Ethical Considerations of Artificial Intelligence in the Mining Industry

Australian Resources & Investment Journal, Volume 15, Number 1, 2021, PP: 14-17

Abhishek Kaul, Data Insights and AI Practice Leader, IBM

Ali Soofastaei, AI Program Leader, Vale

Abstract

There has been increased attention on the possible impact of future robotics and Artificial intelligence (AI) systems in recent years. Prominent thinkers have publicly warned about the risk of a dystopian future when the complexity of these systems progresses further. However, these warnings stand in contrast to the current state-of-the-art of robotics and AI technology.

Digital transformation and applied automation are growing fast in the mining industry. Therefore, it is essential to adapt to the mining industry with the related innovations which play critical roles in the digital revolution.?The core of these innovations is applied machine learning (ML) and AI across the mining value chain.

Many of us would assume that the mining industry would have driven advances in robotics, automation, AI, and ML due to the remote mine sites, the hazardous nature of the work, and the high costs of labor and transport. However, the manufacturing sector has spearheaded most of the technological developments, but it is now the mining sector that is now taking advantage of those proven technologies to help boost its recovery after a significant downturn.

In today's highly efficient mining operations, making the right decisions depends on their 360-degree visibility of the business and the market, combined with accurate demand forecasting.

With huge footprints in remote locations, diverse labor forces, and complex and time-consuming projects, mining companies use enterprise resource planning (ERP) and advanced analytics systems as the technology backbone.

ML and AI are the main part of an advanced data analytics approach, which is increasingly being relied on to make decisions about people, processes, and technologies, accessing worker productivity, exploring the next mine site, or predicting to schedule equipment for maintenance. Although AI and ML-based analytics deliver results, their recommendations for people-based decisions are subject to ethical considerations. For example, issues arise if AI and ML models are biased based on gender, age, or ethnicity and do not provide recommendations.

There are multiple AI policy guidelines available from the US, Europe, and Asia to help organizations build and use ethical AI. ?This article discusses AI use cases in the mining industry with ethical considerations, reviews critical challenges and potential bias mitigation strategies.

Understanding Ethics in AI

AI is intelligence demonstrated by machines, unlike the natural intelligence displayed by humans. AI refers to systems that display intelligent behavior by analyzing and interpreting the data, learning patterns in data, providing reasoning and recommendations, and optionally taking actions with some degree of autonomy to achieve trained goals.

AI systems work very well in use cases where they involve recognizing patterns with large quantities of data. However, AI systems work best together with people, and it is important to understand that AI requires reskilling people, not replacing them.

There are many AI techniques like Supervised Learning, Unsupervised Learning, Reinforcement Learning, Transfer Learning, Knowledge graphs, Reasoning systems, and more. Many of these techniques depend on ML. For example, the ability to automatically learn from historic patterns in data and improve performance over time. The difference between AI and ML can be a little confusing. Figure 1 illustrates the general boundaries between these concepts.

Figure 1: AI and ML boundaries

AI in the Mining Industry

Mining is a complex and fluctuating industry fraught with uncertainty around resource pricing, unpredictable resource fields, and major projects that need to be managed right through their lifecycle.

Controlling costs for mineral exploration, construction, and operation right through to project completion is a monumental challenge, but if the financial elements are managed well, it can help mining companies to be both competitive and profitable.

The key to increasing profits is knowing the precise time to increase production when there is strong demand using resource planning, improving the reliability of machinery with predictive and condition-based maintenance monitoring, delivering clarity with precise financial and operational reporting, and at the same time, providing actionable insights using real-time data extracted from every part of the organization.

There are considerable benefits of using an AI system to improve the quality of work at the mine site and reduce the human failures and hazards at the mine site. AI use cases have been applied across the complete mining industry value chain from exploration, mine management, extraction, processing, and transportation.

Figure 2: Mining Value Chain

Data is fed into the AI systems. The data come from equipment, shift log, operator manuals, operator wearable, CCTV cameras, HR systems, shift rosters, and more.?

