smart Reply only launched earlier this year alongside Google’s big Gmail redesign, the feature was present for about a year before that for Gmail users who used the Inbox service. Google claimed that 12 percent of Inbox users had used the smart feature, prompting it to make it available to more users. When Google moved Smart Reply out of Inbox and into Gmail, it said that it changed the algorithms to make the suggestions more natural.
“Based on our examination of the use of Smart Reply in Inbox and our ideas about how humans learn and use language, we have created a new version of Smart Reply for Gmail,” Google wrote on its A.I. research blog at the time. “This version increases the percentage of usable suggestions and is more algorithmically efficient.”
Google claims that by using a bot to scan your messages, it can apply machine learning to improve Smart Reply and make better, more personalized recommendations in the future. And even though suggestions may improve, for group emails, Smart Reply may not be able to discern whether you’ll want to reply to one person or everyone in the email thread.
For users who love the Smart Reply suggestions, Google is looking at ways to bring this capability to other messaging apps in the future on its Android platform.
The Wall Street Journal is out today with a quick story on the rollout and reception of Gmail’s new Smart Reply feature, and it comes with a few interesting bits of information and anecdotes from random users and Google alike — one of which is particularly humorous…
Deal: Get Pixelbook at 25% off: $750!
The piece notes the variety of responses that people have had to the feature. One notes the feeling of dehumanizing the experience, and an actor noted that the replies felt a bit too casual for his liking:
However, when she realizes she’s on the receiving end of a canned response, she can’t help but feel a little sad. “I feel they didn’t take the time to write a real response,” she said. “Like, we’re at that level now?”
Shane O’Regan, an Irish actor currently in an off-Broadway show, said his suggested replies often include “Sweet!” and “Awesome!”—answers he wouldn’t want to send his agent. “I’m not a surfer dude.”
Perhaps one of the funniest anecdotes from the piece is this one, about the infamous “Sent from my iPhone” phrase that appears at the bottom of emails sent on iPhones by default.
During testing, Mr. Varma said, engineers working on the prototype noticed one gaffe: The algorithm was identifying the phrase “Sent from my iPhone” as a popular response to emails. They fixed the issue before the phrase became a suggested reply, he said.
Another humorous tidbit: Apparently, Smart Reply was fond of recommending “I love you,” which is obviously not an ideal response for most cases in a business setting.
Another one: Google said an early prototype of the feature had “a propensity to respond with ‘I love you’ to seemingly anything,” forcing it to tweak the algorithm. “You don’t want to respond that to your boss,” Mr. Varma said.
Obviously, Google is constantly working and tweaking the algorithm to best fit the context of the emails it’s hoping to help you reply to, but it seems as if Google caught the most glaring of issues early.
(Image: The Wall Street Journal)
How do you feel about Gmail’s new Smart Reply tool? We’re hearing mixed reports. On one hand it’s super handy to be able to tap/click that “Thank you so much!” button when you’re in a rush.
However, users are reporting the suggestions aren’t always on the money. People are seeing “Sent from my iPhone” as a suggested reply, while others have spied “I love you” in the ‘smart’ replies field. Man, that could be an awkward way for your office crush to find out about it, huh?
Also, and perhaps more pertinently, the fact Google knows what a suitable response could include is creepy as hell. It’s an effective punch in the face from Google that says “we’re reading your emails”. We all know it happens, but come on…
Related: Best smartphone 2018
The inaccuracies and the stalker-ish elements have enough folks wanting out.
On mobile this is possible by accessing Settings and switching off the “Smart Reply” toggle switch. However, on desktop, where the Smart Reply feature is arguably less necessary, the option is notably absent.
Thankfully, for creeped-out Gmail users, the switch-off switch will be here in the coming weeks (per Wall Street Journal). There’s no specific date for the rollout, but we’ll keep you posted.
The introduction of Smart Reply came along with Google’s much larger redesign of Google, which was rolled out to its user base of 1.4 billion earlier this summer.
That also included the addition of Smart Compose tool, which aims to predict what you’re going to write. If you agree with a suggestion from Google, hitting the tab button auto-fills the text.
This is another example of Google’s machine learning tools in action and of the user base happy to put convenience over the potential privacy costs of enabling Google to read our personal correspondences.
Are you happy with Google’s Smart Reply feature? Let us know @TrustedReviews on Twitter. I love you.
Google recently made its handy 'Smart Reply' feature available to Gmail's 1 billion-plus users.
But early prototypes of the tool didn't work so seamlessly, according to the Wall Street Journal.
In one case, the algorithms powering Gmail's 'Smart Reply' feature would suggest users send 'I love you' as a response to almost every email.
