Alexander Linden, Research VP specializing in data science, machine learning and advanced algorithms, tackled the trendy Artificial Intelligence topic on November 5th, during the 2018 edition of the Gartner Symposium/ITXPO which took place in Barcelona. He notably focused on the opportunities and challenges linked to the advent of AI.
"The definition of AI and therefore its impact, depends on the perspective," first explained Alex Linden, who continued: "CEOs think it's the #1 tech today, when it's actually many technologies together. Why do they think so? Because it means more opportunities and things that were unfeasible using traditional engineering now becoming a reality". As a matter of fact, the AI landscape of applications goes from Automated control and IoT, to pattern recognition, to customer engagement. He then insisted on the importance of having an approach of collaboration between human and machines, developing the concept of augmented intelligence.
From Artificial Intelligence to Machine and Deep Learning
The concept of AI – machine capabilities that solve increasingly complex tasks – must also be distinguished from the technical disciplines that are Machine Learning, Deep Learning, Data Science, etc, which aims at solving business problems through the extraction of knowledge from data. "In this respect, companies have to deploy what they call AI in a data-driven way, by using Machine Learning and Deep Learning, notably," highlighted Alex Linden. The expert then explained that today's AI solutions are point solutions, with a narrow scope: therefore, when most companies think that they are behind on AI, they are actually not. "Currently, 99% of all AI success is based on supervised learning, which is a branch of ML," added the Gartner analyst. The core of ML is about the creation of mappings from data and can answer to several challenges: loan application, demand prediction, self-driving cars, propensity to buy, failure prediction, customer churn, medical diagnosis, advertisement, etc. According to Alex Linden, "ML is not really rocket science. It starts with data points and finds a model which comes up with a reasonable result for unknown data points. Add neural nets and you can now solve more complex problems". The last 5 years have seen extreme progress with error rates, notably when it comes to image recognition, dropping drastically resulting in Machine Learning being more precise then humans, since 2017. Yet, most "AI" successes rest on the shoulders of few algorithms. Nowadays, in addition to evaluating and classifying, AI becomes generative as it is now able to create content. For instance, you upload a picture of a city, AI can turn it into a Van Gogh painting look alike.
Corporate applications of Deep Learning
Alex Linden then listed some of the many corporate applications of Deep Learning. He started: "When it comes to fraud detection, Deep Learning turned out to be a wonderful tool, combining data and tech to detect specific events, make sense of them and therefore detect potential fraud," explained the Gartner analyst. In the cybersecurity space, DL and AI help make sense of the slightest pieces of information by collecting and putting them all together. Quality can be affected by all kinds of elements during the manufacturing process. What was the reason for misconception or other quality problems? Here, technology can help by analyzing data and events. AI and DL also mean automation and therefore factory control, machine control, tested and done. "Other domains will be impacted and augmented with AI and Deep Learning: advertisement, product recommendation, Predicting outcomes of experiments, disease detections, predictive maintenance and many more," underlined Alex Linden.
Pitfalls and complexity
These opportunities also mean several challenges and pitfalls, the first one being keeping up with innovation and navigating through the tens of buzzwords to find a way and application to actually begin with: "How to pick the best solution approach? Use a decision framework and ask yourself the following questions: is there a common use case that is appropriate or a commoditized solution? If yes, buy it. If not, do you have the in-house capabilities to build it? Do you have a data science team? Etc". This framework can be referred to as the BUY, BUILD or OUTSOURCE framework.
Another complexity is the huge skills shortage in Data Scientists which need to develop several skills: Quant, IT, soft skills and of course business knowledge. "On the other hand, we are witnessing the rise of citizen data scientists as tools are getting easier to use. Yet, expert data scientists will remain rare," added Alex Linden, who therefore advocated the upskilling of collaborators, which starts by mapping tasks against skills, against roles.
Then, the AI expert highlighted the fact that Machine Learning is…Alien, meaning that it involves and requires innovation, investigation, advanced prototyping, but also business execution…and firefighting.
The Gartner analyst ended his presentation by re-listing some of the key benefits of leading AI projects which aim at improving marketing and sales (customer acquisition, customer preference, etc), services support (better resource allocation, automation), products and production, HR, etc. "Use the funnel of the normal project portfolio management: there's no special treatment for AI. Begin with ideation, project management, then ROI estimation to justify the project, prototyping, etc. Finally, do not forget to continuously improve the development and maintain an improvement loop," concluded Alex Linden.