“Artificial Intelligence” has become a trendy buzzword over the last couple of years. Although as shown in my previous articles, the concept is not new or disruptive, as with every embedded technology, it is sometimes hard to measure if and how the technology concerns users. Public belief about AI seems to be that “AI does not concern me”.
According to a report from IHS Markit published in 2019, AI technologies are affecting all sectors of activity, from industry to healthcare, through to consumer goods and the automotive sector. The overall worldwide spending on AI systems reached $1.5T in 2019.
This article will go behind the numbers, to illustrate if and how you are exposed to AI, providing concrete uses of the technology in common products.
AI and the Internet
You are likely to own a smart phone. And, like most owners, you are probably not using it for its original function, which is to make phone calls, but as a text messaging device. There is a high probably that you love the text prediction capability it offers, which accelerates your writing speed.
However, have you ever thought how your phone “knows” which word is likely to come after the one you just typed? How is it possible for this machine to “know” what you have in mind?
This is thanks to AI!
The texting engine has been trained from many existing texts, fed with thousands of words, sentences, texts and books. This training means that your smart phone is able to correct your spelling and grammar, apparently automatically.
There is still more! The phone also has an embedded machine-learning algorithm that trains the smart phone with your own writing, becoming more and more accurate as you increase your use of the phone!
You are also likely to use online services and e-commerce. I am sure you have noticed that your preferred device – whether it is a PC, a tablet or a smart phone – will suggest you some extra products to buy / actions to perform. Although I agree this might become a bit invasive and annoying, we should admit that sometimes, its recommendations are relevant. This is because every modern website embeds some kind of machine Intelligence that continuously analyzes your data and behavior, as it does for the millions of other users. It is then able to profile typical behaviors, to categorize your behavior and finally to recommend actions based on the category you belong to… It is hard to find out that we are not unique, isn’t it? It is harder to discover that “the internet” records / analyzes / categorizes every action we make, although this data is not being processed individually. At least officially…
Still in the e-commerce domain, there is a whole auction system running in every browser to buy and sell advertisement banners. AI is running in the background to determine potential advertisers and agree on the price. This is a matter of milliseconds, and sometimes prices are going to the 4th decimal of cents… However, it is a considerable source of revenue for every website.
Kelkoo is a pioneer in this domain, and most of its R&D on the subject is ran out of Grenoble.
Not mentioning google, that founded its empire on google Ads, google Adsense, and google Analytics…
AI in the financial sector
AI is also applied to the banking industry. When you decide to invest in a financial product in Europe, it is mandatory for the seller to determine your investor profile to offer you products that are compatible with the level of risk you can tolerate. This step is still manual, based on a questionnaire. So, we are still relying on human interactions and intelligence!
However, AI measures the risk of financial products, making heavy use of statistical analysis and predictions: statistical analysis to determine what occurred in the past and machine predictions to anticipate what may take place in the future. Such analyses are very complex since they mix economics, politics and geopolitics to determine risks. Raise Partner is a company based in Grenoble that typically illustrates this use.
Still in the financial domain but with a different angle, it is worth noting Quantics Technologies, a start-up based in Lyon, which makes use of machine learning to automate decision-making for buying and selling on the stock market.
AI in the Industry
Detecting failures in many domains is possible with the multiplication of probes, and most of the time a matter of (big) data management. However, when it comes to complex systems like IT infrastructures that are typically composed of hundreds of components and thousands of instances, it is a mammoth task to be able to identify the root cause for a failure. In many cases, the observation is not the cause of the failure, and the complexity of the system is such that human intelligence is not sufficient to dig into the data effectively. Root cause analysis may takes hours – if not days – and then the problem has to be fixed. AI is of great help in detecting failures, allowing for the identification of the root cause in milliseconds, and problem resolution within minutes, relying, of course, on the spare parts available… Even on stock management, AI has entered the game, predicting failure before it occurs, and even ordering the parts to be delivered to a customer’s premises in advance… A small and medium enterprise from the Minalogic ecosystem, CoservIT, offers exactly this type of solutions to IT managers. Collaborating with AI experts from the local university allowed them to dramatically enhance their solutions, differentiate their product from their competition, and thus win market shares…
AI in Consumer products:
As connected end users, we have seen an explosion of connected devices that are supposedly helping us live better, by providing various recommendations or information about what we do, what we eat, how to dress… Vocal assistants (Alexa, Hey Google, Cortona, Siri, …) are now routinely part of our access to information. Guess what? Speech recognition just neural networks combined with some deep learning applied to Natural Language Processing.
