This blog post is based on a presentation given by Dave Risi at AIHce EXP 2022. An expanded version was published in AIHA's 2022 ebook, The Essentials of OEHS Communication.

Dave Risi, CIH, CSP, has been involved in the development of artificial intelligence (AI) and machine learning (ML) solutions for application in OEHS for some time. After working with software company Oracle to develop its environmental health and safety (EHS) application, he acquired the application in 2001 and formed Spiramid LLC to bring its capabilities to a wider market. Risi now oversees the continuing development of the EHS application at VelocityEHS, which acquired Spiramid in 2019.

Successful deployment of AI projects often requires becoming acquainted with unfamiliar knowledge and collaborating with other professions. These two challenges can be addressed through effective communication, as Risi explained during his presentation at AIHce EXP 2022, titled “Effective Communication in the Time of Big Data, Artificial Intelligence, and Machine Learning.” Even if you are not planning to incorporate AI into your workplace, this blog post may offer insights on overcoming communication challenges that commonly occur when developing and implementing solutions alongside non-OEHS colleagues.

AI Project Management

AI allows OEHS professionals not only to understand what the data says but to make predictive inferences about that data so that potential issues can be addressed before they happen. But although the benefits of AI can be game changers for OEHS professionals, “going through a project like this is not a small task,” Risi noted. “Don’t think it’s just an [OEHS] survey.” In his own experience, bringing data scientists on board to work on an AI project led Risi to realize that an AI project is more than just software programming. “You really need to have the right level of staffing to get involved in these things,” he said.

Risi explained that AI projects may fail for several reasons, including lack of support from leadership, lack of collaboration among key decision-makers, and unreadiness of the organization’s overall culture to transition to AI. AI projects also tend to run over budget and timeline, so project managers must secure and cultivate support and buy-in to ensure projects receive the time and resources they need. The use of biased data, even if unintentional, may also result in an algorithm’s predictions not being as reliable as hoped and failing to answer the questions that the project team and business set out to answer. However, “poor communication is probably the biggest cause of issues with [AI] projects,” Risi stated.

AI Project Roles

For an OEHS professional engaging in an AI project, effective communication starts with learning a new vocabulary of terms and acronyms to more effectively communicate with the new staff members being brought onboard. In addition to the project manager in charge, a typical AI project will involve subject matter experts, data scientists, data engineers, and data officers.

A data scientist’s role is to define solutions, while a data engineer assesses how to collect, analyze, and sort data—that is, how to put data in a framework that can be used by AI algorithms. Finally, a data officer’s job is to ensure that data is reliable and managed responsibly, that collection methods don’t raise concerns among employees, and that employees can be assured of data privacy. All team members should have the opportunity to review decisions.

Project Stakeholder Engagement

Communicating and engaging with key stakeholders is critical to project success. You will need support from all levels of management, from the top down. The supervisors of workers being monitored need to understand how and why you’re collecting data, the benefits that you expect to gain, and how data will be kept private. Employees need to be brought on board to generate support from the bottom up. They will need to understand what you’re doing and why as well as how their private data will be protected.

Other potential stakeholders include workers’ unions and your organization’s legal and human resources departments. Marketing or public relations departments can help you recruit other stakeholders, and it may be easier to accomplish your goals with them on board.

The Steps of an AI Project

Risi’s four steps of an AI project include:

1. scoping or defining the project
2. defining data, establishing the baseline, and labeling and organizing data
3. modeling, which involves selecting, building, and adjusting a model as well as training it to make its predictions more precise
4. deployment, when the model is implemented in the workplace and continuously monitored and maintained

1. Scoping. The first step in any project is to brainstorm. For example, what tasks can be automated instead of performed manually and what high-hazard environments are present? Find out if your organization has AI applications already in use and if they can be expanded or reused with different data.

Write your project’s architectural proposal, which lays out what the algorithm will be and where data will come from. Be sure to define the benefits of your AI solution in a way that clearly demonstrates the value for executives. For instance, will your solution improve safety, improve product quality, safely increase production, or reduce low-value work through automation? Calculate the project’s return on investment (ROI) and the length of time it will take for the project to recover its initial costs (payback period) through the efficiencies and cost reductions it seeks to deliver.

After the architectural proposal comes the “project charter,” which outlines important characteristics of the project. Critically, it defines the team members’ assumptions about where and how they could collect data. If you eventually find you can’t get this data, you can revisit your charter to identify what needs to be changed.