Although ML and AI, by their very nature, are always a form of statistical discrimination, the discrimination becomes objectionable when it places certain privileged groups at the systematic advantage and certain unprivileged groups at a systematic disadvantage.

Objectionable discriminations arise due to multiple reasons in the mining industry. The main reasons are

• defining the business objective of the machine learning problem – for example, if the business objective is defined as maximum throughput without consideration for maintenance or safety aspects;
• unrepresentative data or data with existing prejudice for training – for example, if the AI model training data has been selected from a mine site where the demographic of the population is from the older age group; and
• selecting the attributes or features for the ML model - for example, if operator ethnicity has been included as a data point mobile mine equipment operator behavior in building the AI model.

This article focuses on three main use cases of AI in mining.

• Energy- Ethics in reducing fuel consumption
• Maintenance- Ethics in Predictive maintenance
• Safety- Ethics in using surveillance Video (CCTV) for safety

Details of the use cases, their ethical considerations, and potential bias mitigation strategies will be discussed in subsequent sections. However, first, AI policy guidelines are presented for developing Ethical and Trustworthy AI.

Policy Guidelines for Ethical AI

Many countries have published AI policy guidelines. These guidelines provide a broad objective for the use of AI – to ensure human-centric, safe, and trustworthy AI. Most guidelines make the organization using AI responsible and accountable for their decisions and ask for the same ethical standards in AI-driven decisions as in human-driven decisions.

General Key points achieved from global guidelines are:

• it should be lawful, complying with all applicable laws and regulations;
• it should be ethical, ensuring adherence to ethical principles and values; and
• it should be robust, both from a technical and social perspective, since, even with good intentions, AI systems can cause unintentional harm.

To ensure compliance with Ethical guidelines, AI models need three capabilities. ?

1. Explainability – An ability to explain the behavior of the black box AI model. Multiple algorithms help to explain the model. For example, decision tree rules if A > 50, then stop else continue, are easily understood by people.
2. Fairness – The ability of AI models to report and mitigate discrimination and bias. Depending on the application of the AI model, the appropriate bias metrics should be reported. For example, for hiring, a false-positive (someone unfit for the job is employed) is less harmful than a false negative (someone fit for the job is denied). Further, Bias mitigation algorithms can be applied to improve the fairness metrics by modifying the training data, the learning algorithm, the predictions, the optimization, or the making decision models.
3. Transparency - The ability of the model to be transparent on training data, accuracy and performance, bias, and fairness metrics so that users can understand how AI was trained and deployed.

?AI use cases with Ethical consideration for the mining industry

In this section, three use case details are presented with ethical considerations for AI in the areas of Energy, Maintenance, and Safety. It is important to understand that AI is not about replacing people but reskilling people and the deployment of AI applications will improve the quality of work in mine sites and reduce the human failures, hazards in the mine site.

Use Case 1: AI Application for Energy efficiency - Ethics in reducing fuel consumption [consideration - operator demographics]

Fuel is an important cost contributor for haul trucks in surface mining. Multiple parameters affect fuel consumption like the type of Truck, Payload, Distance, Hours, Weather, and Operator behavior (includes speed, maneuvering, acceleration, and braking). AI techniques like Artificial Neural Networks (ANNs) are generally applied to data to understand the top factors influencing fuel consumption and recommend changes to controllable factors, thereby reducing fuel consumption per ton of ore mined using Genetic Algorithm (GA) in the optimization phase of the project.

Figure 3: Advanced Data Analytics applied to minimize haul truck fuel consumption

One of the predictors for haul truck fuel consumption is the operator (driver) behavior. If demographic data points of drivers are included as attributes or features, then the AI model will identify patterns in demographic data that influence operator behavior. Depending on the training data set, for example, country, mine site, number of operators, this analysis may have bias and may not hold for the general case. For example, maybe the model can get biased to predict low fuel consumption for older male workers based on one mine site operation.

