Ensuring Edge AI Works for Everyone Through Responsible Enablement

While some are thinking about how to make AI work in the first place, NXP is looking beyond, asking the question: how
do
we keep AI working in a safe, reliable and responsible way? This is where Responsible AI takes center stage, working
with technology, government and business leaders to become a reality.

Imagine you are driving in your car to meet with friends, excited to enjoy your favorite dinner. It’s been a while
since
you’ve seen them, so you want to look your best, but as you drive an alarm keeps going off, and you can’t understand
why. The alert notifications you’re receiving are from your vehicle’s driver monitoring system (DMS), telling you
that
you are not paying close enough attention, even though you are driving well.

Unbeknownst to you, the reason that these scenarios happen is due to the training data used by the Artificial
Intelligence (AI) models powering the computer vision in the vehicle. For some reason, somehow, the AI model
misunderstood its live input due to bias in training data that indicates female drivers are more often classified as
“distracted by personal grooming”, which is a result of subtle misrepresentations of people during its training.

Risks to Responsible Edge AI Development

This is not just an example of the risks of using AI to analyze data and make predictions; it’s an example of the
fairness and robustness issues with AI/ML systems and how they can influence modern life. In the same way that an
individual may be denied financial services based on incorrect biases present in training data, edge AI can also
lead to
discrimination when the proper measures and risk assessments aren’t taken. The intelligent edge plays a crucial role
in
connecting the physical world to the digital one. Physical AI, the topic at the intersection of generative AI and
robotics, can only be created through edge devices, and not the cloud alone. Therefore, the risks of AI misalignment
at
the edge require extra scrutiny to prevent physical harm and discrimination.

The world is at a critical juncture when it comes to AI in everyday life. In January 2025, a Boston Consulting Group
survey  found that 75% of C-Suite executives named AI as a top 3 strategic priority for 2025. At the
same
time, less
than
one third of companies have upskilled less than one-quarter of their workforce to use AI, highlighting the immediate
need for education and awareness.

Edge AI, the Responsible Way

While many companies are thinking about how to make AI work in the first place, at NXP, we are looking beyond, asking
the question: how do we keep AI working in a safe, reliable and responsible way? This is where Responsible AI takes
center stage, working with technology, government and business leaders to become a reality.

Responsible AI is not one, distinct and separate technology, nor is it a collection of policies and best practices.
Responsible AI permeates every facet, both technical and non technical, be it machine learning, generative AI and
language models, time-series data, computer vision and voice recognition; all types of intelligent software, sensors
and
hardware. The risks of AI impact businesses and individuals—responsible AI must represent both parties equally.

Therefore, it takes a concerted and comprehensive effort to bring Responsible AI into practice. At NXP, we have
examined
the topic through the lens of edge AI enablement. As a leader in the intelligent edge, we’ve authored a white paper
on
Responsible AI Enablement.

The goal of the white paper is to make recent legislation like the EU AI Act more accessible and interpretable,
discuss
and address risks with edge AI, highlight the roles and responsibilities of SoC vendors and give an overview of how
NXP
is
already contributing to responsible AI through SW and tooling. For example, in the DMS example mentioned earlier,
NXP is
developing Explainable AI (XAI) software as part of our eIQ®
Toolkit that helps detect biases after model training,
before deployment. This will help prevent discrimination, ensure robustness and enable developers to identify risks
early and receive an explanation.

There are many ways in which edge AI can benefit humanity; increased automation and productivity, safe and more
sustainable transportation and more resource-efficient computing. Responsible enablement plays a crucial role in
making
sure the benefits of AI at the edge are maximized while minimizing any possible harm.

Our Responsible Edge AI Enablement white paper was coauthored by
Davis Sawyer, Wil Michiels, Jolijn Martens and Wouter
van
der Loop.