Make sure the person granting approval of the charter is engaged and understands both the project and charter in detail. They will need to comprehend the value of what you’re doing in case you need their approval for a budget increase in the future. In fact, communicating clearly to all stakeholders is important and should occur in person whenever safe or feasible.

2. Defining Data. The team’s data scientists will then determine what data is appropriate for the model that will be built. During this step, you’ll define what types of data will be used, where it will come from, and how much is needed to prove the model. Risi cautioned that he had seen projects in which team members couldn’t collect enough data for the model to be effectively trained, resulting in the project stalling after months of work.

This is also the point to address workers’ legal, regulatory, and ethical concerns about data privacy. Some data collection methods, such as video monitoring, “really could have a negative impact on your workers’ morale,” Risi said. Your workplace culture can be negatively impacted if workers feel that they are being watched or micromanaged, pressured to perform exceptionally well, or encouraged to compete with each other. Being on camera may lead them to feel that they cannot take breaks—even if they must have breaks to avoid injury. In addition, ensure that you address any concerns from workers that automation may lead to them losing their jobs. Be sure workers trust that you’re collecting data for a valuable purpose, such as creating a safer workplace, and address their concerns immediately. “A lack of trust is another common obstacle to project success,” Risi warned. “Successful projects win workers over and engage them as members of the team.”

The team’s data officers are in charge of working with project stakeholders to ensure data is managed responsibly and privately. Informing stakeholders about this should be a key element of your communication plan. Work to make sure that all team members and stakeholders are on the same page, and that the process is transparent. Provide periodic updates, especially when the project reaches milestones laid out in the charter. Celebrating milestones helps the project maintain enthusiasm and momentum.

3. Modeling. When your data scientists have identified at least one algorithm and tailored it to your project’s objectives, your team can begin training the algorithm by feeding data into it. You will need to define metrics that can measure the algorithm’s success. As you progress through training your model, make sure to fine-tune your data and algorithm to meet your metrics targets.

At this stage, Risi advised that it’s important to ensure stakeholders are up to date on the model you’ve chosen—that they understand what it is, why you’re using it, and why it’s important. The team members and stakeholders should all understand and agree on testing protocols and metrics. Continuing to provide periodic updates helps keep stakeholders aware of the model’s progress in terms of your key metrics—which may need to be redefined or reassessed as the project continues—and any changes or delays to the project’s timeline. It’s important to communicate obstacles and challenges, as well as project achievements and important milestones.

4. Deployment. Once the model has been fed enough data to meet your team’s threshold for accuracy, the technology can be rolled out in the workplace. As the model analyzes live data, the project team should measure its accuracy, track trends, seek constant feedback, adjust the data and model as needed, and be mindful of data bias. It may be tempting to believe that deploying an AI solution in the workplace is the last step of an AI project, but “you’re always going to be tweaking the algorithm, getting more data, checking the data, and seeing if it’s biased,” Risi said.

As your team celebrates project successes, Risi stressed that you must be sure to communicate the benefits of implementing the technology to stakeholders. Once again, provide them with key metrics that measure the model’s success in a clear and meaningful way. Be sure that any data shared is mindful of employees’ and other stakeholders’ privacy concerns. You may begin to think about additional ways that your AI model could be applied to other data needs and challenges in your organization.

Achieving “Wins”

Risi warned that the failure of an AI project to deliver as intended may result in disengaged and mistrustful stakeholders and even greater difficulty with getting any subsequent projects off the ground. Avoiding failure is easier for project teams who take the time they need to deliver and refrain from overcommitting in their promises to stakeholders. Particularly if you’re undertaking your first AI project, set realistic expectations and identify targets you feel confident you will meet. Achieve “wins” to celebrate with team members and stakeholders, then work toward more ambitious goals.

To learn more about AI applications in OEHS, read “Keeping Pace with the AI Revolution: Considerations for OEHS Professionals” in the June–July 2022 Synergist.

Resources:

Hive: “15 Fascinating Project Management Statistics.”

QuantumBlack AI by McKinsey: “The State of AI in 2021” (Dec. 8, 2021).

Risi, Dave. “Effective Communication in the Time of Big Data, Artificial Intelligence, and Machine Learning.” AIHce EXP, AIHA, May 24, 2022, virtual. Conference Presentation.