Thomas Jensen, CEO and Dr Barry Norton, Vice President of Research, Milestone Systems discuss the best pathway to successful AI adoption.
The growing use of AI in video security is transforming how the industry protects people, assets and infrastructure.
From smart buildings that automatically adjust access controls based on real-time threats to hospitals using artificial intelligence-powered analytics for patient safety, these technologies have changed how video technology is used in and beyond security.
The transformation brings new, ever-changing rules that security professionals must navigate.
The rise of powerful artificial intelligence capabilities has created significant opportunities for enhancing traditional security systems.
However, it also raises relevant questions about privacy, transparency, freedom and the responsible use of technology.
As technologies become more sophisticated, industry and security professionals face the challenge of getting the most out of artificial intelligence-powered innovations while ensuring compliance within a complex and evolving regulatory framework.
In everyday work-life, security professionals will need to develop new skills focused on managing and using these systems rather than performing routine monitoring tasks themselves.
Or as the new saying goes: “It’s not AI that will replace you, but someone using AI who will.”
Artificial intelligence systems are only as good as the data they’re trained on.
Just like a person learns by reading and understanding information relevant for a given context and not simply by leafing through random books or websites, artificial intelligence models require large amounts of relevant and high-quality data to be reliable.
But where does such high-quality data come from, and how do we ensure that it’s responsibly sourced and legally compliant?
Unlike earlier machine learning models, which are trained on carefully collected and curated data, generative models (such as Large Language Models) are trained on data scraped from the internet, raising questions about consent, bias and quality.
Collecting data from the internet has been necessary because no datasets were available at such a major scale.
But when artificial intelligence models learn from unverified datasets, problems with bias and trustworthiness are inevitable.
The solution lies in steering generative models with datasets that are traceable and which document sources, processing history and usage scenarios and permissions.
Milestone Systems’ Project Hafnia is a platform and a library of trusted video data, curated, tagged and responsibly sourced for artificial intelligence model training.
Rather than relying on unverified data from the internet, this initiative demonstrates how organizations can transform raw video data into artificial intelligence training material.
In collaboration with the City of Dubuque in Iowa, Project Hafnia invested in annotating and curating years of video traffic data to develop a new vision language model (VLM).
The results were significant: artificial intelligence model accuracy jumped from up to 80% to over 95%. Below the 80% threshold, false positives are too frequent.
Another innovation is synthetic data, where entire cities can be built virtually, allowing researchers and developers to train models using simulated scenarios instead of real people’s private data.
And in a semi-synthetic setting, real video data can be anonymized, switching out faces and license plates with synthetically generated ones.
Both approaches protect privacy, as countless variations can be created to improve accuracy without compromising anyone’s personal data.
Most technology buyers now believe responsible innovation will be a prerequisite in the future.
That was the result of a Milestone-commissioned 2023 survey with 150 technology decision makers around the globe.
This finding aligns with a 2022 survey of senior executives and directors by the Massachusetts Institute of Technology (MIT), concluding that “responsible technology is now more than a buzzword”.
As artificial intelligence continues to gain ground, the security industry has also entered the era of agentics, where systems will operate with unprecedented autonomy.
These AI agents are fundamentally different from earlier systems as they can understand contexts, make decisions according to user direction and take actions independently without following prescribed steps.
The revolutionary aspect is how these agents will augment human capabilities.
While previous artificial intelligence focuses on analysis and recommendations, these new systems will increasingly take autonomous actions when appropriate, marking a shift from passive to proactive security management.
AI agents in the security and surveillance industry understand context by continuously processing live feeds from cameras and sensors, allowing them to adapt to changing situations.
Unlike previous models that use static information, agents based on generative AI act in real time, making decisions and triggering responses based on immediate analysis of relevant information.
They can also adapt their behavior over time through adaptive learning, improving their effectiveness without full retraining.
However, this doesn’t mean humans will become obsolete. The human element remains crucial.
While AI can process vast amounts of data and identify patterns that surpass human capability, security professionals must direct these agents, maintain oversight and retain final decision-making authority.
This “human-in-the-loop” approach ensures that artificial intelligence serves as a powerful tool for augmenting human judgment rather than replacing it.
What this means practically is that a security operations center that currently requires a large team of operators might function more efficiently with a slightly smaller team of professionals working in partnership with agents.
The human role will evolve to focus on high-level decision making and handling complex situations that require judgment and empathy.
In this era of agentics, the security industry has a unique opportunity to set an example for other sectors by proving that responsible innovation and technological advancement can work hand in hand.
The companies that will thrive are those that can demonstrate genuine commitment to responsible innovation protecting basic human rights and the people of the societies they serve.
The rapid advancement of artificial intelligence capabilities has prompted governments worldwide to develop and rethink regulatory frameworks.
For the industry, this regulatory landscape means adapting to new requirements around transparency, bias, accountability and human oversight.
The regulatory landscape continues to evolve, with new frameworks emerging that specifically address the unique challenges of artificial intelligence in security applications.
These include requirements for regular system audits, impact assessments and specific guidelines for handling sensitive personal data in security contexts.
AI providers should document how systems make decisions and ensure human operators maintain oversight and final authority over critical security decisions.
The regulations also emphasize the importance of data protection, requiring organizations to establish strict controls over how information is collected, stored and used within artificial intelligence-powered security systems.
We must be realistic about striking a balance between innovation and responsibility.
Being overly cautious can stifle innovation just as much as being reckless.
Countries and companies that don’t strike this balance risk losing their market position. The key is to understand and mitigate all risks to human safety and human rights.
End users have a powerful role to play in shaping the future of responsible technology as well.
By making responsible use one of the key principles in the procurement processes, requiring transparency in artificial intelligence deployments and actively engaging with technology providers on ethical considerations, security leaders can drive meaningful change across the industry.
The key question is not whether to adopt AI, but how to do so in a way that supports trust and accountability.
Striking a balance between innovation and regulation is no longer optional; it’s essential for our industry in the long run.