How smart are Gmail’s ‘smart replies’?

When Seamas O’Reilly responded to all his emails for a week using only Smart Reply, our columnist’s messages suddenly became spookily jaunty. Did his friends spot the difference?

The real me: ‘The messages seemed written in a tone that I have never used in my entire life.’
 The real me: ‘The messages seemed written in a tone that I have never used in my entire life.’ Illustration: Pablo Delcan/The Observer

The philosopher Jeremy Bentham was famed for his panopticon, a hypothetical circular prison that was designed in such a way that its inmates never knew whether or not they were being observed. This would, his theory went, encourage prisoners to presume they were always being watched, and thus act accordingly. No true version of the prison was ever really built, and the word itself only now lives on due to its prodigious utility within breathless op-eds about surveillance culture, mostly written by people who’ve already overused references to Orwell and Kafka.

The genius of today’s boring dystopia has been to offer this surveillance as a feature, not a bug; to cast that all-seeing-eye not as a malevolent shadowy jailer, but as the world’s most boring personal assistant. Nowhere is this truer than with Gmail smart replies, the pocket panopticon that now resides in every inbox. Not only can it see what you’ve already read and written, it has some great ideas on how to make your next contribution, too. But how well does it really know us? How deeply does its unsleeping, lidless eye scan our thoughts and deeds? And could I use this information, the knowledge of a god, to create a stronger, better, smarter me? Seeking answers, I stopped resisting and spent an entire week surrendering to its every whim.

It’s odd that Gmail smart replies have only been around since 2017, since they’re the kind of subtly ever-present tool that now seems always to have been there. Odder still, it feels this way despite the fact I had never used it before. My memory of seeing, and ignoring, the little options below my emails was that they seemed generally accurate, but also glib and oddly worded. More importantly, they seemed written in a tone that I have never used in my entire life, one of indeterminate, cheery abandon that would surely present as false were I to ever use it myself.

But God has his reasons and I endeavoured to discern them for myself. Opening my inbox I found an email from my wife telling me she had booked a restaurant and hoped I thought it would be as great as she did. Below it lay three unassuming rectangles poised for my selection:

– OK, thanks!

– Yes, I agree!

– I don’t think so.

Opting for “OK, thanks!” I began my smart reply odyssey and a descent into what quickly became the world’s most boring Choose Your Own Adventure game.

Starting off, I had thought the main issue would be accuracy. I worried I’d be forced to make wildly inappropriate remarks to longstanding friends and colleagues, like the hapless idiot in an American sitcom, receiving ear-piece instructions to get him through a date with the hottest girl in school. You know the scene: he’s doing great for a while, but eventually comes a cropper when his Bergerac is hit in the crotch by a wheelie bin, or a nest of wasps, which causes our hero to repeat his every swear word verbatim. This, I feared, could be my fate; glibly responding to emails announcing the death of a family friend with blandly inappropriate corporate speak like “Outstanding!” or “Can we change this to Friday?” It turned out I shouldn’t have worried, as the replies were almost universally, even unnervingly, appropriate to the topic. My week was merely a never-ending spree of palatable options that were terrifyingly bland.

The major stumbling block to my ever having used smart replies before was tone. Smart replies always gave me a sense of alien weirdness, so much so that when I did begin seeing them, I’d go out of my way to make sure the wording I did use in my reply was nothing like those suggested. It was as if I was spiting this assistant out of fear I’d be revealed as a Big Data techno-stooge. It’s not just that the replies are curt and impersonal, they’re also so chipper as to sound demented. I have a reputation to uphold, and it’s not as a perky yes man. Nowhere was this more apparent than with my friend Al, who was good enough to email me a particularly chucklesome video he’d found on YouTube. To my horror, my only options by way of reply were:

– Love It!

– Thanks, I’ll check it out!

– Hahaha that’s awesome!

