As discussed in Part 1 of this series, manufacturers can gain a breakthrough competitive advantage from AI through anomaly detection, predictive maintenance, and automated asset management. But the power of AI extends beyond those use cases, supporting a whole new dimension of automation and insight, shares Lori Witzel, director of research for analytics and data management at TIBCO.
Artificial Intelligence (AI) is in the news, as are regulations to manage AI risks. AI regulatory compliance will impact manufacturers sooner rather than later.
Through AI and related technologies, manufacturers can gain a complete, 360-degree, data-integrated view of all operations – from suppliers and supply chains, through equipment, processes and manufacturing practices, to end-product testing and customer satisfaction. It’s the promise of Industry 4.0 realized, and it’s widening the gap between leaders and laggards.
The benefits of AI are no longer without risk, however. Increased adoption of AI across many verticals, including manufacturing, is driving increased technology regulation. U.S. manufacturers must act now to prepare for the changing regulatory landscape.
Trusted AI Is a Best Practice
Building trust and transparency in AI is a core best practice. It is also imperative to ensure compliance with current and future regulations.
Trusted AI is audible, transparent and (at the risk of vastly oversimplifying a complex subject) interpretable. Interpretable AI consists of algorithms that provide a clear explanation of their decision-making processes. This interpretability ensures that humans can evaluate an AI-infused process, so they can apply their own insights and opinions to the logic behind an AI-made decision.
For example, an experienced operations manager may need to understand why certain products coming through production were identified as defective and not others. If AI determines that a product in an image is damaged, this presents a possible interpretable use case – the need for a human to be able to validate the decision. The AI becomes interpretable when the defect location is highlighted visually, so a person can see and verify which of the many visible features in the image represent the defect. It is not interpretable if the AI only indicates that the image contains a defect but does not highlight the actual defect within the image.
Another example of risk specific to manufacturing, noted by McKinsey, is the potential for accidents and injuries due to an AI-mediated interface between humans and machines. If AI-infused systems fail to keep a human in the loop—failing to interpret best practice—equipment operators may not be able to provide needed oversight, increasing physical risk in applications using self-driving vehicles. Other risks for manufacturers, such as faulty AI degrading a supplier’s product, are also consequential.
Interpretable, transparent AI will enable data science teams to respond in ways that even a less technical workforce can understand. This is particularly useful for legacy manufacturing operations, which often find themselves under pressure from digital-first competitors.
See More: A Quick Guide to Smart Manufacturing
Trusted AI is Based on Trusted Data
An example of the value of reliable data for manufacturing is Arkema, a €8B French specialty chemical and advanced materials company. They produce technical polymers, additives, resins and adhesives. The flow of data across customer, vendor and material domains across the business has been revolutionized by their data fabric-like approach to data assets. Jean-Marc Viallatte, group vice president of Global Supply Chain at Arkema, led an enterprise-wide initiative layering a common data framework into an ever-expanding list of products, ensuring that every system deployed draws from the trusted master data hub.
The Arkema team now widely shares standardized, reliable data across the organization, enabling enhanced regulatory compliance, facilitating further growth through smoother integration of data related to M&A activity and supporting flawless customer-centric service. Arkema is an example that US manufacturers can learn from as they seek advantage by using AI for supply chain optimization, anomaly detection, root cause analysis, key factor identification, performance optimization through pattern recognition at scale, and predictive and prescriptive maintenance through advanced equipment monitoring. .
How to Prepare for a Changing AI Regulatory Landscape
As noted by McKinsey, manufacturers that use AI significantly outperform their laggard peers. The examples they cite yield loss reductions of 20 to 40 percent while improving on-time delivery through an AI software agent. But without preparing for AI transparency and auditability, those benefits could be lost to regulatory risk. Although AI regulation is still state by state, in many cases, and is in the draft stage around the world, preparing to meet compliance could include:
1. Data fabric architecture with robust master data management (MDM) for holistic management of the data pipelines that feed manufacturing automation: Regulatory compliance means understanding not only the algorithms used, but the data that was used to train AI and machine learning (ML) models. Data fabric provides the framework to achieve transparency as well as better results.
-
- AI training data discovery and management: Your data science teams may not only use data from the organization, including IoT data – they may also use publicly available data sets. Whether internal or external in origin, the lineage of the data, and the observability and transparency of its use, are key components for regulatory compliance.
- Disclosure and management of Personally Identifiable Information (PII): To ensure AI regulatory compliance, the organization must understand if there is PII in any AI system used by the organization. A robust MDM can help identify what PII data is in which systems and how that PII is masked or otherwise protected.
2. Data virtualization to help scale and reduce friction in AI training data preparation: The enormous amount of training data required by ML and AI systems requires agile, scalable data preparation processes. Data virtualization can reduce friction in data preparation by reducing the impact of data silos on scalability and access.
3. Baseline and ongoing algorithm revisions: Identifying and documenting algorithms used across manufacturing automation and supply chain processes is an important step towards the transparency required for regulatory compliance.
-
- Algorithmic transparency and explainability: An integrated platform approach to data analytics and data science will make it easier to identify and document algorithms used. It will also help ensure that these algorithms are transparent and explainable – key aspects of AI compliance.
- Business partner and vendor algorithm documentation: Manufacturers should also ask business partners and technology vendors for documentation of any algorithms that may be used by the manufacturer’s own systems and processes. Boston Consulting Groupamong others, it is recommended to implement a responsible AI framework that includes vendor management because a manufacturer can be held responsible for non-compliant AI provided by a business partner or vendor.
Just as the benefits of AI for manufacturers cross silos and extend across the organization and its business partners, so too should preparations be made for the regulation of these technologies. AI can be pivotal in enabling manufacturers to leap ahead of the competition. As you prepare to make that leap, make sure you have the governed, transparent AI processes in place – along with various stakeholders – to be able to adapt to a changing regulatory landscape.
What AI compliance strategies are you implementing to adapt to the evolving regulatory landscape? Share with us on Facebook, Twitterand LinkedIn.
Image source: Shutterstock
