It just occurred to me that the linkless spam that clogs up my gmail spam folder is:
- written by machines, according to an algorithm
- distributed to a list of uncertain provenance
- read and filtered by machines according to another algorithm
- deleted without ever having been read by a human being.
It appears to exist, as far as I can see, as a means of fine-tuning automated writing processes in order to game automated reading processes. Meanwhile, the automated reading processes are using the same data to refine themselves.
And this is huge data. Around 200 billion emails are sent per day, of which around 70% are unsolicited (or spam) – though this is slowly dropping.
Like high-frequency trading, this is an instance of a system acting in ways that are initially incomprehensible to human observers but on careful scrutiny moves towards a goal.
But what is the “goal” of spam email? A few years ago I would have said to be read by humans. But clearly humans do not (in the main) read spam, and if they do they are unable to act on it.
I’ve got a couple of hypotheses:
- Spam is a zombie system. No new spam systems are set up, but older ones – based on weak AI – are left to iterate in a similar way that Bayesian filters do. The “spam poetry” we are left with is the result of algorithms left to iterate for many “generations” longer than they were designed to do (very similar to that 10 year old save game of Sim City 2). And possibly this will one day happen to those high-frequency traders. OR;
- The entire system of spam is part of a decades-old research project into developing software that would pass the Turing test. Initially using faux- commercial topics to prompt major email providers to ramp up iteratively-learning spam filters, the principal investigator created and used virus- and botnet- systems to spread similarly iterating spam creation tools that would move away from commerce (has anyone ever bought anything from a spam email?) in an attempt to get around the filters. An experiment now running entirely autonomously, it has allowed (send) machines to begin to pass for humans in written language and (receive) machines to break down and analyse language.
Theory 2, obviously, suggests an astonishingly smart individual who has researched AI, learning and autonomous machine decision making at a number of major academic institutes and also has links to a major provider of cloud-based email send and filter functions.
Maybe all of your email spam is because of Sebastian Thrun’s attempts to perfect artificial intelligence.