As CIOs have continued to lead digital transformation initiatives across their organizations in recent years, computer vision AI technologies with environmental health and safety (EHS) applications have quickly evolved from relative obscurity into must-haves for workplace safety. The growing adoption of these solutions signifies amazing progress in closing the longstanding EHS innovation gap – but it also means that CIOs must now contend with far more prospective vendors (and, in many cases, far more confusing marketing claims) than ever before.
Choosing the right vendor demands careful consideration not only of technology capabilities but also of factors ranging from ethics to data management, to operational efficacy. Here are four key questions CIOs should ask prospective vendors to help guide their selection process:
1. Who sees our images and videos?
Why it matters: Transparency and security are paramount when dealing with sensitive data. It's essential to scrutinize whether the vendor relies on third parties for essential processes like data labeling. Any third parties involved must adhere to stringent security protocols to prevent data breaches – however, verifying this can be challenging, and third-party risk exposure will likely persist regardless.
Additionally, I've also seen firsthand the complications that can arise when third-party involvement delays the timely processing of hazard detection and alerting functionalities. Ensure your vendor has robust, direct control over data processing or – at the very least – ironclad agreements with all involved parties to safeguard data integrity and ensure operational responsiveness.
2. How do you ensure a diverse dataset and ethical AI practices?
Why it matters: The effectiveness and fairness of AI-driven solutions depend heavily on diverse datasets and ethical development standards. Diverse datasets should encompass a wide array of environmental conditions, equipment types, physical processes, and human factors, such as different skin tones, body sizes, and types of attire. This variety helps AI systems to be effective across various real-world scenarios and prevents biases that can arise from homogenous data.
What to look for in vendors
• Dataset diversity: Inquire about the origins of the vendor’s data. Effective vendors should source data from a variety of settings and conditions to ensure the robustness and adaptability of their AI models. Check whether they incorporate a broad spectrum of human-related factors and physical environments.
• Ethical AI practices: Determine how the vendor addresses privacy and ethical considerations. Look for use of anonymization techniques that protect individual identities without sacrificing the utility of the data. Transparency about data collection, usage, and storage is essential to prevent misuse and foster trust.
• Bias mitigation strategies: Ask about the vendor’s practices for detecting and mitigating bias. Vendors should demonstrate continuous efforts to evaluate and improve their models using fairness metrics and regular retraining with new, diverse datasets.
• Regulatory and ethical compliance: Ensure that the vendor adheres to industry standards and ethical guidelines, potentially overseen by an ethics committee or similar governance structures. This oversight helps guarantee that AI development supports a safe, productive, and trust-based workplace environment.
By thoroughly vetting potential vendors on these fronts, CIOs can ensure that the AI solutions they adopt not only enhance workplace safety but also do so in a fair and ethically responsible manner.
3. How many AI models can you scale per camera and per facility?
Why it matters: An often-overlooked yet reliable litmus test for an AI system is its scalability. For example, a technology that can allow a single camera to simultaneously detect the multiple types of hazards that may occur in its purview — such as PPE non-compliance, vehicle-pedestrian interactions, and restricted area access — will be more cost-effective and simpler to manage than multiple systems.
Scalability also involves considering bandwidth implications to ensure responsiveness as more functions, use cases, or other variables are added. The solution’s ability to adapt and integrate multiple AI models without compromising performance is key to ensuring 24/7 visibility into the full range of hazards to which your workplace is predisposed.
4. How do you validate accuracy after deployment?
Why it matters: Initial testing and continuous validation of accuracy are vital for the successful implementation and operation of any computer vision AI-driven solution. During rollout, it is normal for inaccuracies to emerge as the AI models adjust to real-world conditions and learn their environment.
Active involvement of users in reporting and addressing these inaccuracies is crucial. A vendor that encourages feedback and provides straightforward mechanisms for reporting issues and verifying accuracy demonstrates a commitment to partnership and continuous improvement. This collaborative approach helps the system evolve in line with the actual safety needs of your workplace.
Above all else, it’s important to remember that choosing the right computer vision AI vendor for your organization and EHS function requires rigorous evaluation of not only their technology but also their ethical standards, integrity, and commitment to operational excellence. This decision has profound implications that cannot be overstated – but with a detailed, informed approach, CIOs can ensure they identify and partner with vendors who are not just providers but true allies in the quest for a safer, healthier, and more successful workplace.