Article written by Bilal Abdallah and Brian Kenneth Swain
Risk is the effect of uncertainty on objectives. It takes many forms and spans a wide range of severities, from the trivial (traffic might make you late to the party) to the profound (an errant political decision could start a war). There has been risk—and attempts to manage it—since the dawn of civilization, and the tools available have evolved at more or less the same pace as the level and variety of the risks themselves. But one thing has not changed much over all those eons: the risk management process that starts with awareness, whether that’s awareness of predators lurking in nearby woods or a changing business climate that might make your product less desirable.
The latest generation of tools for managing risk is built using artificial intelligence (AI) capabilities that are uniquely well-equipped to identify and mitigate business risk. AI-enabled tools conduct analyses and generate action recommendations with speed and efficiency that humans cannot match and they do so objectively while continuously learning and improving their performance.
The management of risk is a constant challenge, regardless of what business your company is in. And the better you understand the various tools available to address this challenge, the less exposed your organization will be, whether the risks you face are internal and controllable or external and uncontrollable. The AI-based cycle of Observe, Reason, Act, Learn, and Adapt natively matches and supports the established risk management process comprising risk identification, assessment, control, and continuous improvement.
Observe—What is happening?
Effective risk management begins by observing the business domain and monitoring a vast array of performance indicators, some critical to system/process operation, others less directly related but still potentially capable of affecting the efficiency or safety of the process. One of AI’s core strengths is its ability to effectively digest and analyze immense quantities of performance data, automatically seeking out relationships that would not be apparent to humans conducting manual analysis. Examples of the sorts of data that can be evaluated by AI models include (but are by no means limited to):
- Structured data such as weather forecasts, market forecasts, financial indices, process parameters, lab results, analyzer data, etc. Machine learning-based technologies like SparkCognition ML Studio enable companies to ingest and process immense quantities of equipment and process data in real time.
- Textual data like news, experts’ speculations, material safety data sheets, exit interviews, incident reports, near misses, safety observation cards, work order closing comments, operating procedures, work permit applications, medical reports, etc. Natural Language Processing (NLP) technologies like SparkCognition DeepNLP can read handwriting and printed text in many languages, gleaning important insights and providing advance warning of potential issues or threats.
- Visual data from images and camera feeds can now be analyzed in real-time to identify unsafe activities, behaviors, and man/machine interactions using advanced contextual scene understanding and knowledge of the relationships between people, objects, and the environment.
Reason—What should we do about it?
Presented with detailed information about potentially risky situations, AI-enabled algorithms connect the dots to identify linkages between actions and situations whose risk outcomes may not be apparent at first glance. For example, having gleaned from written technical documents (using NLP) that the presence of a certain chemical substance on or near another material can cause a hazardous situation, the AI algorithm can deduce that a component of the process includes the likelihood of such an interaction occurring, and can thus proactively raise a flag about the impending risky situation long before it ever occurs in real life.
Following this analysis, we must next decide which mitigative actions will have the greatest impact.
- Correlating observed situations with accident/failure probability
- Performing computations necessary to support decision making
- Evaluating likely severity of potential accidents/failures
- Detecting deviations and anomalies from prior risky occurrences
- Understanding which measures were more/less effective in past situations
Act—Take steps to address the risk
Having proactively identified potentially risky situations and identified a range of potential responses, the AI algorithm then seamlessly selects the specific actions most likely to effectively mitigate the situation based on past experiences and its comprehensive library of potential actions, either automated or manually performed.
These responsive actions can take many forms—some immediate, some longer term.
- Automatically operate fire extinguishers and sprinklers when a camera detects a flame, smoke, or electrical spark/arc—well before heat and smoke sensors detect it.
- Notify managers/workers of safety hazards/non-compliances when a camera detects:
- Trip hazards from objects left in walkways
- People/objects under suspended loads
- People/objects in the vicinity or in pathways of moving vehicles/forklifts
- Non-compliance with personal protective equipment (PPE) requirements
- Carrying loads incorrectly
- Worker-down events (injuries, unconsciousness, etc.)
- Missing fire extinguishers
- Too many people in confined areas
- Unauthorized people in restricted areas
- Objects blocking emergency exits
- Waste/garbage, left-behind objects
- People not washing hands after leaving toilets
- Use of cell phones or smoking in restricted areas
- Absence of people required at a site, such as safety officers or site supervisors
- Recommend safety measures when processing work permit applications
- Recommend topics for future safety campaigns
- Update asset registries to document newly discovered hazards, e.g., to note that mixing two new incompatible chemicals can cause fires or explosions
- Recommend auto-classification of incidents based on corporate governance and incident description, e.g., if an ambulance is mentioned in an incident report, a valid medical report must be attached within a certain time period
- Identify composite risks and potential Swiss cheese model incidents by correlating minor risks related to the same location but in various layers of protection. For example, AI can provide warning of a major risk if cameras/observations detect:
- Lamp cover removed because of dust and resultant insufficient lighting, providing a potential source of ignition
- Valve left open because of the frequent need to drain, causing a single point of failure and leak risk
- Alarm wire disconnected because of looseness that caused numerous false alarms, disabling its function when a true alarm is required
Learn—Did our actions help the situation?
AI learns about the effectiveness of its recommendations in various ways:
- Direct end user feedback/training: Users manually update the classification of incidents, explicitly labeling AI-recommended results by the degree of effectiveness and timeliness of outcome.
- Automated outcome monitoring: AI monitors the implementation of its recommendations automatically (or implementation of different recommendations per user discretion) and the outcomes of such actions, scoring each action and its outcomes.
Adapt—What should we do differently next time?
AI adapts its recommendations to improve the outcomes from future incidents/threats and recommendations. Subsequent risk situations/threats thus benefit from knowing the outcomes of all prior analogous occurrences, progressively increasing the effectiveness and timelines of future recommendations.
Modern risks demand modern response tools
AI-enabled tools offer unprecedented capabilities for the effective management of industrial and business risk. Such tools perform analyses and deliver recommendations at scale and speed unachievable by humans while also providing objectivity in reasoning and action and the ability to continuously learn and adapt.
SparkCognition’s products address risk management in numerous useful ways. Visual AI Advisor (VAIA) provides advance warning of unsafe conditions, risky personal work practices, and potentially dangerous interactions between people and machines. Oil and Gas Maintenance Advisor uses the power of machine learning (ML) and normal behavior modeling (NBM) to provide advance warning of impending equipment failure or performance degradation. Only through advance knowledge of potential problems can risk be managed effectively. Artificial intelligence provides the toolkit for achieving the requisite level of awareness to act effectively and adapt future behaviors to ensure optimal business performance.