The ring

Japan Internet Report No. 64 Spring 2002

Extract:

The flood of junk mail and unsolicited, automated single-ring-and-hang-up solicitation calls (``one-giri'') sweeping through Japan's cellular market has raised the mobile annoyance factor to unprecedented heights.

Of the 900 million messages that go through DoCoMo's servers each day, 880 million (98 percent) are spam, according to the company. The problem is that, regardless of the source of the message, subscriber phones ring (or vibrate) every time mail arrives.

-- Daniel Scuka, Japan Inc. Wireless Watch No. 60

Of course you can have three categories of mail: rejected mail, mail that appears in the device but is too suspect to ring, and mail that does deserve the ring.

Here are some simple strategies that the machine might use.

• only friends on a whitelist can send ringing messages. Pretty simple, right? Analagously, many voice handsets can assign unique rings to phonebook entries, so you can implement a blacklist by setting entries to ``no ring,'' but I don't know about a whitelist, like ``only caller ID matching my phonebook may ring.''
• pattern-recognition sifts out all those ``forwards'' that relatives and distant friends with nothing to say send you in place of personal correspondence. The most notorious forwarders usually don't do any editing, so programs like SpamAssassin should adapt easily to this task. The math and the method is done already, but we do need some harness to invoke SpamAssassin separately, once to find spam, and a second time to find annoying mail, and treat the two differently. This is hard because it needs user configuration, while many ISPs do spamfiltering that isn't user-confiugrable.
• users can rule in advance on the ringworthyness of a thread---all replies will obey the verdict delivered to a thread's parent article. This could help tremendously when you have five friends bantering about something over Cc:'ed email, by turning twenty rings into one initial ring.
• set up two addresses---a ``main'' email address that doesn't ring and a separate adjunct address that both rings your wireless and also sends a copy to your main account. The main account archives wireless messages along with all the other trivial crap you get, while the wireless account acts as an ephemeral read-and-delete newsspool for special messages sent to the ringing address. This means you have to read wired email, too, because mail sent to the ``main'' address not only doesn't ring, it doesn't even get sent over the air.

  I think this is dumb because all mail should come to one inbox, then get sorted out by a good email-reading program (an MUA). The two-address strategy reminds me of the foolish people who used to refuse to buy celfones because ``I don't want to be reachable all the time.'' Fine. then turn off the handset, or use caller ID to screen calls. If you're leaving your house to avoid phone calls, then maybe the wrong people have your phone number. Why would anyone object less to invasions and interruptions while at home than while in public? I would expect the opposite.

  Adopting the dual-email-address trick may not be quite the same as this technomidget fallacy, but it's definitely similar in that it mistakenly blames annoyance on wireless rather than on the Ring.

There are adaptive systems, too. Right now machine learning people play with these mostly for web browsers, but I think they could apply to ringing email. A hypothetical adaptive system might watch how long you spend reading a message and, if you re-read the message, when you read it, so it can attempt to guess urgency and interestingness separately. It might give you virtual highlighter and anti-highlighter pens so that you can mark exceptionally interesting and useless body text explicitly, but in general explicit training is a fascist concept that works poorly with machines---I think the user tends to be either too lazy or too angry and condescending. If the keitai has a press-and-turn scrollwheel, adaptive scoring might know that you're reading the message when you're pressing the wheel and turning it slowly or not at all. If you let up the wheel, you're probably distracted, and if you spin it quickly you're probably skimming. In this scenario, the small screen turns into an asset.

The email client may have an adjustable threshold to set how interesting a new mail has to be before it rings at this particular moment. Similar to how Ericsson handsets have a shortcut to enter or exit ``silent mode'' by holding down the 'CLR' key, you could tell it ``ring only for very interesting messages,'' or ``ring for anything.'' Or the keitai might have different ring patterns to indicate interestingness.

In addition to the simple phonebook-whitelist ``ring anything this person sends, always,'' you probably want a checkbox when you send a message, ``ring any replies others make to the message I'm sending that arrive within the next 24 hours,'' in case you are expecting an email for some immediate plan-making purpose.

Microsoft will of course be the first out of the gate to add an idiotic sender-settable ``make my message ring when received'' check-box for people sending mail, analagous to their 'X-Priority' and ``Return Receipt'' idiocy, but I think this trivial solution is deceptively appealing and is quite wrong, even inappropriate and rude, and the sort of thing only a clueless buffoon would dream up.

Why is sender-settable so bad? First of all, because the recipient's machine intelligence can easily outperform the herd of senders. But there is a more fundamental reason: because this is part of the great leap forward that ringing email offers us. Wouldn't it be nice if we could score and filter our incoming phone calls so that some of them ring and others go to voicemail? Ring is not the sender's decision, and because this is a matter of etiquette, whose decision it is should be transparent to both sender and recipient.

Another sender-settable system is to use many addresses for the same inbox. Most open mail systems support ``plusing,'' a scheme popularized by CMU's Andrew network. For example I can receive mail sent to <carton@Ivy.NET>, <carton+spam@Ivy.NET>, or <carton+[anything]@Ivy.NET>. I can tell which address the sender used by looking at the 'Delivered-To:' header. Multiple addresses is not really a useful filtering technique so much as a way to catch people who share your address without your permission. Look at it this way. There are two fundamental ways to filter mail: based on who-sent-it, and based on its contents. If you had hypothetically ideal perfect control over privacy of personal information so that no one shared your address with anyone, this plusing scheme would be equivalent to filtering based on who-sent-it. If that's what you want to do, what advantage does the plusing scheme have over filtering on the 'From:' header, as with a whitelist?

Besides, email addresses are fashionable tokens of vanity. People compete for cool addresses in congested namespaces. Addresses shouldn't be chosen based on mundane filtering issues. And addresses should describe a person semi-permanently---they should not follow the metaphor of a self-addressed stamped envelope that one can only use to ``reply'' rather than to reach or to find.

Mail aliases and sender-settable GUI widgets don't substitute for advanced filtering in the MUA. Ringing wireless mail needs this filtering. I'm not thinking of literalist, primitive boolean header-scanning, ``if from: = jojo and subject: = lunch''. Forget that junk. I mean stuff like SpamAssassin that can have hardwired rules to recognize complex generally-offensive patterns like forwarded chain letters, or the sneaky adaptive scrollwheel system that spies on you and tries to learn what you find interesting.