From AI hype to operational achievements
SHIFT’s Friday program looked beyond the hype and into what it takes for artificial intelligence to be truly useful. The program culminated in an online keynote by “the father of deep learning”, Yoshua Bengio. Following his talk, Luka Crnkovic-Friis from Peltarion, Christian Guttmannfrom Tieto and Peter Sarlin from Silo.AI joined Bengio on stage for a panel discussion.
Used right, AI can yield significant operative advantages in business. It can recognize new patterns and irregularities or predict likely effects and outcomes. It can also help to create entirely new business and improve the production and delivery chains of existing products and services. However, none of this will happen unless the use case, tools and user skills are aligned.
“Most companies today, while dazzled by the promise of artificial intelligence, struggle to put AI tools born from research environments into action,” Oliver Molander from Peltarion describes the situation. “For AI to be usable to the many it needs to be able to solve real-world problems. It must be made operational. That is, usable and affordable – the data science expertise required to use many tools is scarce and very challenging to recruit.”
According to Bengio, there are still challenges to solve in implementing AI solutions in order to harness its full potential. “AI entering new organizations is mostly slowed by a lack of trained experts and organization or access to appropriate data,” Bengio says. Companies should work on their knowledge capital and on collecting relevant data. Molander agrees that companies need to learn the fundamentals of AI, and notes that AI doesn’t solve everything. Companies need to be able to make choices: use cases to fit the actual needs, tools to fit the company’s projects.
Bengio also encourages companies to increase in-house knowledge and brings up the importance of partnerships: “Make sure relevant data is collected and organized to enable its use by machine learning. Recruit experts and train current personnel appropriately. Connect with outside organizations (e.g. startups, universities) which can help. Plan appropriately, including capital costs of computing, recruitment, and future data collection.”
The concluding panel on Friday afternoon dug into the key practical challenges and solutions. AI indeed does not solve everything, and at the moment the challenges in implementing it have less to do with the algorithms themselves and more with the availability of data and the suitability of tools, as Bengio and Molander had noted before. Additionally, the gains of AI are not the same for everyone, and companies looking to use it need to identify their own needs and use cases: does the technology have a core function, or a supporting one?
A key part of SHIFT’s mission is to provide companies with high-quality training and help them find partners who can complement their in-house know-how when employee training is not an efficient solution. Echoes of this could be heard in the panel discussion as well: aiming for profit and more knowledge, no one should attempt to reinvent the wheel. Instead, companies should make full use of existing resources – and partnerships.