How to phase artificial intelligence into companies?

Although the takeup of artificial intelligence is still in its infancy, there are already many companies who fall into the “early adopter” group with interesting use cases to tell. IDG Research Services sees AI as the next great disrupting factor in business, which cannot be put off or ignored.

For this reason IDG has put on AI Roadmap 2019, in which several top companies have explained how they have integrated AI and a panel of professionals has given its take on the current state of play and the sector’s future prospects. José Carlos Baquero, Big Data and AI Manager of GMV’s Secure e-Solutions sector, was one of these guest experts. IDG also took this chance to present the report "AI Machine Learning Roadmap 2019", drawn up to help companies set up their AI strategies.

AI is nothing new for GMV; for years now the company has been using cutting-edge technology to show up bank fraud, pinpoint threats and streamline industrial processes, as well as for predictive maintenance, client segmentation and scoring, knowledge management, conversational interactive voice response (IVR) technology, voice recognition and indexing, facial identification, training, technology consultancy, etc.

GMV advocates the co-creation model, which allows for tailor-made developments, combining, on the one hand, the client’s inside knowledge of its own business and problems with, on the other, GMV’s expertise and resources to connect precisely with the best use case. We at GMV help our clients to identify those use cases where AI could be applied, either to improve a process or product or even to create new products from data.

The advantage of automating the decision-making procedure is obvious. But it is also crucial to take into account the possible partiality deriving from algorithm biases inherent in the data itself; this situation could lead unknowingly to unfair, discriminatory decisions.

To tackle this problem we at GMV are analyzing different techniques in three phases of algorithm design: firstly, prior data processing, when the data per se already has biases; secondly, during training, using algorithm-penalizing techniques that preempt any learning from these biases; and, finally, in the post-processing procedure by means of thresholding.

It is vital for AI decisions to be seen as even-handed, not only to abide by regulations on this matter but also because society itself should demand responsible use that watches out for individual human rights. GMV is now rising to this new challenge and is already pulling off eyecatching results.