Scroll down for video
Google recently made its handy 'Smart Reply' feature available to Gmail's users. But early prototypes of the tool didn't work so seamlessly, with one goof suggesting 'I love you'
HOW TO SWITCH GMAIL'S SMART REPLY OFF To disable the 'Smart Reply' feature on a mobile device, simply go to 'Settings' and uncheck the box next to 'Smart Reply.' But for now, there's no way to disable the feature on the web. However, users can return to the 'classic' version of Gmail on the web by selecting that option under 'Settings.'
Smart Reply rolled out to Gmail users as part of a broader redesign of the email client.
Soon, the feature will become a default setting for all email accounts - an update that has both excited and annoyed Gmail users.
It offers short responses to emails, like 'It was great seeing you too,' 'I'll look into it' or 'Sounds good, thanks for the heads up!'
Others are even shorter, like 'Nope, that's fine' or 'Yes, I'll be home.'
Phrase suggestions are formulated using machine learning, which scans an email for different types of words, sentences and predicts possible responses.
Approximately 10 percent of responses are generated using Smart Reply, Google told the Journal.
When Smart Reply was in the early stages of development, it faced its fair share of goofs, according to the firm.
The feature's algorithms noticed that 'Sent from my iPhone' was a common phrase in emails and began suggesting it as a reply.
It didn't take very long before Google's engineers noticed this and rectified the issue.
In another case, the prototype had a 'propensity to respond with "I love you" to seemingly anything,' Ajit Varma, director of product management at Google, told the Journal.
'You don't want to respond that to your boss,' he added.
Soon, Smart Reply feature will become a default setting for all email accounts - an update that has both excited and annoyed Gmail users. Pictured is an example of how the feature works
It offers short responses to emails, like 'It was great seeing you too,' 'I'll look into it' or 'Sounds good, thanks for the heads up!' Others are shorter, like 'Nope, that's fine' or 'Yes, I'll be home'
While Smart Reply sometimes struggles to discern context when it scans an email, Google says the software does learn over time.
Gmail alters the tone and suggested responses the more you use the feature.
For example, if a user prefers to use a period at the end of each phrase instead of an exclamation point, Gmail will alter that user's canned responses accordingly.
The firm also explained that while it does read emails to generate responses, that it keeps 'privacy and security paramount.'
It doesn't look at a user's activity on other apps and sites, even Google-owned ones like Google Calendar, to generate responses.
Smart Compose, a feature like Smart Reply that predicts sentences as you type, will also be available in Gmail later this month.
Users can opt out of the Smart Reply feature, even if they're using Gmail on a desktop computer.
It's unclear if they can opt out of Smart Compose as well.
A few days ago, I received a short, effective email in my inbox: "Sounds good!"
I had to pause. Although that was the response I wanted — I was arranging a meeting — I wondered: Did he really send that, or did he simply hit Google's automated response suggestions at the bottom?
In recent weeks, Gmail's "Smart Reply" feature has been released to an ever-rising amount of the web email service's over 1.4 billion users, meaning that soon a large percentage of the world's population will get access to Google's cheery, direct suggestions when it becomes a default feature next month.
The actual language of the suggestions is created with machine learning, a family of technologies that is sometimes called artificial intelligence in the non-technical world.
Google's software crawls billions of emails on Gmail, and then uses software to suggest phrases drawn from that database that you might want to use to confirm a coffee at 4 PM at Blue Bottle. 10% of responses are generated through the Smart Reply feature, according to The Wall Street Journal.
But it hasn't always worked well enough — a director of product management at Google, Ajit Varma, shared some funny early bugs with the Journal.
From the story:
Google said an early prototype of the feature had "a propensity to respond with 'I love you' to seemingly anything," forcing it to tweak the algorithm. "You don't want to respond that to your boss," Mr. Varma said.
Imagine trying to organize a conference call with a client and accidentally sending "I love you."
Another funny suggestion the software kept repeating is a nod to Google's neighbors in Silicon Valley, Apple. Apparently, the software thought "Sent from my iPhone" was a good way to respond to emails.
That's the default signature on the default email app on iPhones, so it shows up all the time in Google's data. And Google's machine learning software, as smart as it was, isn't actually a human, so it didn't realize that was a clever bit of branding, and not any kind of useful information.
Same with "I love you" — while Smart Reply is impressive, it doesn't understand social context yet.
But the software does learn: Google says that as you use it more, the style and tone will become more personalized to you.
From: Corinne Purtill (Quartz)
To: Corinne Purtill (Gmail)
Mon, Oct 1, 2018 at 11:08 AM
At its annual developers conference earlier this year Google introduced “smart compose,” a new Gmail feature that helps users complete their sentences. Smart compose scans the content of users’ messages and suggests phrases and words, based on things that Google knows about English (i.e., sentences that begin with H often turn out to be “How are you?”) and things Google knows about you. It builds on Gmail’s “smart reply” feature, in which quick responses like “Thanks!” or “See you there” can be sent with a touch of a button.