Our kids are playing with companion robots, while the elder generation is extending their time in their own homes, thanks to care robots. Such robots are full of sensors which interact with their environment, collecting data which is analyzed by AI algorithms in order to decide on the best possible action.
Hoomano is a start-up from Rhône-Alps that specializes in man-machine interactions; they are convinced that it is a mistake to duplicate the look and feel of humans when designing a robot, and that its function should guide its design principles. Making a robot look like a human creates expectations as people will want to interact with the robot as they would expect to with a human. However, interacting with a task-oriented robot, designed for its purpose, sets expectations at the proper level.
AI in Transportation
AI also has several applications in the transportation domain.
I will put aside autonomous vehicules for the time-being, for the simple reason that, while being AI-techno intensive, you may not consider autonomous vehicules as being widely available, and therefore may not consider yourself as being concerned… I will not argue this point, but there is actually a good chance that the car you own already has some embedded intelligence to assist you in your driving (GPS and trip optimization, adaptive speed control (traffic, limitations, lane positionning assitance), parking assistance, …)
However, AI is widely used by the transportation industry, particularily in solving the “traveling salesman problem”, one of the most famous problems in computer science. The problem consists of finding the most optimal route for a salesperson having to visit a number of customers in different locations. Despite its simplicity, as of today, there is still no determinstic algorithm to solve it Although initially expressed for salesmen, the problem applies to a number of different use cases: waste collection according to the fullness of bins, etc…
Using brute force – evaluating all possible solutions and choosing the best – is not an option due to the complex nature of the problem (factorial order of magnitude). Indeed, 71 different destinations lead to more than 5 × 1080 possible solutions, which is approximately the number of atoms in the entire universe… Brute force does can not be applied even with a small number of destinations, because it leads to endless computations which are incompatible with the dynamicity of our use cases.
AI proposes alternatives to not necessarily find the best route, but – at least – to approximate it.
For example, NeoVision, a consulting company and expert in AI, helped one of its customers, a school bus operator, to optimize buses routes, reducing its daily milage by one third, without compromizing the service to schoolchildren.
AI is also widely used in logistics. Every letter and every parcel is dispatched automatically according to zip codes in postal sorting centers by sorting machines. Solystic, a subsidiary of Northrop Grumman, based in Valence, is amongst the top 3 leaders of logitics solutions in the world. Besides sorting machines, Solystic also provides autonomous mobile robots to move letters and parcels. All of their products make great use of AI to recognize handwritten countries and zip codes, and navigation algorithms for the robots to move and collaborate with humans in sorting centers.
Similarily, Meanwhile is a young start-up that builds automotous robots for various environments (Industrial, medical, open to public…). Their differentiation is that they put collaboration between robots and humans at the center of their design, inventing a new category: collaborative robotic or cobotic. Their robots have a number of embedded sensors in order to “feel” their environment, and a lot of AI-based algorithms to make sure the robots behave properly without humans.
AI for Energy andHome
We all live in houses, grouped in cities and our daily lives are are based on energy consumption.
The way we produce and consume energy is key to sustaining the planet’s resources, and AI technology is once more part of our daily routine. Many algorithms predict our energy consumption, optimize the buildings we live and work in, and make the energy mix between fossil and renewables sources possible.
Smart Grids for example are a revolution in the energy domain that have been possible thanks to AI.
