Manufacturers were among the first industries to embrace robotics. So it should come as no surprise that the sector is jumping aboard the artificial intelligence bandwagon. But is it moving too fast?
Along with the obvious benefits of AI comes a slew of uncertainties, including the potential impacts on a company’s security, reputation and regulatory stance. Says Mason Clutter, AI governance partner at the law firm of FBT Gibbons: “It’s really just a question of responsible development and employment — to strike the right balance.”
The rush to adopt AI isn’t unique to manufacturing, Clutter notes. The lure of enhancing efficiency in turbulent times is too strong for most businesses to resist. “I do think, though, that they need to question, and document the ‘why’ behind each deployment,” she says.
AI projects are often launched from the top of the corporate ladder, but those on the lower rungs aren’t always given sufficient guidance as to how and where top management intends to pursue them. Clutter says it’s vital to define specific use cases at the outset, and ask questions that will mitigate the potential risk of deploying a technology that even its creators are unsure about where it’s going.
Such questions relate to many of the fundamental concerns that drive a business: the safety of proprietary data, confidential information and intellectual property, not to mention a company’s state of regulatory compliance. All can be bolstered or undercut by increasingly sophisticated AI models.
By understanding the practical and ethical considerations of using AI, manufacturers can better evaluate the tools being offered by a dizzying array of vendors. Clutter says it’s important to know whether a given supplier has its own set of ethical standards about how the customer’s proprietary data will be used. Maybe it has promised not to use the underlying data to train its models. Such assurances, however, need to come at the very beginning of a vendor relationship, and can be helpful in filtering out the riskier alternatives.
Clutter recommends the formation of a cross-functional governance committee to review prospective AI models. It should consist of stakeholders from across the organization, including legal, IT, security operations, and marketing and sales. “Everyone needs to be on the same page before a single model even touches your data,” she says.
At the same time, there needs to be one person tasked with overseeing the effort. That individual’s title can vary according to the company’s organizational structure. At the Department of Homeland Security, where Clutter previously served as chief privacy officer, the job was handled by a chief AI officer, although relevant decisions “weren’t made in a vacuum.”
Appropriate use cases for the application of AI can be determined by ascertaining the level of risk that comes with any given AI system. In a manufacturing setting, for example, a tool that manages predictive maintenance through anonymized sensor data might quality as a low-risk use case. But one that automates supplier qualification or quality control carries a much higher degree of risk.
“Those are places where you want to make sure the right stakeholders are involved,” Clutter says. Questions to ask include what kinds of data the particular tool ingests, where that data is stored, and how decisions are made by the model that might affect company safety and regulatory compliance.
Some organizations are drawing “red lines” that rule out any dependence on AI for certain critical tasks. That might be the case where a manufacturer is making product for a federal government agency, whose safeguards are so strict that they can’t be given over to an autonomous system. Or the organization might need to prohibit the use of any proprietary data by the AI model whatsoever.
In any case, Clutter says, the need to build strong guardrails into AI systems should always be “top of mind.” The protection of sensitive data on customers, employees and products must be built into vendor contracts at the time of purchase.
At the minimum, vendor agreements need to contain clear guidance and restrictions on the secondary use of data by AI vendors. Ultimately, though, the responsibility for ensuring that those commitments are upheld rests with the primary user. Individuals within the organization must be trained and assigned appropriate responsibilities over AI models on an ongoing basis.
To Clutter, it’s a case of humans remaining in the loop. “The risk goes up the more you default to the tool,” she says. “It is not a replacement for your human judgment and your expertise.”
