Critical Failure Detection, Prediction and Risk Analysis

We did related work which included risk analyses on solutions of many types, including AI machine learning and classical analytics.   Even those that could be considered less than 'critical'.  Typically using elements of predictive analyses.   Prediction could also be used to produce test sets for failure and recovery analyses.   Simulation also essential.  So the approach here is quite interesting. Can see the methodologies being intertwined.

AI researchers devise failure detection method for safety-critical machine learning
Researchers from MIT, Stanford University, and the University of Pennsylvania have devised a method for predicting failure rates of safety-critical machine learning systems and efficiently determining their rate of occurrence. Safety-critical machine learning systems make decisions for automated technology like self-driving cars, robotic surgery, pacemakers, and autonomous flight systems for helicopters and planes. Unlike AI that helps you write an email or recommends a song, safety-critical system failures can result in serious injury or death. Problems with such machine learning systems can also cause financially costly events like SpaceX missing its landing pad.

Researchers say their neural bridge sampling method gives regulators, academics, and industry experts a common reference for discussing the risks associated with deploying complex machine learning systems in safety-critical environments. In a paper titled “Neural Bridge Sampling for Evaluating Safety-Critical Autonomous Systems,” recently published on arXiv,  https://arxiv.org/abs/2008.10581  the authors assert their approach can satisfy both the public’s right to know that a system has been rigorously tested and an organization’s desire to treat AI models like trade secrets. In fact, some AI startups and Big Tech companies refuse to grant access to raw models for testing and verification out of fear that such inspections could reveal proprietary information ...."

Irving Wladawsky-Berger: Why Some AI Efforts Fail

From a former IBMer what we worked with on enterprise AI the first time around.   Also insightful for many kinds of emerging tech.  I have now followed up on this question for several major AI projects.   Reading the report mentioned below now.  More to follow.

Irving Wladawsky-Berger
A collection of observations, news and resources on the changing nature of innovation, technology, leadership, and other subjects.

Why Some AI Efforts Succeed While Many Fail
Winning with AI, - a 2019 report based on a survey jointly conducted by the MIT Sloan Management Review and the Boston Consulting Group, - found that 90% of respondents agree that AI represents a business opportunity for their company.  The global survey attracted over 2,500 respondents from 29 industries and 97 countries, and conducted interviews with 17 executives leaders of AI initiatives in large organizations.

The report classified the total survey population into four subgroups based on their understanding of AI tools and concepts and their levels of adoption of AI applications: Pioneers (20%) are leading-edge organizations that both understand and have widely adopted AI; Investigators (30%) understand AI but have not deployed applications beyond the pilot stage; Experimenters (18%) are learning by doing, conducting pilots without a deep understanding of AI; and Passives (32%) have not adopted AI and have little understanding of the technology.

“Many AI initiatives fail,” was the report’s overriding finding.  “Seven out of 10 companies surveyed report minimal or no impact from AI so far.  Among the 90% of companies that have made at least some investment in AI, fewer than 2 out of 5 report obtaining any business gains from AI in the past three years. This number improves to 3 out of 5 when we include companies that have made significant investments in AI.  Even so, this means 40% of organizations making significant investments do not report business gains from AI.”

Why is it so hard to realize value from AI?  Why do some efforts succeed while many more fail?  To help answer these questions, the study looked for patterns in the survey data and in the executive interviews to uncover what the companies that are succeeding with AI are doing.  It found that the companies generating the most value from AI exhibit a distinct set of organizational behaviors.  Let me summarize these findings.  .... " 

Coke Rewards Failure

But not failure to learn.  We did not do enough to understand and document precisely why we failed.

Coke CEO: Why we have an award for projects that fail   By Catherine Clifford

Coca-Cola was invented in 1886 by Dr. John S. Pemberton in Atlanta, Georgia. That year, sales of the drink totaled $50. In fiscal year 2018, The Coca-Cola Company reported $31.9 billion in sales.

Today, the company has more than 700,000 employees and over 500 beverage brands — from Fanta and Minute Maid to Honest Tea and Odwalla to SmartWater and Dasani — sold in 200 countries.

When a company becomes to be so massive, it can be hard to keep the creativity and innovation that helped it grow in the first place. And that’s largely because with so much success people begin to fear failure, according to Coca-Cola Co. CEO James Quincey.

“Fear of failure is often the biggest hurdle for innovation in large organizations,” Quincey said in a recent Harvard Business Review Analytic Services (HBRAS) report commissioned by Mastercard.  ... .... "

Data Mining of Failure

Reminds me of the AAR: After Action Review, which we did for a while.  But the important thing is to gather the data consistently, often an issue.

How the data mining of failure could teach us the secrets of success in Technology Review
These data researchers found that for startups, scientists, and terrorists alike, learning too little from experience spells doom.

by Emerging Technology from the arXiv 

Thomas Edison is often described as America’s greatest inventor. His successes include electric power generation, sound recording, and the electric lightbulb.

But Edison was no stranger to failure. He famously tested 1,000 different designs before settling on the carbon filament that became the first commercially successful lightbulb. This tenacity set him apart. “Many of life’s failures are people who did not realize how close they were to success when they gave up,” he said.

Many groups and individuals have studied the nature of success. These studies have yielded varying degrees of insight. The flip side—the nature of failure—is much less well studied but arguably more important. Little is known about the mechanisms that govern the dynamics of failure. ...."

AI Failures in the Past Year

I don't call these AI failures, because almost by definition they are not 'intelligent'.  But it does give some good examples about how things can fail,  just as people can.  Don't think we will ever rid ourselves of these.  Intelligent is not perfect. A worthwhile set of cautionary examples.

Artificial ignorance: The 10 biggest AI failures of 2017

From self-driving car accidents to Face ID hacks, artificial intelligence didn't have a flawless year.
By Olivia Krauth in TechRepublic ... "

Predicting Power Outages

Researchers develop model to predict and prevent power outages using big data
Texas A&M Engineering News   by Shraddha Sankhe

Researchers at Texas A&M University have devised a predictive risk analysis framework that can forecast a potential vulnerability to utility assets and map out the location and time of a possible outage. The researchers say this feature enables the trees in the most critical areas with the highest risk to be felled first. "Dealing with aging infrastructure assets adds another layer of complexity that utility companies face," says Texas A&M professor Mladen Kezunovic. "Any kind of environmental data that has some relevance to the power system can be fed into this prediction framework." The model's applications can be tailored by feeding it data such as a utility company's operational records, weather forecasts, altitude, and foliage around the power systems. The team says they use the goals of the power system to select a large volume of input data from multiple sources and conduct a risk analysis, adding to the reliability of the system and its operations. ... "

IOT Failure Prediction

Attended this webinar of interest, with a largely nontechnical approach, other parts will follow: " Industrial IoT and Failure Prediction on event signals | Part 1 " featuring Adurthi Ashwin Swarup, Senior Data Scientist, DataRPM.

Here's the replay link.

Celebrating Failures

From Knowledge@Wharton:    After action reviews?

The digital transformation of a company requires not a mere shuffling of the organizational chart, but rather a “chemical” change in the culture and business practices, says Ganesh Ayyar, CEO of Mphasis, a major IT services company. But it is easy to say and more difficult to do. One place to start is by encouraging experimentation through the celebration of failures, adds Wharton marketing professor Jerry (Yoram) Wind. Another is to learn to co-create with clients. As always, the CEO and other senior executives set the tone: The old command-and-control style of managing is becoming passe, replaced with a more collaborative model recognizing that good ideas can come from anywhere in the company. ... " 

Fail at Scale

In CACM: Reliability and the science of graceful failure.  Abstract, full article requires registration.