Quick overview: Sandy Pentland subscribes to the kind of ecological view that a lot of my recent sources have espoused- an emphasis on the relations between objects, instead of on the objects themselves. He argues for a “computational theory of behavior”, using Big Data and a system of collection/observation that he calls “reality mining”: the point is to grow the fullest, richest models of social behavior possible in order to capture all of the details, the micro-patterns, that traditional methods would overlook in favor of averages and stereotypes, and to capture them in “real life” as it is lived, instead of in a sanitized environment. Causation can be worked out through “quantitative, time-synchronized predictions” to make other explanations implausible, and then we might seek supporting quantitative lab experiments.
The two most important concepts in social physics are Idea Flow and Social Learning.
- Idea Flow exists within social networks, and can be “separated into exploration (finding new ideas/strategies) and engagement (getting everyone to coordinate their behavior)”.
- Social learning is “how new ideas become habits”. Learning can be “accelerated and shaped by social pressure.”
Some more specific applications of this information can be found in the previous installment of notes.
This is part two (of presumably five?) of Pentland’s Social Physics. I’m roughly following the sections that the book lays out, in order.
- Part 0: Some immediate personal/social context, what I’m thinking about before reading.
- Part 1: A broad summary of the book, the methods involved, etc.
- Part 2: Heavier details on the foundations of Social Physics (Social Learning, Idea Flow, Engagement)
- This part will be about Organizations (a favorite topic on this blog!)
- Part 4 will be about “Data-Driven Cities” (a recent favorite topic on this blog!)
- Part 5 will be about Policy (mostly avoided on this blog…)
This time around, I will probably be a little sparser with references to the studies that backup the results.
Organizations are sustained systems of information flow. Group intelligence, which we will measure by idea flow, is “as important a factor in performing group performance as IQ is in predicting individual performance.”
“The largest factor in predicting group intelligence was the equality of conversational turn taking.” The second largest factor was the social intelligence of the group’s members.
Briefly, on method: “The badges used in this experiment, and in other research studies from my lab, produce detailed, quantitative measures of how people interact. Typical variables measured include: the tone of voice used; whether people face one another while talking; how much they gesture; and how much they talk, listen, and interrupt each other. By combining data from individuals within a team and comparing it with performance data, we can identify interaction patterns that make for successful teamwork.”
What this sociometric data showed was that the pattern of idea flow by itself was more important to group performance than all other factors and, in fact, was as important as all other factors put together [eg. individual intelligence, personality, skill, etc.]
Pentland (with colleague Wen) identifies three simple patterns that account for “approximately 50 percent of the variation in performance access groups and tasks.” Highest performing groups include:
- Large number of ideas (many short contributions instead of a few long ones)
- Dense interactions: a continuous, overlapping cycling between making contributions and very short (<1 second) responsive comments (“good”, “aight”, “what?”) to validate and invalidate ideas and build consensus.
- Diversity of ideas: everyone contributing ideas and reactions, with similar levels of turn taking among the participants.
- The decision needs to be made now. Discussion may not be available. We need a tyrant.
- High-running emotions, socially broken team. We need a facilitator.
The interventions to resolve these exceptions should take as little time as possible.
Pentland speculates an evolutionary basis for why our ideal social dynamic is the way that it is. Essentially, language is a new layer of communication in species-time and older methods of signaling were visual, quick consensus-building activities to get basic things coordinated. No need to dig too deep into that bit.
(Social) engagement is the central predictor of productivity. Study example: Pentland et al. improved call center metrics by allowing whole teams to go on break at the same time, instead of scheduling breaks on an individual basis. The result was more inter-team informal communication, and higher productivity during work.
Pentland posits that there are two distinct processes that create the “star-shaped network” that is most “intelligent” (i.e. “best for idea integration, and behavior change.”).
- Exploration, when the team members interact with other teams and diversify, and
- Engagement, when team members interact with each other and converge
“Qualitatively, this is what the Bell Stars study [found]: Star performers became familiar with different perspectives of their work. Senior management, customers, sales and manufacturing groups all have different views, and the combination of their ideas with those already in their work group were a major source of useful creative thinking.The difference today, of course, is that with sociometric badges we can now actually measure this exploration and ensure that it is both frequent enough and sufficiently diverse.”
A summarization of the Kahneman-ian theory behind these findings:
The connection between creative output and diversity of experiences seems to be due to the power of unconscious cognition. There is considerable evidence in the scientific literature showing that unconscious cognition can be more effective than conscious cognition for solving complex problems. Our fast thinking seems to work best when our logical, slow-thinking minds aren’t interfering, ssuch as during sleep or when we are turning an idea over in the back of our minds. Because fast thinking uses associations rather than logic, it can make intuitive leaps more easily by finding creative analogies. It can take the experience of a new situation, let it soak in for a while, and then by association produce an array of analogous actions. In contrast, our attentive, slow-thinking mode provides insight into our actions, helping us detect problems and work through alternate plans of action.
“The social physics view of organizations focuses on patterns of interaction acting as a kind of “idea machine” to carry out the necessary tasks of idea discovery, integration, and decision making.” This view contrasts with person-centric views of the organization. “…Interaction patterns within [organizations] typically account for almost half of all the performance variation between high- and low-performing groups.”
Pentland suggests that more people use his methods to visualize and manipulate interaction patterns to shape idea flow and improve productivity and creative output. Again, it is the promise of the method that is the unique stuff here- the results line up with plenty of social science canon. I’m framing the results as Pentland does.
Echoing some of the interesting stuff out of Delanda’s Deleuze, Pentland argues that economics, social science, and much of our study of the world is “dominated by the equilibrium state”, where everything is in balance- and in trying to model that ideal we may actually miss out on the richness of the actual, messy reality. Before recently, data about our social world was totally improverished. “With the coming of digital media and other big data, all this has changed. We can now watch human organizations evolve on a microsecond-by-microsecond basis and examine all of the interactions among millions of people. When we observe fine-grain patterns of interaction within an organization, we find mathematical regularities that allow us to reliably tailor the organization’s performance and predict how it will react to new circumstances.
The Red Balloon Challenge was won by MIT by again using a social network incentive structure: there were ten balloons all over the USA to find in the quickest time. The prize was $40k. $2k was promised to the first person to correctly identify a given balloon, but $1k went to the person who invited that balloon finder onto the team- $500 went to their inviter, and $250 goes to their inviter. (The remainder money went to charity). All of the balloons were found in under nine hours.
In contrast with old-fashioned market thinking about crowdsourcing (eg. Amazon Mechanical Turk), Pentland argues that “the point isn’t that it’s possible to get lots of people to work [but] to get people to build an organization that does the work.” Networking people was as important as the searching work. “We used a standard individual economic reward to motivate people to report balloons but a social network reward to get them to recruit more people.”
Pentland briefly discusses more life-examples distinguishing the social network reward from the common market-oriented reward. He discusses Wikipedia and it’s social network incentives, he uses a study to demonstrate intensified engagement as organizations change and force rapid new habit-generations and social ties. In the next seciton of the book he intends to describe another view of the data-driven city. More on that next time.