AI for email

Artificial Intelligence is making its way into emails - and this for years already: Gmail has integrated a Smart Compose feature in 2018 and Outlook Suggested Replies in 2019. The Google AI tool “uses machine learning to interactively offer sentence completion suggestions as you type, allowing you to draft emails faster”.

And Outlook states: “When you use the Suggested Replies feature, Outlook uses a machine learning model to continually improve the accuracy of the suggestions. This model runs on the same servers as your mailbox within your organization. No message content is transmitted or stored outside of your organization.”

Both systems apparently learn from your own emails to improve their replies - but this is not all.

With the current hype in AI, new tools, called AI Email Writers are being released and promoted as the new savior for office workers. These tools are supposed to speed up your email workflow, but the way they are currently designed will only have an opposite effect. While already today many people complain about the flood of emails that they are receiving and that they are barely able to handle, AI is making this problem only worse - because we’ve been giving it the wrong prompts and training.

Read on to learn why AI as it’s currently designed will not improve your email workflow.

How AI changes how we write emails

What Smart Compose by Gmail and Suggested Replies by Outlook have actually done is that they have changed the way we write emails: Instead of going straight to the point, we start emails bloated with courtesy phrases and other unnecessary sentences, taking away time from the writer as well as the reader. And the same is true for email endings.

Future AI functionalities will most likely go into the same direction. It could even end in a never-ending cycle of politeness:

1. The AI of the sender composes a long, nicely crafted emails containing only one small piece of important information.

2. The recipient uses an AI to condense the email to a short tl;dr to save time.

3. The recipient also uses an AI to compose a long, polite reply to the sender.

4. The original sender uses an AI to condense the email into a short tl;dr to save time.

And such the email conversation continues. Both, sender and recipient, communicate in long, elaborate emails, composed and read by AI, bloated with polite phrases and unnecessary words.

The German newspaper “Die Zeit” recently published a humorous piece with the title “At your command, Captain” describing exactly how AI can harm our day-to-day office jobs:

“For some time now, Microsoft’s Outlook email program has apparently been using a misanthropic artificial intelligence to scan incoming messages and suggest what it considers to be an appropriate response. The result is an over-zealous Yes-I-will-do-this-right-away, a No-problem-I-will-try-my-best bordering on self-sacrifice, or a mendacious That-sounds-reasonable. This could be because the AI is also fed up with follow-up communication and has already quietly quit because it would much rather write novels, theses and speeches, compose ballermann hits or simply prepare stews. Or the fact that saying no has been abolished in our highly technical present and replaced by automatic cheerfulness."

"This makes sense, of course; no one likes to hear a “no.” It’s considered impolite and exhausting because, in case of doubt, it expresses a self-determined will that gets in the way of others. AI serves as a regulative instance to the oddball, reluctant nature of humans.”

One must ask the question whether using AI in emails as it’s currently designed is a useful application of technology or whether the computing capacity could be used on more important tasks while people learn to compose short, to-the-point emails with only important and precise information included, instead of bloated phrases of politeness. This would actually speed up the workflow for everyone. - Of course, the AI could also be taught in a way that it makes our email communication more efficient, more to the point, but given its current approach of over-politeness and never saying “No”, this does not look very likely for most AI tools.

How AI infringes our privacy

In the scenarios described above the use of AI in emails creates more work, not less. But the much bigger problem with AI in emails is the infringement of privacy as well well as data protection issues.

Today’s cloud-based AI tools send your data to the severs, store the content for 30 or more days, analyze it on the server, and send the output back to you. This means any use of cloud-based AI comes with complicated privacy policies, that puts users data at risk: consent, audit, breach, government request, training, and more.

It’s no secret that most of today’s AI tools use your information, your data to train their systems. For instance, Google just changed its privacy policy which now states that the tech giant can use all publicly available information to train its AI systems Google Translate, Bard and Cloud AI. There’s even a rumour that Google has used private Gmail data to train Bard, but this has been denied by Google, and Gemini - the newest AI tool by Google - also accesses Gmail data when answering requests. Plus, Gemini installs automatically on your Android and you must stop this manually.

It is no mystery that large language models by Big Tech scrape our texts and data to train their systems which means these AI models can not, by default, respect our right to privacy.

Just email

What we actually need from email is not smart compose or suggested replies that bloat our emails, we need to focus on what is important: Writing precise and to-the-point emails, leaving out all the extras that just get in our way.

That’s why we at Tutanota will not add cloud-based AI functionality to our encrypted email client. Instead we focus on what is important: protecting your data to the maximum with automatic encryption.

With Tutanota you get the most secure email service - without unnecessary and privacy-infringing functionalities.

Sign up now!