Artificial Intelligence (AI) is growing fast, upwards and outwards, out of its adolescence. We have moved beyond the cats and dogs phase of Machine Learning (ML) where two furry animals, each with four legs, teeth and claws might be mistaken for each other.
The era of Supervised Machine Learning (SML) where we had to tell our computers that cats are typically (but not always) smaller than dogs, come in a more uniform total variety, only have pointy (not floppy) ears and are more likely to be photographed sat staring at a wall is fast becoming a memory.
We’ve been able to ‘expose’ ML to enough datasets (pictures of different cats and dogs) that it has started to enter the era of Unsupervised Machine Learning (UML), where it can make more sophisticated inferences and start to provide us with more AI power.
While there’s still a market for tracking cats and dogs with AI in some areas, we are now taking those same core object identification technologies forward and developing massively more complex algorithms for massively more complex tasks. This is the sort of intelligence required by modern software applications used in international defense by governments and the military, with wider use also in high-end civil engineering projects.
Un nouveau savoir-faire know-how
Bidding to bring a new kind of savoir-faire to the know-how behind AI algorithms used to analyze geospatial data and satellite images of planet Earth is French company Earthcube. The Paris-based startup was formed in 2016 with a specific mission to develop a new set of broader, wider and more complex algorithms designed to analyze and identify objects and ‘situations’ from satellite images, geospatial data or data from open sources of data, such as social networks or other freely available public data streams.
With a team of 50 people based in Paris, Earthcube has been working with aerospace giants such as Airbus to bring AI to its IT operations layer. Its solutions are also used by several countries for strategic intelligence and different organizations for economic intelligence.
“Our objective is to be the first European Defense Tech unicorn,” asserted co-founder and CEO Arnaud Guérin without a blush. Guérin was quoted alongside his co-founder CTO counterpart Renaud Allious on French media title Challenges in an interview with journalist Vincent Lamigeon in March this year.
Lamigeon makes note of Earthcube’s not inconsiderable rounds of funding and market interest which has seen it approached by a French investment fund of the Ministry of the Armed Forces, Definvest, but also by the fund of the CIA, In-Q-Tel.
A ‘plurality of information’ to enable new geospatial AI
Earthcube’s AI algorithms are able to detect even highly pixelated images i.e. images in an extremely low definition at poor resolution. “Until now, the majority of the military analysts’ job was to monitor strategic sites (e.g. the Strait of Hormuz, the Chinese sea, Libya…) where nothing happens most of the time,” said CEO Guérin. “Our software computes huge amounts of satellite images and raises alerts on dates where [any] abnormal behaviors are found, such as equipment in motion, or the arrival of transport aircraft.”
So how does it work?
The software works by ‘establishing’ a database that brings together a set of ‘geographical indicators’ in the form of data. The more information about physical images for a given location on our Earth are collected (whether they be topographical landmass images, or images of industrial facilities, automotive or airborne vehicles and so on), the more details can be provided about them. Earthcube says that it is then a question of maximizing these images through a ‘plurality of information’, whether it is environmental, administrative, or even military.
The AI here then builds up an ‘image footprint’ of any single location based upon a multiple layered stack of information. The layers of a location are not simply a map coordinate. Instead, they are composed of various tiers including image metadata supplied from image proprieties and satellite information, geographical location, surrounding road network density, description of soil occupation and even snow cover to note the presence of snow at a given date.
The information for the AI analytics performed is then extracted from the combined images’ (plural) footprints.
“The results will depend on the presence of data within those footprints. Indeed, the collected data can take multiple forms due to the plural and various sources. It can easily be seen as a ‘superposition’ [as in quantum superposition] of multiple rasters and vector layers that cover the entire surface of the globe, on which we cut the only parts that have an interest to us,” notes Earthcube, on an explanatory blog posting.
Another great French technology?
Earthcube now plans to expand internationally, particularly (and notwithstanding the arguably highly disruptive effects and consequences of Brexit) to the United Kingdom.
France may perhaps not always be thought of as a technology leader first and foremost, but do remember that the French gave us Ubisoft (Assassin’s Creed video games), Orange (telecoms) and Schneider Electric (energy IT). Before that, the French gave us the parachute, pasteurization, the pencil sharpener, the Etch A Sketch… and, after all, they did come up with a term for tech innovation that we all use every day – entrepreneur, n’est-ce pas?