Contrary to popular belief, Artificial Intelligence (AI) is not new. In fact, the term was first coined almost 50 years ago in the famous Dartmouth conference. After a few periods of reduced funding and interest, aka an “AI Winter”, we can all agree that we’re now in an “AI Summer”. The widespread use of neural networks and the impressive progress in generative AI, particularly with transformers, speak for themselves. The pace of change and adoption is remarkable, and many of us have been testing or using these technologies. At Thales DIS, we have been working on edge computing in AI for years. They are fundamental to our biometric products, and we have developed AI algorithms to embed into smart cards. We believe that many of our products could benefit from AI to enhance or complement their current performance.
Given the widespread use of AI, I would like to share a few thoughts.
First, AI’s longevity depends on its security. Like many technologies, AI’s success hinges on trust, which involves several aspects. The fuel of neural networks is data, so trusted neural networks require robust data protection at all stages: at rest, in transit and during processing. This is why Thales DIS has been focusing on confidential computing. As Greg Lavender, Intel Executive VP and CTO, stated:
“Our collaboration with Google and Thales endorses Intel’s commitment to zero trust principles. The combined power of Google’s new cloud offering with hardware-based Intel® Trust Domain Extensions (Intel® TDX), Intel’s independent verification service, Intel® Trust Authority, and Thales’ Cipher Trust Data Security Platform provides a robust and secure solution for end-to-end data workload protection.”
Combined with strong authentication, this is particularly relevant for AI, offering our customers an end-to-end data protection journey that safeguards against data poisoning, input manipulation, training or inference attacks.
Trust and edge computing in AI also encompasses protecting the infrastructure on which neural networks run, including public clouds, and ensuring proper data governance in line with national regulations, especially regarding privacy. Additionally, we must protect the model themselves from new types of cyberattacks, such as inference attacks, backdoors, adversarial attacks, and prompt injection for Large Language Models (LLMs). This highlights the extensive work required. The industry will progress by combining fundamental research with decades of cybersecurity expertise. This is Thales DIS’s roadmap and is why we’re spearheading “Cyber for AI” as a topic.
Second, AI must be deployed at the edge to succeed. While we discuss LLMs with hundreds of billions of parameters, the efficiency required by use cases, in terms of latency and performance, or device constraints (the size-weight-and-power paradigm), will drive edge computing in AI. This mirrors the balance sought between the core and the edge in cloud and mobile networks. The same will apply to AI.
This is good news. AI’s potential benefits to society, from health to safety, anomaly detection (e.g., fraud use cases), and process efficiency (e.g., support functions or operations), must align with environmental constraints. The energy consumption driven by AI and the growing demand for new data centres illustrate the “rebound effect.” How can we contain this? While virtuous eco-design principles are essential, they are insufficient. A true “frugal AI” approach is needed, moving edge computing to AI, utilising smaller, more focused models, and balancing data usage to avoid unnecessary network traffic.
Finally, AI cannot exist in a black box. AI must not operate as a black box because transparency is essential for trust, accountability, and regulatory compliance. Understanding AI’s decision-making processes helps identify and correct biases, ensuring fairness and enabling improvements. This is where Thales’ expertise and differentiation lie.
In Thales Air Traffic Management solutions for example, our AI integration prioritizes error-free performance, transparency, and explicability, always allowing human decision-makers to retain ultimate control. Leveraging decades of expertise in critical design and operating safety, Thales ensures that AI inputs are clear and understandable, facilitating informed decisions. With a robust team of over 600 experts and 100 doctorates specializing in AI, Thales has been at the forefront of neural network development since the 1990s, reinforcing the essential role of human oversight in AI-driven solutions.
In the IT world, AI is a focal point at trade shows. However, I often wonder why cybersecurity is rarely mentioned and why the industry, which advocates ESG objectives, does not emphasize it regarding AI. At Thales DIS, we aim to address both topics, advocating for the responsible use of AI in every sense.
Further reading:
- AI-powered smart cities: Navigating challenges and opportunities
- Trusted AI: a strategic challenge
- Intel, Thales, and Google Collaborate to Provide End-to-End Data Protection for Sensitive Workloads
- Air Traffic Management
- Thales teams up with Google Cloud to provide organisations with an extensive set of cyber detection and response capabilities