It is recommended not to include demographic data in the analysis in such applications and rely solely on the unique mine equipment operator identifier. Unless essential, if demographic data is needed, de-biasing techniques like reweighing and adversarial de-biasing should be applied with visibility on fairness metrics like statistical parity difference, Thiel index, and more should be enforced. Ideally, the business objective of the AI model should be finetuned to provide guidelines to an operator to influence their behavior – like speed, acceleration, etc. and provide a mechanism to monitor for deviations in operator actions to AI recommendations.

Use case 2: AI Application in Maintenance - Ethics in Predictive maintenance [consideration - operator shift logs / NLP]

Equipment downtime significantly affects the productivity and safety of mining operations. The main goal of predictive maintenance is to shift the unplanned breakdowns to planned maintenance activities, increase the equipment lifetime, optimize maintenance schedules, and ensure safe operations. AI techniques in machine learning like Cox regression, Logistic regression, Gradient boosting, Neural nets are applied to predict the remaining useful life (RUL) and predict the equipment's health score using historian, maintenance data. Further, Natural Language Processing (NLP) techniques like Word2Vec, BERT are applied on operator shift logs to gain a deeper understanding of operational events like faults, trips, overriding, noise, resetting observed, actioned, and documented by the operator, which are then co-related to maintenance failures to provide deeper insights.

Figure 4: Advanced Data Analytics applied to predict mine equipment failure using equipment data and operator logs

One of the data points used in the analysis to discuss further is the operator shift logs. Operator privacy is one of the considerations for analyzing logs, however beyond privacy, when analyzing operator logs, if language linguistics analysis is used for deciphering personal traits, personal attributes, like modeling, and then to co-relate maintenance failures; inclusion of bias becomes a relevant topic, and it is subject to ethical considerations.

In such applications, it is recommended to use the co-relation of events (nouns, verbs) with maintenance failures for providing slack time in operation for operators to do necessary maintenance inspections. Further, it is recommended to de-bias the NLP models, which may cause some drop-in accuracy points but helps to keep the recommendations fair.

Use case 3: ?AI Application in safety - Ethics in using surveillance video (CCTV) for safety [consideration – surveillance video data]

Video surveillance data (CCTV cameras) are used in many work areas to ensure the security and safety of the site. Typically, hundreds of cameras feed data to the site security/ safety office. Since it is impossible to review the feed from all cameras in real-time by the human operator, the use case for video surveillance tends towards post-facto video retrieval and analysis for historical incidents and disputes. With advancements in Computer vision technologies, AI models are trained with the surveillance video feed to perform automated analysis for object detection, object classification, object tracking and raise proactive alerts in real-time to detect and mitigate any safety or security violations.?

Figure 5: Advanced Data Analytics applied to detect non-compliance behavior using CCTV footage

Surveillance is itself an ethically neutral concept. What determines the ethical nature of a particular instance of surveillance will be the considerations which follow, such as justified cause, the means employed, and questions of proportionality. While it can be argued that monitoring remotely via a camera is no different from historical times when security personnel were physically present at the worksite, there are local privacy laws and regulations which must be complied with while using surveillance, face recognition technologies.

AI technologies in computer vision are leapfrogging every few months, like understanding the Spatio-temporal relationships of objects, which can be used to monitor people's behavior based on the change of posture, time of day, relationship with equipment, movement between areas, and more. Ethical considerations when employing AI for surveillance monitoring should balance workers' Privacy-Trust-Autonomy and workers' Safety-Security-Behavior.

Conclusion

Understanding the implications of ethics in AI is important for mining companies to remain fair to their workforce in human-driven decisions.

Mining companies should adopt and build AI solutions that follow leading policy guidelines and are explainable, transparent, and fair. Further, when evaluating the AI solution, they should understand how it built the data used to train the AI models.

In order to enforce Ethical considerations for AI, mining companies can appoint an AI ethics officer or committee to review each AI application being developed or purchased from vendors and requests for disclosure on the fairness, explainability, and transparency metrics.