And a more challenging one yesterday, in a locally-small, though genuine ‘emergency’ situation. Which led me to these thoughts http://is.gd/BGkw about the behavioural aspects that need to be instilled if we are to get the best out of memes, and avoid some of the time-wasting they can generate.

Erik – the firehose is the thing, isn’t it? From an operational point of view, I feel more like we suffer from data droughts rather than information floods (paper available here, now out of date), but once again it depends what you need the information for… information for informations sake is not our goal!

1) We know now that there exists a lot of information out there on the web in the form of images from flickr, streaming updates from twitter, txt messages from Ushahidi and the traditional news from the various bureaus.

2) we believe that there exists valuable information within this stream of events and that we can extract the signal from the noise.

3) we believe that the current state of the art sorting algorithms and filters can be combined with human intelligence to create a tool that probabilistically assigns a likelihood to an event.

The science is not at the point where we know for sure whether points (2) and (3) will turn out to be true. The key problem is how to assign weight to an individual story, how to combine the various available metrics to determine reliability of the source. This can be a combination of crowdsourcing, algorithms, geographic details, past history of user, repeatability of the information. Not an easy set of variables to combine together and indeed a difficult problem, but one that if solved could be a very important tool. As with anything of this nature it is trying to determine where the bar of certainty lies – too low and you get noise, too high and you miss out on information.

As to the final point about the twitter meme – i.e. “someone might abuse the tool” – this is always a possibility. As engineers you create tools and they have the power to do good and do evil. What has to be controlled is the algorithms and community structure to ensure the probability of false events becoming ‘real’ is kept to a minimum.

Neither way can be completely accurate, and neither way will be completely accurate when combined either. However, my hypothesis at this point is that one (or both) can lead to having a greater probability of telling “better” information from “worse” as we deal with a larger and larger firehose of incoming data.

It’s a beginning, and we’ll likely learn much along the way that proves many of our presuppositions wrong. Such is the beauty of trying new things, many times you end in failure, but with continued iteration, something good can/will finally come of it – even if it’s not from us.

Perhaps there’s a very useful place in something like SwiftRiver for “old” sources of information and authority to contribute to rating or verification of the crowdsourced river. This is part of the Ushahidi model anyhow: the concept that there are trusted entities (like NGOs) with sufficient knowledge, resource and credibility to make judgements about information.