9.2 Modern Methods
135
Big data relies on machine learning, isolates variables to identify patterns, reveals
insight. Big data gains insight from scale of data points, but loses resolution details.
It does not tell you why those patterns exist.
Thick data relies on human learning, accepts irreducible complexity, reveals
social context of connections between data. Thick data gains insight from anecdotal,
small sample stories, but loses scales. It tells you why, but misses identifying complex
patterns.
Focusing solely on Big data can reduce the ability to imagine how the world
might be evolving. Big data only is not sufficient for risk assessment, and in particular hazard identification (http://www.riskope.com/2017/04/05/hazard-identificationscience-art/). It can create a distorted view of the risk landscape surrounding an entity.
Known Knowns and Unknown Knowns
If one is seeking a map of an unknown risk territory (risk landscape (http://www.riskope.com/2016/06/23/business-intelligence-platform-helpingtailings-stewardship/)) and data are scarce, then thick data is the tool of choice. As
data availability grows, on its way to becoming “big data”, integrating both types of
data becomes important. In the case of innovative companies that combined insight
can be highly inspirational.
When performing risk assessments of dams, mining operations, etc. we always
collect and analyze “stories”, anecdotes, loss reports to gain insights of “pre-existing”
states of the system. The combined insight can tell us that a dam that “looks wonderful” actually has a “congenital defect” that raises the probability of failure.
Big data would not be capable of showing that, but could probably reveal a pattern
between third-party observations and meteorology. In fact it could reveal a patterns
between any other groups of variables, which could sound an alarm on “shorter-term”
emergent hazards.
Integrating Big Data or Thick Data
Working successfully with integrated big and thick data certainly enhances any risk
assessment. Over the years we have found ways to integrate data from multiple
sources and of various natures in our risk assessments. We routinely use incomplete thick data sets in conjunction with expert opinions and literature to generate first a prior estimates of the probability of occurrence of hazards and failures. This immensely increases the value of the first-cut risk assessment, which can
then be updated using big data and Bayesian techniques (http://www.riskope.com/
knowledge-centre/tool-box/glossary/) (see Sect. 10.1.2).
The combined approach also makes it possible to enhance the value of big data,
avoid capital squandering, and the reduce the running cost necessary to obtain
big data. Recent studies have shown that without that approach data oftentimes
remain virtually “unused” (https://www.inc.com/jeff-barrett/misusing-data-couldbe-costing-your-business-heres-how.html).
135
Big data relies on machine learning, isolates variables to identify patterns, reveals
insight. Big data gains insight from scale of data points, but loses resolution details.
It does not tell you why those patterns exist.
Thick data relies on human learning, accepts irreducible complexity, reveals
social context of connections between data. Thick data gains insight from anecdotal,
small sample stories, but loses scales. It tells you why, but misses identifying complex
patterns.
Focusing solely on Big data can reduce the ability to imagine how the world
might be evolving. Big data only is not sufficient for risk assessment, and in particular hazard identification (http://www.riskope.com/2017/04/05/hazard-identificationscience-art/). It can create a distorted view of the risk landscape surrounding an entity.
Known Knowns and Unknown Knowns
If one is seeking a map of an unknown risk territory (risk landscape (http://www.riskope.com/2016/06/23/business-intelligence-platform-helpingtailings-stewardship/)) and data are scarce, then thick data is the tool of choice. As
data availability grows, on its way to becoming “big data”, integrating both types of
data becomes important. In the case of innovative companies that combined insight
can be highly inspirational.
When performing risk assessments of dams, mining operations, etc. we always
collect and analyze “stories”, anecdotes, loss reports to gain insights of “pre-existing”
states of the system. The combined insight can tell us that a dam that “looks wonderful” actually has a “congenital defect” that raises the probability of failure.
Big data would not be capable of showing that, but could probably reveal a pattern
between third-party observations and meteorology. In fact it could reveal a patterns
between any other groups of variables, which could sound an alarm on “shorter-term”
emergent hazards.
Integrating Big Data or Thick Data
Working successfully with integrated big and thick data certainly enhances any risk
assessment. Over the years we have found ways to integrate data from multiple
sources and of various natures in our risk assessments. We routinely use incomplete thick data sets in conjunction with expert opinions and literature to generate first a prior estimates of the probability of occurrence of hazards and failures. This immensely increases the value of the first-cut risk assessment, which can
then be updated using big data and Bayesian techniques (http://www.riskope.com/
knowledge-centre/tool-box/glossary/) (see Sect. 10.1.2).
The combined approach also makes it possible to enhance the value of big data,
avoid capital squandering, and the reduce the running cost necessary to obtain
big data. Recent studies have shown that without that approach data oftentimes
remain virtually “unused” (https://www.inc.com/jeff-barrett/misusing-data-couldbe-costing-your-business-heres-how.html).