To understand how these features influence communication, I’m writing in two places: here on my non-smart work email, where I must create sentences and replies from scratch as if this was 2015, and my personal Gmail account, which has smart compose turned on and ready to go.
Are you willing to play along, Gmail?
From: Corinne Purtill (Gmail)
To: Corinne Purtill (Quartz)
Mon, Oct 1, 2018 at 11:10
Sounds good to me!
From: Corinne Purtill (Quartz)
To: Corinne Purtill (Gmail)
Mon, Oct 1, 2018 at 11:16 AM
Smart reply, which this month will become a default feature for all Gmail users, has proved divisive since its April soft launch. Some users say that it spares them time, superfluous taps, and the consequences of their worst impulses (“Got it, thanks” is almost always a safer option than a clever-yet-snide reply one might be tempted to compose). Others say that the suggested replies are either inappropriate—early version suggested “I love you” with unnerving frequency—or, more often, fail to capture the sender’s tone.
For example, you, Personal Gmail, also suggested “Like it!” and “Love it!” as potential replies to my previous message, both of which felt a little too eager.
About one-quarter of the world’s emails are now opened in Gmail. If Google itself is helping dictate the content of those emails, will they still read like the people who wrote them? Or will we end up emailing each other in “Gmail-ese,” a generic English dialect that conveys the substance of messages quickly at the expense of individual style? (For the time being, smart features are only available in English.)
From: Corinne Purtill (Gmail)
To: Corinne Purtill (Quartz)
Mon, Oct 1, 2018 at 11:57 AM
I’m going to ignore the three possible replies Gmail has suggested here—again: “I like it!” “Love it!” and “Sounds good to me!”—even though the latter is, upon reflection, something I actually say quite often when ready to exit a conversation. Instead, unless it completely derails the substance of the discussion, I’m going to accept any suggestion Gmail tosses me as I type this. (I think that final “this” is superfluous; Gmail apparently does not.)
I was discussing this earlier today with Quartz colleague Cassie Werber in London, who pointed out that the feature has the potential to erode the diversity of the English language. If we’re all being encouraged to “grab” lunch, for example, when typing “Let’s” to a colleague around noon, will we receive fewer invitations to “do,” “have,” or “get together for” lunch—all subtle variations that convey minute but possibly significant changes in tone and style?
As is often the case with new technologies, smart email may also come with built-in (if inadvertent) biases. Smart reply was developed with the help of a bot that analyzed the contents of billions of Gmail messages, a Google product manager explained to the Wall Street Journal. Gmail may have a minority stake in the overall email market, but it’s the overwhelming favorite among younger users, with 61% of emailers aged 19 to 34 and 74% of those aged 14 to 18 preferring Gmail in 2016, according to one survey. In early versions, casual responses like “Sweet!” and “Awesome!” were suggested frequently, which many users found off-putting in professional exchanges.
Speaking of which: The few words or phrases Gmail has suggested as I type have all been sensible. The thing that comes up most frequently in the ghostly gray type of the suggestion box is “?”—a gentle nudge, perhaps, to just hurry up and end a sentence that’s clearly going to be a question.
So: Are we doomed to inboxes full of Gmail-speak?
From: Corinne Purtill (Quartz)
To: Corinne Purtill (Gmail)
Mon, Oct 1, 2018 at 12:31 PM
I don’t think so. AI has been lurking around our digital communications for a while now, helpfully correcting typos in some cases and unhelpfully garbling names and meanings in others. Identifying situations when it’s in our interest to overrule the machine isn’t particularly onerous. iOS has long offered a set of prefabricated text messages to send when declining a call. You only have a split second to decide if “Sorry, I can’t talk right now”—with a period at the end!—will make you sound like a cold-hearted monster to that particular caller, and really, that’s all you need.
Smart email is good at the quick, perfunctory exchanges that fill up much
Google will no longer continue support for Reply, an experimental app that offers smart reply responses to various messaging apps such as Slack, Hangouts, and Messenger. The app launched earlier this year as part of Google’s Area 120 division that incubates and tests wacky apps.
“As you know, Reply was an experiment, and that experiment has now ended,” Google wrote in an email to beta testers. “While it might still work for the next few months, you may encounter bugs, or see that the suggestions aren’t as good.” The company says it will incorporate what it’s learned from the app in other Google products, which likely includes Gmail’s Smart Compose and Smart Reply. (Smart Reply is also available on Android Messages.)
When I tested the app earlier this year, I found that Reply’s automated responses were often bland, leaned toward affirmative answers, or just simply parroted the noun from the message that was received. It also tended to offer “I love you” as a default third response regardless of context.
Reply was also supposed to roll out a Maps and Calendar integration that could offer appointment suggestions or commute estimates based on a user’s current location and traffic conditions, but neither of these suggestions was available in my short time using Reply. If Google plans to still roll out these features, however, expect to see them in Gmail and Android Messages first.
Did our AI mess up? Flag the unrelated incidents