The Optimal Segment

Yesterday’s post got me thinking again about the current state of AI and the limits of automation. Services like CrowdFlower or Mechanical Turk are proof enough that it’s still much more viable to build a massively scaled human infrastructure than to make use of the – comparatively – tiny advances in AI we made since the first Turk was built. We’ve been waiting for the AI breakthrough for decades, and might have wait for decades to come.

Now a crowd, per se, is not particularly intelligent just because it’s a crowd, cf. The Content Anthill. Rather, a segment of a crowd might be intelligent enough relative to a given task. The issue, then, and the business case, is to match the task to the best of the many possible segments.

The larger and more diversified your crowd (company), the more possible matches of tasks (say, content creation or governance efforts) to segments (teams). And as soon as a potential match is found, it’ll take precious days until its quality is determined, and you never know if you’ve arrived in a local optimum or found your dream team.

Maybe finding the dream team is just a result of swerves after all.