By Ryan McGreal
Published November 17, 2010
One of my favourite eponymous rules is Clarke's Third Law, named after Arthur C. Clarke, the engineer and science fiction writer who came up with the concept of geostationary communications satellites and whose plethora of novels includes Rendezvous with Rama and 2001: A Space Odyssey. It states:
Sufficiently advanced technology is indistinguishable from magic.
Earlier this year, an essay published on lesswrong.com borrowed this formulation for an ingenious take on the rise of content mills like Demand Media, the company behind those quasi-useful eHow articles:
Sufficiently advanced spam is indistinguishable from content.
I do urge you to read the essay, which does a great job of introducing the useful concepts of superstimuli and optimization by proxy.
A supernormal stimulus, or superstimulus, is an artificial stimulus that produces a stronger response than the natural stimulus it emulates and exaggerates.
For a canonical example, a chocolate bar is a superstimulus.
Humans have evolved a positive response to the taste of sugar, salt and fat because we need foods containing those dietary substances for survival. Foods naturally occurring in the ancestral environment - fruits, vegetables, roots, animal meats, and so on - contained these substances in concentrations that allowed for reasonably good nutrition and health if humans followed their taste buds.
However, a chocolate bar contains those flavours in concentrations far beyond any food existing in nature, and humans respond to the flavours much more intensely than we do to the flavours of naturally occurring foods.
As a result of the superstimulus, humans tend to crave and over-consume chocolate bars (and candy, and chips, and deep-fried batter, and so on) relative to our actual dietary needs.
This brings us to optimization by proxy. Human evolution has optimized for health - a difficult outcome to measure - via the proxy of taste - a criterion that is much easier to measure and correlates pretty closely with health in naturally occurring foods.
That optimization of health by the proxy of taste was close enough for government work before our sources of food were co-opted by processed food manufacturers.
The big pitfall of optimization by proxy is that it allows the opportunity for third parties to manufacture superstimuli that co-opt responses and crowd out natural stimuli. In our example, chocolate bar manufacturers co-opt the human attraction to sugar, fat and salt to crowd out fruits and vegetables.
The lesswrong article applies this phenomenon to the content farms that design their pages to function as superstimuli for search engine ranking algorithms.
Search engines want to return search results based on quality. A search should result in content that was produced to high standards and gives searchers what they are looking for. However, it's extremely difficult to measure what qualifies as a high quality search result.
Instead, search engines use popularity - i.e. the number of inboound links - as a proxy for quality. The premise is that high quality content will attract more inbound links.
Content mills design their pages to appear to search engines as though they are highly popular, which bumps up their rank in search results and crowds out more legitimate content.
When you do a search and get an eHow page as the top result, it's the equivalent of searching for food by taste and getting a chocolate bar instead of an apple.
I'd like to propose here that comment trolls do the same thing: they create comments that act as superstimuli for people looking to have a real discussion.
In my years editing and curating for RTH, I've had some thoughts on trolling and how to deal with it. What trolls crave above all is provoking outraged replies.
The term "troll" comes originally from the fishing method of dangling a shiny lure out the back of a boat while putting along just quickly enough to catch the interest of a fish. The fish that can't resist the lure is snared on the hook and eaten.
Crude trolls subsist on the meager attention given to their vulgarity; but sophisticated trolls can keep a debate going for days by dancing on the fine line between feigned reasonableness and deliberate obtusity. They are by far the more corrosive to online discourse.
Here's the critical insight: What makes trolls disruptive is not the trollish comments themselves, but the chain of outraged replies they manage to elicit.
That's where optimization by proxy kicks in. Trolls post statements that are more provocative than normal comments. They reply with a persistence that goes far beyond normal discussion. They shift arguments and evidence with amazing fluidity so that there are always more points to address. They press emotional buttons that weaken the rationality of their opponents. At their most sophisticated, they feign reasonableness without ever settling on reasonable conclusions.
In this vein I'd like to repurpose Clarke's law yet again:
Sufficiently advanced trolling is indistinguishable from discussion.
A discussion containing too much debate with trolls is like a diet with too much chocolate and fried dough: ultimately unsatisfying and even corrosive to good health.
Just as good nutrition in an environment rife with superstimuli requires a more calculated approach than simply eating what's tasty, so too does good discussion in an environment rife with trolls than simply challenging every ridiculous comment with a reply.
The reason RTH instituted that comment voting feature was to break the chain of well-intentioned participants endlessly debating trolls by giving the community an opportunity to establish shared awareness that a given user is not debating in good faith.
However, the system doesn't work perfectly (after all, comment voting itself is simply a form of optimization-by-proxy) and is sometimes abused to express disagreement with an unpopular opinion rather than disapproval of an inappropriate comment.
Trolls, in turn, exploit this mismatch by insisting that votes against their comments amount to the former rather than the latter. This plays to the sympathies of legitimate commenters with strong inclinations toward free speech and, combined with a careful application of reasonablish comments at key points, allows them to go on displacing real discussion for a long time.
I'm trying to figure out what enhancements or changes we might make to the RTH community moderation system to make it easier to "out" trolls - even the sophisticated ones - while drawing a clear distinction between trolls and legitimate contrarians making sincere arguments from evidence. Any ideas on how best to do this are welcome!