By 2035 artificial intelligence (AI) will be everywhere, but it will be invisible. Behind the scenes, it will help steer the choices we make, offering us the best options to lead a good, healthy life. It will help customise the services and products we buy to best match our preferences and all of this will be achieved by its lightning-fast ability to process huge swaths of data and extract knowledge.
Imagine that at some point in the future you are taken down to the hospital with what appears to be a rare, complex condition. Immediately, the doctors run your personal history and medical parameters through their AI system, searching for identical patterns in a worldwide database of anonymous patient data. They find a dozen matches and see potential treatments and their effects. So there is less guessing and reliance on solely human experience, and luck being in the right hospital. Wherever you live, and who ever you are, you’ll get the best available diagnosis.But wait … chances are that you wont have to go to hospital. Why? Because with your physical parameters you are constantly monitored, and you’ll be offered intelligent choices at every step in your life. Whether it’s about the food you eat, the exercise you take, your career choices – all of this will keep you healthy for much longer.
One of the most successful branches of AI is machine learning. Machine learning (ML) algorithms allow computers to learn and detect patterns from huge amounts of data and establish the relationship between inputs and outputs, between huge swaths of data and meaningful conclusions.
|NASA used artificial intelligence to discover planets outside our solar system, such as a recently discovered eighth planet circling Kepler-90, a Sun-like star 2,545 light-years from Earth. (courtesy NASA)|
ML can learn to identify individuals in camera footage, steer cars away from moving objects, detect planets around distant stars, or recognise clusters of health parameters that will help experts predict a disease.
And once they have learned their ‘trick’, they can apply that learning at lightning speed, without pause or getting tired.
Spreading insights instantaneously
Taking another scene 15, or so, years into the future. You’ve just been picked up by a self-driving car that covers part of your journey to a conference on, what else, AI.
It’s raining heavily and while the car picks up speed on the highway, it suddenly has to swerve to avoid a tree branch that has blown into its path. The vehicles next to and behind your car have to break and there is a short moment of chaos. This near-miss, although very rare, could certainly be a possibility. Overnight, the data of the vehicles involved are analysed and an update is sent to all cars worldwide on how to handle this situation in the future.
Of course, by then you’ve long reached your destination – unaware of how your journey influenced, even improved the driving behaviour of all cars way beyond those involved in the incident.
Humans can change their mind, adapt their behaviour to new circumstances and new learning but so too will intelligent agents, such as cars. And because the world of 2035 is tightly interconnected, the new knowledge can be spread to all intelligent agents almost simultaneously. So there’s no risk of colliding with a car that’s running on last year’s intelligence.
As has to be expected with such a pervasive technology, there are technical and ethical caveats.
One is the issue of explainable AI: if a critical system takes a decision, we humans should be able to track down its reasoning, to understand why the system did what it did.
Another issue is that machine learning is only as good as the data it is fed. Therefore, technologists are continuously on the lookout for biases that may pop up in behaviour of smart systems, or for biases that are added with malicious intent.
Examples are recognition or profiling on the basis of ethnicity or gender, or seeing as global what in effect are only local customs or behaviours, or even just temporary, commercial hypes.
And last there’s the concern that people should remain free in their choice to contribute or retract personal data, or to act upon the suggestions of AI systems.
Global but individualised
Of course in 2035 your clothes are made to fit to perfection. When you need a new pair of shoes, your local factory will consult your digital twin, deriving all possible parameters and produce a pair of shoes that are unique, but more importantly, costing no more than you used to pay for your average size 11 shoes.
But there’s more. You just bought and attached a sport’s sensor that’s now breaking in. Give it a few more hours with you, learning the very intimate relation between your blood pressure, heart beat, temperature and many more metrics and it will have become part of you, a sensor that matches up with no other person in the world but you.
The industry is no longer making a small range of average products. Instead, they are able to make separate, individual products for everyone. Like the good, old cobbler used to do for you as an individual, but now at the cost and speed of mass-manufacturing.
And some products even keep on changing and learning after you buy them.
It’s machine learning but no longer trained at the manufacturer’s with labelled input, but on your body with unlabelled data.
Budding AI wisdom
You’ve arrived at your holiday destination to find that your luggage has gone missing. You call the airline’s helpdesk and are put through to an operator, whose voice and body language are immediately comforting and reassuring. Within minutes, even while you are speaking, your luggage is located and an appointment is scheduled to have it delivered at your hotel the same evening. You full heartedly thank the operator, who smiles and wishes you a good holiday.
For a split second the thought registers that this was probably a bot, but by now you’ve become so used to being helped by imaginative, empathic bots that you’re rather pleased.
Machine learning is only apparent intelligence. ML systems still have to be trained by humans, who supply it with the training data and determine the questions to be solved. That makes for hugely useful systems, but not really intelligent ones.
But by 2035, we’re also seeing a first budding of really intelligent systems, systems that show some measure of reasoning, creativity, imagination, common sense, and above all empathy.
Leveraging its expertise in hard- and software, imec is setting up an ambitious AI program – together with industrial partners that are active in domains as diverse as personalised healthcare, smart mobility, the new manufacturing industry, smart cities and smart energy.
As for imec’s approach to bringing AI to the sensors at the edge of the Internet of Things (IoT) we are looking to introduce a pipeline of innovative hardware and software that – instead of using hundreds of watts – will consume less than a watt, or even mere milliwatts.
We will also be looking to develop machine learning applications that can be customised for specific uses and for individual people – on the spot, instead of with pre-learned parameters.
To that end the Flanders Government has earmarked a considerable sum to AI research, industrial application, and policy and imec has itself signed a collaboration agreement with the French R&D centre CEA-LETI to advance both AI and quantum computing.
Rudy Lauwereins is VP digital and user-centric solutions at imec