I have already discussed the problems with email. Two of the weakest features are the inbox – which tends to grow out of control and hide from me what I care about most – and folders, which are just about more trouble than they’re worth. I have proposed three essential dimensions of email – time, people, and topics. Now I will discuss these dimensions from the point of view of human beings, look at how current software addresses them, and suggest where we can look for improvements.
Previously I suggested that email isn’t about messages. Instead of treating email as a collection of individual messages, we should instead focus on people. In fact, we should shift the focus from messages to all three of these dimensions.
Time is the most obvious way to categorize email, and indeed all clients use it as the primary view. Unlike people and topics, time is continuous, so sorting by time is natural. However, this is already the default sort order for messages and it has not solved the Inbox Problem. Perhaps sorting is the wrong way to look at time.
Recent messages – especially but not only unread ones – are fundamentally different from older messages. Currency divides all messages into two natural categories: current and archived. This concept of currency is critical, and it’s what the inbox is all about. The inbox metaphor itself points to this: an inbox is for new correspondence; older material is archived in folders. Unlike time, however, currency is not continuous: a message is current, or it is not. It does not necessarily follow that current messages should be sorted by time. Ideally, they should instead be sorted by priority. Priority is not easy to determine, but it is almost certainly not simply a matter of ordering by time. (For archived messages, time can be even less important. We search for the sender and subject of an email at least as often as when it was sent. Here it is legitimately used for sorting once we have already selected a group of messages by person or topic.)
Remail and the 2003 version of Outlook have addressed time by grouping messages by day, and in the case of older correspondence by week. These groupings correspond to the natural cycles of our lives, and are also used by our minds to divide memories. The most obvious human rhythms are days, weeks (due to work), seasons (for which months are a good climate- and latitude-independent proxy), and years. For the more distant past, our memories tend to compress time, so larger groupings are more practical. These natural units of time are a very useful for dividing a huge volume of correspondence into meaningful chunks.
The meaning of an email message is not only its subject and content. When we send or receive email we are participating in a conversation with one or more people. For human beings, the identity of the sender is often enough to determine what the message is about and how important it is. In fact, the content of the message may not be the most important piece of information it carries: the act of sending a message is itself meaningful as a means of keeping in touch. A message from a long-lost friend is important simply because it is sent and received, regardless of what it has to say. The sender is useful not only for sorting, but for much more detailed analysis of messages.
IBM’s Remail project provides an important insight with their correspondents map, which displays a list of the user’s contacts. These are colour-coded according to whether the most recent message from that person has been replied to, and if not how long ago the most recent message was received. The insight is this: the person is what’s important, and the most recent message from a person is almost always the most relevant. Like the latest issue of a newspaper, it casts in shadow what has gone before.
Email clients have far more information about a person than just their name. This is because email has a history. While an individual conversation with someone may be brief, we often hold many conversations over a period of time. This pattern of correspondence is a rich source of information about the relationship between ourselves and the people we are in contact with: Do we reply quickly or at all to their messages? Do we read them immediately or pospone them until we have dealt with other matters? Do we write long messages or short? These are all clues to the importance of the relationship. They offer the possibility that the software can take advantage of some of the clues we use when we prioritize our mail.
The most important aspect of a message is usually its content. As an indication of this the topic of a message is very useful, and there have been a number of attempts to capture topic information. The most common are folders and message flags, but these techniques require human intervention to categorize messages and have met with limited success.
The problem with topic is a limitation of software. Topic is about meaning, something which computers don’t handle well. If the computer cannot determine topic, then the effort is left to the user. This can be obtained in three ways. The first is for the sender to provide it, and this is done in a limited way by the subject line. The second is for the recipient to add the information when the message arrives, e.g. by flagging it or dropping it in the appropriate folder. The third approach is to not add the information at all, but to search for messages using keywords and other information which corresponds to topic.
These two problems – of encoding information in a structured way that computers can use, and of searching existing information – have been the subjects of much useful research. Most email software allows the user to search for messages containing specified words. This is useful for archived messages, but takes too much effort to use for current mail. It would be wonderful if the software could automatically prioritize and group messages according to content.
Oddly enough, this is already being done, but only for one kind of message: spam. The more advanced spam filters use artificial intelligence techniques to “learn” how to recognize spam. Mass mailers attempt to trick the filters, yet the spam filters are quite accurate. This same technology could be applied to other incoming mail – it might even be more successful, since legitimate mail is not written to trick the software.
Even without artificial intelligence, messages already include some structure. The main clue we have to the topic of a message is its subject line. This can be used to group messages with the same topic. Threaded message views take advantage of the fact that each reply message encodes within it which message it is responding to. The limitation of these approaches is that they work well for long conversations, but are not useful for analysing brief independent exchanges. Nevertheless, both techniques – particularly threaded views – can be invaluable, but in practice are seldom used. The reason, I believe, is views, which I will deal with next.