It’s finally here—the email program that not only receives your email but also automatically generates responses and not simply with the standard out-of-office reply. Smart Reply, at least in theory, is smart enough to generate responses to the emails we don’t have time to respond to ourselves. Sound good? Perhaps, but Google’s Smart Reply also raises serious questions about best practices in business correspondence.
Machine Intelligence versus Human Intelligence
Researchers at Google, describe Smart Reply as “a deep neural network that writes email.” When they went into development, the goal was simple. What if they could develop an email program capable of automatically determining if and when an email is something that can be handled with a short automatically generated reply? The company’s researchers were up to the challenge and turned out what they believe is a viable way to help busy people handle the massive amount of email they receive on a daily basis. So how does Smart Reply work?
On the company’s blog, Google researchers explain: “A naive attempt to build a response generation system might depend on hand-crafted rules for common reply scenarios. But in practice, any engineer’s ability to invent ‘rules’ would be quickly outstripped by the tremendous diversity with which real people communicate. A machine-learned system, by contrast, implicitly captures diverse situations, writing styles, and tones. These systems generalize better, and handle completely new inputs more gracefully than brittle, rule-based systems ever could.” What does this mean? In short, it means that rather than spit out one-size-fits-all responses to specific messages, the program is smart enough to build replies—one word at a time. But does it work?
Google engineers admit that at first, their program was a bit quirky. When asked a question, it would often give three responses that were in fact all identical, such as “How about tomorrow?” and “Wanna get together tomorrow?” and “I suggest we meet tomorrow.” In other words, rather than offer three choices, the program would offer one choice phrased in three different ways. Google’s engineers also discovered that their early prototype was overly amorous; the program would respond with “I love you” to almost anything. As they explained during Smart Reply’s launch earlier this month, “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.” Unfortunately, while most humans would already know (or we hope they would) that responding to “The documents just came back from the printers” with “I love you” is inappropriate, the program was having difficulty grasping which depth of affection is appropriate for which situation.
After all this trial and error, however, Google’s Smart Reply was rolled out, and so far, reviews have been more or less positive. The question that remains is what we are giving up when we give in to machine-generated communication?
The Limits of Machine Intelligence
As discussed in several previous eLeaP posts, machine intelligence, while rapidly evolving, continues to be limited. If you’ve ever talked to Siri, you’ll already know that machine-generated language is often stilted, at times riddled with non-sequiturs, and occasionally just off-putting and none of these things are qualities one wants to bring to their business correspondence. More specifically, when we rely on machines to generate responses, we downplay the soft skills that are what great business communications are all about.
First, there is no question that great communication is intuitive. After all, it is often about reading between the lines or beyond the literal. It’s about being able to understand what someone is really asking for, even when they are failing to do it directly. Intuition, however, is something one gains through experience. Only time will tell if Smart Reply will become increasingly intuitive over time. It seems likely, however, that while Smart Reply may become more appropriate over time, it will never be truly intuitive.
What is sincerity? At its most basic, it is a feeling generated when we believe that someone truly believes what they are saying. According to linguistic philosopher J.L. Austin, the author of How to Do Things with Words (a seminal text in linguistic philosophy and a key text on what is commonly known as speech act theory), sincerity is a key part of communication because in order for speech to do more than describe events—in order for anyone to actually “do things with words”—an utterance must be sincere (the speaker must mean what they are saying and the statement must be believable). But can a machine-generated reply be sincere? Might the way in which a response is generated compromise the sincerity of a speech act? Perhaps, these questions are best left to linguistic philosophers, but what we do know is that trite, predictable and cliché responses often appear less sincere than felt responses (e.g., those that appear to have taken time and thought to compose). For this reason, it seems likely, if not inevitable, that the rise of machine-generated language at the very least also risks compromising sincerity.
Cultural Nuance and Sensitivity
A final and by no means incidental issue is whether or not the rise of machine-generated responses will take cultural nuance and sensitivity out of business communications. As discussed in many recent eLeap posts, promoting diversity in the workplace is a complex and critical endeavor. Among other things, promoting diversity in the workplace requires paying additional attention to language. This may be as simple as using what some people might describe as politically correct language but in other cases, it can mean creating a space for linguistic differences and cultural vernaculars to circulate in the workplace too. Will Google’s Smart Reply be able to handle the true cultural and linguistic diversity of today’s workplaces or reify a specific version of “Business English” already in circulation? Again, whether or not machines can learn to speak with cultural nuance and sensitivity (e.g., to recognize that a word that once had a negative connotation has been reclaimed or to recognize that someone is using a word ironically that may be considered offensive in another context) is yet to be seen.
Is Smart Reply the Only Way to Deal with Email Overload?
While machine-generated email responses may represent a viable solution, the real problem is that we all have too much correspondence and as a result, frequently feel overwhelmed by our email. We can all help keep email to a limit by following a few simple rules:
- Only copy people who are implicated by the message; don’t go overboard cc’ing everyone in the office.
- Be clear the first time; one clear email is always more effective than sending several unclear emails.
- Use detailed subject lines; say as much as possible in the subject line and don’t recycle old subject lines for new subjects.
- Only address one subject per email; avoid sending emails that include too many questions about too many projects.
- Write emails as if you were writing a journalist article; put the main point in the opening paragraph and the least important point in the conclusion.
- Avoid letting your email back up; respond as quickly as possible but if you don’t have an answer, don’t be afraid to wait (avoid sending too many, “I’ll get back to you later” messages).
- Put limits on your email; despite the move to a 24/7 work life, know your limits and maintain them.
For more on effective business correspondence, also see the following eLeaP courses: