Swiss Info recently published an article on a Swiss Idiap Research Institute team’s insights on some of the many illusions of artificial intelligence.

The term “artificial intelligence” itself is subject to scepticism, especially when it comes to how it is used today. Many argue that there is in fact no such thing as “artificial intelligence” since no system can nor will ever be able to reflect human intelligence or common sense.

Although hesitations with AI exist, over the past decade it has become established across a multitude of business sectors and is “increasingly contributing to decision-making processes”. In fact, Bourlard, who took over Idiap in 1996, suggests that there are three aspects to AI that make it unique and powerful: (1) computing power; (2) mathematical models; and (3) vast and ubiquitous databases.

Artificial intelligence is able to produce powerful computations and draw insights better than traditional mathematical methods. AI is able to detect and predict patterns much better than a human or multi factorial regression. However, these computations and insights require a vast volume of data that must be clearly labelled and organized by a human to enable a machine to make sense of it. Thus, “the limits of machines depend on the limits of data”. In this sense, Liebling, head of Idiap’s Computational Bioimaging Group, brings to light a solid argument that “the threat lies in the way the data is managed” as opposed to the impending world take-over by “sci-fi machines”.

Consequently, the massive amounts of data fed to giant tech companies like Facebook are what worry the scientific community. Resultantly, the AI community is now pushing to develop “ethical, explainable, safe, and fair models” that can explain their inferences and how they reach certain conclusions. In turning complex algorithms into something accessible and understanding to the external world, AI becomes much less intimidating. Additionally, on-going discussions surrounding AI have highlighted how poor our traditional methods are of making sense of the world and drawing predictions from conventional knowledge in economics, mathematics, and in the science of open systems.

“Machine learning,” as Bourlard prefers it, or “augmented intelligence” are better-suited to describe the current technological era as opposed to “artificial intelligence” as there is very little base intelligence being used. Computers use “neural network models” inspired by the human brain and are only as powerful as the data they are running on – the black box is just that. Computer intelligence as the term “AI” lends is possible, but needs improvements in the hardware, software, and mathematics behind the operations.

Read the full article from Swiss Info.