Computer vision technology: Navigating its risks and rewards

[MattQuirk]

Computer vision technology: Navigating its risks and rewards

A growing cohort of OEMs and innovators are building advanced computer vision technology solutions to reshape how their customers operate. From improving safety to automating quality control to enhancing retail shopping experiences, combining AI and machine learning with live video and vision supercharges the ability of machines to interpret and respond to the physical world.

HPE OEM partner Fogsphere, for example, is revolutionizing worker safety with its computer vision-powered technology. Combining AI computer vision, CCTV, and IoT technology, Fogsphere solutions monitor hazardous work environments to check that workers are consistently following safety protocols and using protective gear and equipment. Its computer vision system can identify in real time whether workers are wearing hardhats, gloves, and other equipment, and instantly issue safety warnings.

Meanwhile, Sensei is making cashierless shopping possible with its computer vision technology. Its solutions can track which products in-store shoppers interact with, yielding insights that help retailers reduce friction, eliminate stockouts, and delight their customers.

Computer vision superpowers also come with risks
As with any powerful new tool, the use of computer vision technology also presents risks and challenges, including:   

• Privacy concerns: Monitoring people in workplaces and public spaces and storing large amounts of visual data raises privacy and data security questions.
• Bias and discrimination: Biases included in training data and algorithms can produce discriminatory outcomes and higher error rates for certain demographics, presenting ethical, legal, and accuracy challenges.
• Accuracy and reliability issues: Computer vision AI isn’t foolproof. Errors, misidentifications, and false positives can have critical consequences when they drive incorrect decisions and actions.
• Security risks: AI computer vision systems and their data could offer new targets and exploits for cyberattackers.
• Ethical and regulatory challenges: Managing issues of consent, transparency, and accountability is crucial for responsible use and regulatory compliance.

To successfully navigate the complexities of AI computer vision, OEMs and enterprises need a holistic and forward-thinking approach to system design.

Proactively design AI with people in mind

An effective strategy for avoiding many computer vision risks and challenges before they crop up is to use a human-centered design approach.

This design philosophy places human needs and experiences at the forefront to help ensure that AI systems enhance rather than undermine human capabilities and interests. It prioritizes building AI applications that can adapt to diverse contexts, reducing the risk of bias and discrimination. It also designs AI systems to be transparent and explainable, so users can understand how and why decisions and automated actions are made. As a result, humans can monitor, review, and intervene in AI’s decisions and actions.

Build on a foundation of responsible AI

It’s far more effective to build AI systems on a solid foundation of well-considered ethical standards and strategies than to try to bolt on fixes and implement new policies once problems arise.

At HPE, we’ve developed a comprehensive set of AI ethics and principles to guide how we design, build, and use new AI innovations. These principles are designed to ensure responsible AI use, prioritizing privacy, security, and minimizing bias and errors. They include:

• Having human oversight to prevent misuse
• Promoting inclusivity and fairness
• Incorporating rigorous quality testing and safeguards
• Insisting on transparent, responsible AI that allows for accountability and the ability to challenge AI outcomes

Developed and maintained by the HPE Advisory Board, these principles and practices underpin all of our AI activities. HPE OEM partners and customers can rest assured that our AI-related products and services are built upon this responsible foundation from the start.  

Lead with transparency and continuously audit computer vision solutions

To ensure privacy and combat bias with AI computer vision systems, businesses should follow practices that include securing consent for data use, anonymizing data, providing opt-out options, and regularly auditing algorithms and results for errors and bias.

As mentioned above, transparency into how video data will be used and explainability of AI output are crucial for trust and avoiding the perception that the system is a mysterious black box. Building in mechanisms for human oversight of AI decision-making is an important check on errors and bias. In addition, with the recent introduction of the EU AI Act, explainable AI is fast becoming an essential component in AI-driven systems.

Of course, no tech discussion is complete without talking about the critical importance of robust cybersecurity. Computer vision systems and their stores of visual data will be irresistible targets for cyberattackers. Including advanced encryption methods, strict access protocols, and regular security audits from the outset is crucial for safeguarding against unauthorized access and adversarial attacks.

Sharpen your focus on computer vision

Ready to start building computer vision solutions? Think about beginning with an HPE AI Service Transformation Workshop, where HPE AI experts will help you identify the right use cases, define desired outcomes, and highlight potential challenges. You’ll come away with practical AI implementation strategies grounded in HPE’s AI ethics and principles.

HPE OEM Solutions also has a dedicated team of AI experts ready to help you achieve your AI objectives. Let’s start working on transformative and responsible AI computer vision solutions for your customers today. If this sounds good to you, please get in touch.