AI: it all begins with the algorithm

How does AI applied to critical missions differ from consumer AI?

The spectacular progress we read about in the media is mainly happening in consumer applications of AI, such as video games, connected objects, online shopping and advertising, image and text generation and automatic translations.

But a company like Thales has customers and partners who are engaged in activities of vital importance for individual citizens and society at large such as air traffic, defence, security services, and cybersecurity. In other words, mission-critical sectors.

They need to make time-sensitive decisions in extremely constrained environments, and those decisions have a direct impact on human life, physical security and the ability of businesses and major infrastructure to operate. That makes all the difference when it comes to developing and using AI technologies. 

The critical nature of these tasks poses a whole range of specific challenges that very few technology companies are capable of meeting.

Two ways of learning

Under the hood of every form of AI there are mathematical algorithms, some of them data-driven, others based on models, laws of physics and mathematical principles.  

The AI that's talked about most — AI for consumer applications — is data-driven. It uses deep learning algorithms that need to be fed phenomenal quantities of data on a permanent basis. 

For AI used in critical environments, the situation is a little more complex. In some cases, critical systems and their many sensors generate even more data than consumer applications. But in other cases, there may be very little data available — or none at all, quite simply because the situations we are trying to understand and master have never happened before. Data-driven AI can work well in those first cases, but it has a fundamental flaw in that it doesn't explain what it's doing. Indeed it can even produce false results or be misled by its data, and nobody understands why. In these cases, model-based AI can be extremely useful. 

One of Thales's major strengths is the ability to develop both these complementary types of algorithms, supported by expertise in the technologies that are driving the digital revolution today, namely connectivity, IoT and cybersecurity. Cybersecurity expertise, in particular, is a key differentiator for the Thales Group, making it possible to capture, analyse and transmit data securely and reliably in applications where security is a fundamental requirement. 

The energy challenge

One aspect of AI that receives less media attention is energy consumption.  

In consumer AI applications, vast quantities of often useless data is stored and analysed in gigantic datacentres, preferably located in cold climates because of the enormous amount of energy they consume. 

Thales was one of the first companies to privilege smart data, which, contrary to big data, only collects the data we really need. 

In Thales's markets, energy constraints are even more critical, particularly for onboard systems. Imagine the complexity, for example, of embedding AI applications on board a fighter aircraft!  

Meeting the energy challenge is crucial for the future, and not only in the digital world. The Internet consumes more energy today than the much-maligned air transport system. And an AI system consumes between 10,000 and 1 million times more energy than the neurons of a human brain.