Our Approach

Sharing knowledge - measuring conviction.

How Sapient enables groups to become swarms.

The Software

SapientSwarm is a real-time, online tool for 4 to 40 participants. Each participant is represented by a Barracuda that is part of a ‘swarm’, or more accurately, a ‘school’. After the question is asked and 2-8 possible answers are presented, the game will start after counting down from 3. The process that follows, which can be seen here, often produces results that differ from that of polling all participants individually. Swarming filteres the majority opinion of the group through a unique dynamic arising from the slow, deliberate and transparent collective process of ‘schooling’. Instead of submitting their answers in isolation, participants can see their peers and use their vote wisely if their favoured choice is not in the running for top spot. An example: there are four political candidates for mayor and you do not know the likely winner. You cast your vote for the person whom you like best. It later turns out that two candidates received about 20% of the vote each, while the top two got 32% and 28% respectively. The winner is now announced and represents less than a third of the popular opinion. In an open ‘tug of war’ style process, 40% of the voters would have had the chance to switch candidates and put their weight behind either one of the top contenders, and the result may have been very different. In fact, the ‘loser’ 2nd place candidate could theoretically receive 68% of the vote.

The Great Barracuda (Sphyraena barracuda) is a solitary pelagic predator that can opportunistically form large schools for hunting or avoiding apex predators such as orcas. Schools benefit from safety in numbers, shared vision (meaning the movement caused by any fish perceiving a threat will trigger movement in others who do not yet perceive the threat themselves), the ability to encircle prey, and increased hydrodynamic efficiency. It is an apt metaphor for the human ‘wisdom of crowds’ in that it increases efficiency and reduces risk, while being voluntary and opportunistic.

A third benefit of the delay between initial positioning and arriving at the answer is that well informed but indecisive participants will be presented with a shortlist of answers. The UK Brexit vote is a good example. Six possible stock market reactions to the vote announcement between ‘rise sharply’ and ‘fall sharply’ were boiled down to the two extremes between decided participants. Undecided participants who observed the premature relief rally correctly picked ‘fall sharply’, regardless of the actual outcome, due to the superior risk/reward ratio. Instantaneous voting on the other hand would likely have resulted in an almost useless ‘even-steven’ type of prediction, such as ‘neither rise nor fall’ outcome, which would have been inconsistent with the volatility forecast of every single participant!

The Process

Our swarm forecasting process is much more than the online tool. We train our participants in the principles of 'Superforecasting', make them aware of typical biases and pitfalls in group decision making in general and the subject under discussion in the specific, and measure all swarm participation data over time. This allows us to coach participants when and how to lead with conviction, when to mediate and when to object. We also provide assistance with data analysis where needed and have a vast array of calibration options to suit all group sizes and styles.

Our Team

Sapient Predictive Analytics brings over 40 years of combined derivatives and commodity trading experience covering equity index futures and options, oil, gas and electricity, as well as several agricultural commodities and fixed income products. Our team is well-versed in the programming languages most used in machine learning, data science and web development. We are totally obsessed with solving hard questions and impossible challenges and you will find our team members engaged in a broad spectrum of competitions and competitive games out there. We love open source, wikis and decentralization, and hope our work can contribute to building a fairer, more sustainable information society.

Necessary conditions for smart groups.

Open source collaboration and huge crowds of humans walking with surprisingly few 'collisions' in a shopping mall are examples of wise crowds. 'Tulip mania' and mob justice show the dangers of crowds with the wrong motivation or composition. So what makes a crowd wise?

About the Canvas animation: This is a lightweight 2D JavaScript implementation of Craig Reynolds' Boids algorithm. It's a classic example of emergence and a suprisingly simple way of mimicking not only flocks, but any form of swarm or herd or crowd. Thanks to Hugh Kennedy, Github: hughsk