Daniel Lemire's blog

, 3 min read

We need more than spam filters: we need bona fide assistants!

I don’t know how other professionals behave, but if you email a researcher at 11pm or even 2am, there is a good chance he will get back to you within two minutes. Many of them are workaholic, at least as far as email goes.

According to my Google activity report, not counting spam, I receive 5000 emails a month and I send back about 1000 emails.

To preserve my mental health, I have decided to set some time aside every day for my family and for relaxing. I play video games, I read novels, and I watch TV shows (on a tablet). For this unwinding to happen, I need to avoid email in the evening. Otherwise, the work never ends.

The net result is that I have accumulated emails in the morning. So the first thing I do at 9pm is reading my emails. As I write this blog post, it is 10am and I am still grinding through emails that were sent last night. As I process these emails, more tend to appear.

The net result is that I often finish reading my morning emails at noon. The time I spend on email tends to increase, year after year.

It is not as bad as it sounds: most of my work involves writing emails anyhow. For example, much of my research is coordinated through email. Also, the emails I spend time on are truly important. They come from students, research collaborators and key colleagues.

Nevertheless, I often feel guilty. After all, giant scholars often avoid email altogether. Maybe I would produce brilliant work if only I did not spend so much time with my email.

To cope, I try to process emails in batch. That is, if you email me, you can expect long delays before I answer. It can take a day or a week. This seems to violate some social convention as I routinely find that people decide that I have ignored them because I haven’t gotten back to them immediately. But if I did try to get back to everyone at once, I would morph into some kind of email robot. I would never got to write code or research papers anymore.

This being said, as the amount of interesting emails I receive keep on increasing, and people’s expectation increase, I worry. It seems that there is intense pressure to get back to more people faster. I might soon have to declare email bankruptcy. I will simply receive too many worthy emails for me to process it all.

Yes, we have fantastic spam filters. And these help a lot. And my email client (GMail) tries its best to identify the most relevant emails and put them in a priority inbox.

However, it is quite clear that we are going to need even more help. CEOs and important folks have human assistants to sort through their mail and answer common queries. Yet I am not an important person, I am just a regular Joe. Still, I too need assistants. The answer seems obvious: we need clever AIs that can process the bulk of our email.

How could this work? Let me run through some examples…

  • A lot of my students email the same queries. Very often, they are asking for more time to complete their assignments. I tend to grant these requests. A software assistant could recognize this pattern and help me process these emails with less effort.
  • When I work with a student or a collaborator on a research project, we send a lot emails back and forth. I find that we routinely “forget” about past discussions and turn in circles. Moreover, it is hard to track ongoing tasks. Who is doing what and when? If I want to cleanly archive discussion, I need to do it manually. I get no help in organizing the discussions from my email client. If I want to be reminded of a deadline, or of the need to follow-up on what a student is doing, I need to set reminders myself. Overall, email discussions are much more effort than they could be.

It seems clear that there is a huge need for more email-related AI. These problems are absurdly difficult given the current state of our knowledge, but I am depressed at how little effort seems to be put on these important practical problems.