Since it was, at least superficially, closest to the feeling engendered by the video, I went for “Hahaha that’s awesome!” making it, I think, the first time I’ve ever used that latter adjective in my entire life. One thing that struck me was how Gmail had worked out that the video itself was funny. I mean, it was funny – it was a wedding band from the Irish midlands who appeared to have made a music video for roughly the price of a Quavers multi-pack – but that’s not the point. The fact is, this wasn’t stated anywhere in the email itself. How did it know? Would “Hahaha that’s awesome!” still have been an option if the video he’d linked to had been Kofi Annan giving a dignified reading from Angela’s Ashes, or a compilation of old ladies receiving larger-than-expected gas bills?

The other big take away, and one that speaks to my larger point, is the rather poorer job it had done of approximating my own speech. It knew enough to suggest it was funny, but not enough to stop me from sounding like a hyperbolic youth pastor on bad speed. This was to be a constant theme in my week of robo-replying, the sense that Gmail wasn’t just reading, but remodelling me in its own ghastly image.

“An exclamation mark,” wrote F Scott Fitzgerald, “is like laughing at your own joke.” God knows, then, what he would have made of my output this week alone. “Yes!”, “Great, thanks!”, “Will do!”, “Got that!”, “See you then!”, “Of course!” Now I was laughing at my own joke a dozen times a day like some half-mad cyborg, rendered dizzily incontinent by even the mildest exchange of words. Google, it turns out, splits its sides for emails confirming dental appointments or lunch dates. Somewhere in the beeping digital ether, the hive-mind wipes hot tears from a dead digital eye, weeping with laughter at the thought of a Thursday lunch meeting. My emails suddenly presented me as a giddy type, pummelled into hysterics each time an attachment was received as stated. Running on the spot and punching the air at the thought of buying new bin attachments from Hackney council by the weekend.

Google made a big point of claiming smart replies respond not to your own style of writing, but to a generalised neural network of everyone’s responses to similar cases. This was, I presume, intended to placate those who felt the opposite was too creepy, but I began to find it unnerving, too. After a week I started to accept this was not Google’s idea of who I was, but their conception of some new, better me they felt I should become. No more Séamas the recalcitrant Scrooge, indifferent to the ecstasy of being asked, for a sixth time, if I was attending a social function. Step forward my new, better self, one who screams: “Yes, I’ll be there thanks!” with something like the manic ebullience of someone hoping to deter you from entering the bathroom they’re in.

If it were just the case that smart replies made me gormless and weird, that would be one thing, but there were other, unexpected hurdles. One email I received was from an event organiser addressing an overdue payment for a talk I’d given. I like this person and readily understood it wasn’t their fault, so was only too happy to offer my polite thanks for the update. But I was then surprised to find I had no choice either way. In response to being asked if it wasn’t too much hassle for me to be paid at the end of the month, my options were:

– That’s fine, thanks!

– Of course!

– Nope, that’s fine!

Rather like the child who wakes up with bold plans to clean his room, only to have this new regime interrupted by a parent demanding they tidy up, Google had removed any choice in the matter and, in the process, made me resent options I may well have embraced on my own. Gritting my teeth, I selected: “That’s fine, thanks!” and reflected on the fact that Google’s responses to my non-payment ran only from moderately cheerful to fizzily delighted. What might that say about the silicon values now beating within my inbox’s beige, beeping heart? Whatever the case, I was beginning to chafe against my Big Data overlords and felt ready to call it a day.

Following a week of emails in this vein, I was looking forward to reverting to normal and letting a few of my more common recipients in on the gag. When I did, I was horrified to discover no one had noticed a thing. Not even Al, who’d seen me use the word “awesome!” and hadn’t raised an eyebrow. I found myself offended that the ruse I had pulled on unsuspecting friends had worked so well, forced to reckon with the fact that my sense of individuality was, clearly, somewhat overrated.

The blandness of Gmail’s smart replies had been their camouflage, the processing power of 10m moon landings, disguised in a sea of inane pleasantries and errant exclamation marks. Worse still, I can’t stop scanning messages I receive, detecting the hidden, mechanical hand of Uncle Google in every misplaced Americanism, or brisk and jaunty phrase. I no longer know which of my friends, colleagues, or bin collection services I’m even really talking to any more. A week inside the machine has changed all that. We must, as always, be careful gazing long into the abyss, because the abyss gazes back. And That’s fine!