Jean-François Puget, a machine learning and optimisation expert with over 25 years’ experience, said that he believes the human factor is not discussed enough when the media reports on machine learning, AI systems and deep learning.
The IBM distinguished engineer, who carries this special title for those who have outstanding technical achievements, said he felt the human context is necessary to derive the most value from the automation of processes. He said: “With these technologies, you can build a system that delivers value. But it’s not magic,” adding that human reactions and experiences must be taken into account.
"Machine learning and AI is used to help humans make better decisions."
He said: “Even with a good platform like IBM Data Science Experience, it’s not enough because very often machine learning and artificial intelligence is used to help humans make better decisions and to guide them. And this is where the problems can arise.”
Puget used the case study of an IBM partner that worked with a global hotel chain wanting to improve its room pricing system. It is well-known that most hotels operate an elastic pricing structure, with room prices varying according to the date of booking and proximity to the date of room occupation, as well as demand for the rooms. He added that the room prices are set to maximise the margins for the hotel and also to please its frequent customers.
"Those young kids called data scientists won't teach me my job."
Puget said that IBM’s partner did a good job and developed a pilot that was able to improve the hotel company’s margins by more than 2%. However, the pilot did not get a rapturous reception all round. “When you develop an application with this return on investment, you would think that people would adopt it overnight,” said Puget. In reality, he said the new application was met with resistance from managers around the globe, who all said the same thing. ‘I know what I’m doing. I’ve been doing this for 20 years. Those young kids called data scientists will not teach me how to do my job.’
The IBM partner had to convince the managers that theirs was a better way of doing things, and did so through gamification. They created a simulated hotel as well as a simulated competitor hotel. The managers were asked to set a room price and compete against the automated pricing system. “The automated pricing system was consistently beating the managers by a few per cent margin. After they tried several times to beat the system, they trusted it, they saw the value. And then the adoption began,” said Puget.
"Make sure that users trust and love the system instead of fighting it."
He said that the lesson is, when one starts an AI or a machine learning project, one must think about how to go from data to models and from models to a production environment, and even more so when humans are involved. “You must include how humans will react from the start,” he said. “You need to make sure that users will trust and love the system instead of fighting it. Think about how to make people adopt what you’re building from the start and not as an afterthought.” With this example he drove home the point that the human context should be not just considered but embraced when aiming to build better and smarter data systems.