Can you tell us about your background?
"I have a master's degree in artificial intelligence from Université Laval, during which I did several internships. I am currently doing a doctorate in artificial intelligence, and more specifically a thesis on the automatic processing of natural language. In the case of Google Translate or Siri, language modelling is the automatic generation of textual content using the deep learning method or deep neural networks. In addition, I lecture deep neural network classes, and I give training at Université Laval’s Research Center to both private companies and the general public. I am also an AI Scientist at Vooban, where I have the chance to oversee application development within the artificial intelligence team. "
In simple words, can you explain what deep learning is?
“Deep learning uses the same neural networks as in the human brain. It is a complex mathematical function, an efficient way to model a set of data. In other words, we use a learning algorithm or a mathematical function to recognize objects in order to make decisions. Here’s a classic example of deep learning: on a given image, we want to identify a dog or a wolf. For that, we need a database; an assortment of images of dogs and wolves and an algorithm that will allow us to determine what the sets of pixels represent. Deep learning makes it possible to classify, identify and generate images or content, and to predict a numerical value.”
How long have we been talking about this set of machine learning methods?
People started talking about artificial intelligence in the 1950s. In the 1990s and 2000s we had a lot of expectations towards AI, but because of computational constraints and data complexity, it didn’t deliver the results we hoped for. Yann LeCun is one of the founding fathers of deep learning. He developed one of the first neural networks, but he encountered major computational problems. Humans weren’t able to ‘teach’ a machine."
What are other methods of machine learning?
Deep learning is one of the methods of machine learning. The latter is made up of several algorithms. We can divide them into two main aspects: supervised algorithms (which make a prediction, and if they are wrong, we will update our mathematical function) and unsupervised algorithms (we collect data that are similar and representative of subassemblies). There are more interpretable methods, such as decision trees and nearest neighbor search. With a decision tree, we will teach the algorithm certain rules, like true or false, and we will create a sequence of decisions so as to obtain a prediction. Nearest neighbor search cases are intuitive. In a database, if k = 5, I search for the 5 closest examples of k and identify clues of similarities, then I will be able to decide based on these 5 cases. Here’s an example: a patient comes to see his doctor, and based on his symptoms and using a database, we look for similar cases to make a diagnosis. "
Concretely, how can companies apply it?
In healthcare, for example, it is possible to detect skin cancer using images. A picture of the skin is taken, then an algorithm compares it to a data set composed of benign or malignant tumors to establish a diagnosis.
Is it only meant for large companies or can SMEs benefit from it? And in which sectors?
This makes me think of an aquaculture start-up business in Quebec City. They wanted to count microorganisms (fish eggs or microscopic shrimps) and calculate their size inside a bucket. Naturally, this is impossible to do with the naked eye because the microorganisms are tiny and constantly moving. With AI, we can take pictures of the contents of the bucket and the algorithm will be able to calculate the size and quantity of the microorganisms with incredible accuracy.
In a completely different area, there is Labplas, a manufacturing company of 85 employees located in Sainte-Julie. They are currently undergoing a major workforce problem that is slowing their growth. To counter this, Vooban is training AI algorithms that will reduce the number of operators required per production line and allow them to improve their quality control.
In fact, any company can make use of artificial intelligence to optimize their processes or to obtain recommendations based on data analysis. To do that, however, they need a good amount of data to feed the algorithm and make it powerful. Large companies often have a bigger data set. With the addition of a small AI model or algorithm, this allows us to improve processes and make decisions."
Why is Quebec becoming the new hub of artificial intelligence?
“The province of Quebec offers many types of training in AI, which makes our graduates very competent and coveted. In fact, we have the largest concentration of AI researchers in the world, with $ 1 billion invested in university research and more than 20 world-renowned professors. The Canadian government made major public investments, and Montreal was chosen as the headquarters of SCALE IA, an AI innovation supercluster unique to Canada. Many world leaders (such as Google, Facebook and Microsoft) have also chosen Montreal to develop their expertise in this area, in part because specialized labour is less expensive compared to Silicon Valley. Montreal not only attracts the big players of the industry, it is also one of the best AI incubators and accelerators in the world for a start-up company."
Will artificial intelligence overtake human intelligence? Will robots replace employees?
“To be honest, I don’t think we’re there yet. On the contrary, artificial intelligence will help some employees in their work and create jobs. We can compare this to the Industrial Revolution, with its arduous and repetitive assembly-line work.
A robot is much more constant and impartial than a human being. Devoid of emotions, it can make decisions according to data without any biases.”
What are the possible malicious uses and other abuses? (Deepfakes, for example)
“There will always be malicious uses that will result from new technological advances. Unfortunately, there is no way to counter this except through education and prevention. It is therefore important to teach people about fake news and abuses by making them understand how AI works.”
Can you explain machine learning? What is the difference between artificial intelligence, machine learning and deep learning?
“Machine learning regroups several types of algorithms according to the kind of problem to solve. As mentioned above, it’s composed of supervised and unsupervised learning. For a machine to learn, it needs data and an algorithm that will make decisions out of them.
Artificial intelligence is the evolution of expert systems. Before, humans analyzed data and programmed a computer or software. Nowadays, the computer learns from the data on its own, and humans are able to make decisions based on it.”
What are the professions that will disappear or drastically change with the democratization of artificial intelligence?
"All jobs that are very repetitive and methodical will be affected, especially those that do not require reflection, since computers only need data to complete their tasks. But don’t you worry. Robots are not capable of creativity. They can’t demonstrate humour, show empathy or any emotional intelligence whatsoever, so jobs in which there are human interaction probably won’t be replaced by artificial intelligence.”