As an illustration, I can cite Odit-e, a start-up near Grenoble, that provides a solution dedicated to utility providers to monitor, analyze and recommend potential corrective actions to optimize their electricity networks.
In industry, AI is more widely used to anticipate when to intervene on machines. This is referred to as “predictive maintenance”. Without predictive maintenance, technicians can either change parts on machines before they become unusable, or when failure occurs. In both cases, this approach is not optimal. Predictive maintenance aims at anticipating when to stop the machine before failure, optimizing the parts’ lifespans. This prediction is made by observing several months of operations. This is the approach taken by D-Analyse Signal for example, a SME in Roanne. However, when it is not possible to gather enough data, other companies are using Bayesian algorithms derived from game theory. This technique relies on getting enough accurate data from fewer samples. This is what ProBayes, now part of Groupe La Poste, does. In general, such approaches rely heavily on a fleet of sensors to acquire and combine various data from industrial machines. Another interesting solution, from Adeunis, consists of deploying microphones instead of sensors. Thanks to some advanced sound processing, Adeunis’ solution is able to detect faulty behavior on machines, based on subtle differences in their functioning sounds. This technique found its roots by observing how experienced technicians perform machine maintenance and is particularly cost-effective.
AI in the Military sector
The military sector also uses of AI in its products. There are many applications for Artificial Intelligence in this domain, and we are more or less exposed to civil solutions.
Video surveillance is probably the most common civil application. Digital cameras produce so many images nowadays that there are not enough human eyes to watch them. AI allows us to automatically review them, seeking for and detecting abnormalities such as traffic congestion, suspicious individual behavior in crowds, etc…
Similar techniques are also used to control access, based on facial recognition for example.
In some countries, we have seen some autonomous robots assisting police in maintaining law and order.
The use of augmented soldiers is also a vast domain of application for AI technologies, from target tracking to observation drones…
Active actors are large corporations, like Safran or Thales in France, or defense departments, like the DGA. All these actors are permanently seeking innovative technologies and are conscious that their market is not sustainable solely for military applications. They pay special attention to dual (civil + military) use of the technologies they rely on, to ensure the longevity of their solutions.
AI in Health
Intuitively, we sense that AI probably has a tremendous number of applications in health: AI has the ability to digest billions of diagnostic data every millisecond, and not a single doctor on earth could accumulate as much experience as machines are able to, maybe except Dr House, who, given his social inability, we may legitimately doubt he is human… So theoretically, a machine could establish the perfect diagnosis, including very rare diseases, providing they are able to carry out a good symptom analysis.
However, the patient/caregiver relationship is so built around non-verbal information that, as of today, humans are still ahead!
However, this does not mean that AI technology is not present in the health sector: medical image interpretation, robot-assisted surgery, personalized treatment are all solutions with embedded AI technology.
We have all anticipated, and hoped, that AI would help us anticipate, fight and eradicate pandemics. Unfortunately, technology and AI do not provide a solution for every problem on planet Earth, and it is now clear that humanity will have to rely on itself to find solutions to solve a crisis like CoVid#19… This being said, there is obviously some disappointment about the (in)capacities of today’s AI.
Myth or Reality?
So to answer our initial popular thinking: it is a strong NO: AI is definitely not SciFi and most of us are interacting with AI multiple times on a daily basis.
Our myth is clearly “Busted”! AI does have a strong impact our daily lives, sometimes with us being unconscious of this impact. It is up to everyone to decide whether this “intrusion” is good or not
I would like to thank warmly Jean-Eric Michallet who supported me in writing this series of articles, and Kate Margetts who took time to read and correct the articles.
I would also like to give credit to the following people, who inspired me directly or indirectly:
- Patrick GROS – INRIA Rhône-Alpes director
- Bertrand Brauschweig – INRIA AI white book coordinator
- Patrick Albert – AI Vet’ Bureau at AFIA
- Julien Mairal – INRIA Grenoble
- Eric Gaussier – LIG Director & President of MIAI