This newsletter is designed to bring you developments that might otherwise be missed in the tsunami of digital economy information. This week we go to Finland to look at AI innovation there.
In 2017, Finland was one of the first developed economiesto publish an national AI plan. The Finnish Government recognised that Finland did not have the resources to match AI developments in the US and China, so its goal was to be a world leader in the practical application of AI in economic and social life. The following AI projects from the Finnish Centre for AI (FCAI) help illustrate what this means in practice.
Making robots more dexterous
Robot ‘hands’ are a bit like BBQ tongs – strong grip once engaged, but awkward in grabbing (and a few dropped sausages in the process).
Multi-finger robotic hands are essential to many of the tasks we want robots to do, such as aged care. But this is very challenging because the AI has to control and co-ordinate more fingers. Currently, multi-finger grasping by most AI is painfully slow, and it can take minutes to generate a single grasp (by which time your robot-made tea has gone cold).
FCAI researchers have recently developed two fast finger grasping methods, called Multi-FinGAN and DDGC. Multi-FinGAN synthesizes, directly from camera images, high-quality grasps in about a second on individual objects – 20 to 30 times faster than current AI.
DDGC is even more impressive. The AI can pick through a table top full of other things and grasp onto the targeted object.
But probably the innovation of wider application is how the robots are trained in grasping and handling objects. Like most other AI methods (and ourselves as children learning to use our own hands), Multi-FinGAN and DDGC require a lot of training data to work well. However, collecting data with real robots is highly time-consuming and wears out the robotic hardware. FCAI uses completely synthetic data collected from simulation – which their experiments with the programmed robots shows is close to use in a real human-centric environment, such as in aged care.
Spectral imaging in your pocket
Colour can tell us a lot – from the health or maturity stage of a plant, to performing a dermatological analysis of your skin, to getting the perfect colour match for paint on your walls at home.
However, we each perceive colour differently – and variable external factors such as light also have an impact.
Hyperspectral technology achieves very precise analysis of colour by examining each pixel against a continuous range of wavelengths. This technology has been around for some time but has required specialised, clunky equipment.
FCAI researchers have developed ‘in your pocket’ hyperspectral technology which combines a smartphone, a small snap-on peripheral device, and an app that determines the colour drawing on a cloud-based service.
When your digital assistant becomes your personal assistant
Digital assistants like Siri and Alexa are already embedded in our lives, but the FCIA researchers believe these crop of assistants have two limitations: first, they are based on only limited data, usually the service provider’s products or websites; and second, they do not recognise a user’s ‘evolving intent’ as the user’s information needs change.
FCAI researchers are building an ‘EntityBot’ that recommends ‘entities’ related to the user’s tasks, and these ‘entities’ are the broad universe of information and apps such as documents, files, emails opened, and keywords.
EntityBot learns what we are doing and recognizes the task that we are working on, as well as how, other ‘entities’ such contact information, applications, and documents relate to it. Its recommendations are always based on what the user is currently doing.
EntityBot does this – and here is the spooky bit - by 24/7 digital activity monitoring of our usage across all our devices, with a screen capture made every 2 seconds. The FCAI team has flagged the obvious privacy issues, and are looking at how to build in accountability measures, such as the EntityBot explaining how it is making its recommendations.
A potential application of the EntityBot is to support aging workers by, in effect, underpinning their performance by ‘nudging’ them through the recommendations it makes.
Cranes and humans can be a dangerous mix in large scale industrial halls.
FCAI is working with Kronecranes on Artificial Intelligence for Industrial Vision, which focuses on developing a computer vision technique known as visual SLAM (simultaneous localisation and mapping) which tolerates feature poor areas, varying lights and reflections.
The biggest challenge, as the lead FCAI researcher says, “is getting computer vision to recognise humans. This requires deep learning.”
The solution uses 3D multi-object tracking to find the humans as they move about in the warehouse space. This technology also has been around for some time, but the key to the project is combining this technology with navigation and positioning software so the crane can always work out where it is relative to the humans.
Perhaps one of the most innovative elements of the Finnish AI program is the recognition that the general public need to be brought along with the development of AI. It is well recognised that public trust is critical to the widespread use of AI, but tangible steps to build public confidence and knowledge are scarce on the ground.
Finland has set up a free online course run by the University of Helsinki and a private online education provider, Reaktor, the Elements of AI. Over the last couple of years, over 2% of Finland’s population have completed the course. There are now over 700,00 users globally are enrolled in the course, and 40% are women (twice the level of typical computing courses).
To see Finland’s latest national report on AI.