Resilience is a Verb

David Woods presentation discusses how resilience engineering studies in real settings have revealed patterns in adaptation in complex networks and discovered surprising new fundamentals about how all adaptive systems work — fundamentals that overturn assumptions made by many disciplinary areas of study.

Adaptive Universe, David Woods, May 2019

graceful extensibility is a positive capability to stretch near and beyond boundaries when surprise occurs

opposite of brittleness

graceful extensibility trades off with optimality

Biological universe (not physics)

"life is a verb" and "life developed by networking not combat." - Lynn Margulis

no role/title is exempt from the rules of the adaptive universe

exemplified by multiple levels: cells (glycolosis); healthy heart; sick heart, neurobiology, joint cognitive systems, socio-technical systems, organizations, social systems

all lines of inquiry contribute but are partial, incomplete, and off axis to the new emergent synthesis

special character of control laws, especially for layered networks. (you can violate control laws, but you cannot escape the consequences of those laws. e.g. high center-of-gravity car)

Viability requires Extensibility

Studying adaptation and surprise in risky high performance settings.

Adaptive Cycles

People searching for advantage on their goals, brittleness/complexities appear, people adapt to fill gaps and workaround bottlenecks.

stories of change reveal the congestion, cascades & conflicts that arise when apparent benefits get hijacked

We have to get good at two kinds of adaptive capacity. The ability to improve our confidence envelope, to get better with respect to frequently encountered events and variations. At the same time the ability to recognize and respond to novelty and surprise at the boundaries of our confidence envelope.

As capabilities increase, extensive and hidden inter-dependencies arise

cartoon

animation

High Frequency Trading table comparing events in the cycles of adaptations on the left, and annotations on the right of what kinds of adaptive phenomena they exhibit.

If you find something where you can specify all the nodes and links as if they were definitive effects, you have already mis-modeled a tangled layered network. You have already misrepresented, in a fundamental way, the adaptive universe.

SNAFUcatching is a euphemism for graceful extensibility.

How do adaptive units and tangled layered networks outmaneuver complexity since there's no place to hide?

What are we explaining? (brittleness – sudden failures against a background of improvement and how this doesn't happen more)

What's new? (graceful extensibility – remember viability requires extensibility, net adaptive value, a general parameter capacity for maneuver or risk of saturation,...)

What are we predicting? (how unintended consequences emerge from units locally adapting)

What do we need? (architectural principles and general mechanisms... how do we balance and not trade-off optimality with graceful extensibility? How can we have those tensions work together to create a better net adaptive value)

review of Graceful Extensiblity

Adaptive Capacity is about the future. Before things go wrong, you have to already be poised to adapt. You have to have the potential to modify plans to benefit the situations actually encountered. The potential for adaptation must exist before changes and disruptions call upon those capacities.

Biology does not skip graceful extensibility.

We engineers fall under the illusion that this generation of new technology is better than the last and so we don't have to build in graceful extensibility.

You have to have some and you cannot have enough. Those are fundamental constraints on any adaptive unit.

So the key thing in a network architecture is to set it up so neighbors will help.

Short on time... skipping over many slides... stops to emphasize:

Organizations rationalize under-investment in graceful extensibility on three grounds (all three of which are wrong empirically, technically, & theoretically):

Rarity: orgs wrongly believe SNAFUs occur rarely given the organization's design, so investing in SNAFU Catching is a narrow issue of low priority

Prevention: orgs wrongly believe there is a record of improvement that reduces the likelihood, severity, or difficulty of SNAFUs

Compliance: orgs wrongly believe when SNAFUs do occur, the poor response is due to people who fail to work the rules for their role within the organizational design. (e.g. wrongly believe the design of the system is sound – it isn't – and it was humans who failed – they didn't)

We need to build a humble but different set of engineering capabilities.

We have plenty of pragmatic examples. The problem is the examples don't fit the world view. They're networked, not linear. People asking for pragmatic examples want us to give them something that still fits with their linear fundamental organization.

How do you build and sustain reciprocity? Reciprocity is necessary to align performance, align behavior relative to the goals and goal conflicts.

He skips passed a slide at 46m15s

"TGE reframes optimality. The pursuit of optimization is a form of pressure on units that arises from other units/levels in the network. The need to respond to changing pressures, given conflicting goals is omnipresent in networks of adaptive units. Computations, however justified, miss the reality of being caught in a squeeze between conflicting goals as pressures ramp up. The assumption is the right computations use a policy that provides a best solution to the conflict and pressures – the end of the story. But adaptive cycles are always stories about how conflicts stimulate adaptations. As pressures ramp up, squeezes intensify; one critical question is what goals get sacrificed first and which last. This process of re-prioritization gets lost in the pursuit of more optimal computations."

"Optimality" is often phrased "better, faster, cheaper". One part of the system demands better, faster, cheaper. Another part of the system demands code freeze (or similar). Under the extra pressures, what goals get sacrificed first?