How intelligent is Artificial Intelligence? Right now, you might think that AI is going to turn the world upside down, make everything much more efficient, and soon take jobs away from people – from analyzing x-rays to driving cars to automated service calls.
However, in the past weeks, there are increasing signs that AI is not yet ready, and in most cases – people are merely fooled by the hype word. A new study by the Technical University of Berlin (TU Berlin), the Fraunhofer Heinrich Hertz Institute (HHI) and the Singapore University of Technology and Design concludes that about 50 percent of today’s AI systems cheat. Every second supposed AI model would use only a few smart solution strategies.
“Smart Hans” instead of real AI
“We were very surprised by just how broad the range of learned problem-solving strategies is. Even modern AI systems have not always found a meaningful approach, at least not from a human perspective, instead sometimes adopting what we call ‘Clever Hans’ strategies,” says Wojciech Samek, group leader at the Fraunhofer HHI. The term “Clever Hans” dates back to around 1900 when a horse that could supposedly count and do sums became a scientific sensation. It later turned out that in about 90% of the cases, the horse guessed the answers based on the questioner’s reaction and no other sensational reasons.
Scientists in various AI systems discovered such “clever Hans” solution strategies. The researchers give an example:
“An example is provided by an AI system, which some years ago won several international image classification competitions pursuing a strategy that can be considered naive from a human perspective: It classified images primarily by context. Images featuring much water, for example, were assigned to the category of “ship.” Other images featuring rails were classified as “train.” While the copyright watermark correctly classified further images. This AI system did not solve the task it was set, namely to identify ships or trains, even if it did ultimately correctly classify the majority of images.”
It could be dangerous
Such AI systems are utterly useless in practice. Indeed their use in medical diagnostics or security-critical areas entails enormous dangers,” Klaus-Robert Müller warns. According to him, it is quite conceivable that as many as half of the AI systems currently in use implicitly or explicitly rely on such “Clever Hans” strategies. Safety and security should be major topics. In particular, in medical diagnostics or safety-critical systems, such supposed AI developers should better be avoided.
The marketing concept is misleading
The startup world is also handling with the term AI very generously. An analysis by London-based venture capital firm MMC Ventures concludes that around 40% of start-up companies, which run as AI companies have nothing to do with Artificial Intelligence. The crux of the story: You often get risk capital more easily because investors are also so excited by Artificial Intelligence.
Alas, AI has unfortunately degenerated into a marketing term, which has nothing to do with real artificial intelligence. “Because there is no precise definition of what AI is, the demarcation is equally subjective. Some startups actually work a lot with artificial intelligence, and some companies use the term AI for marketing purposes,” says Rasmus Rothe, founder of startup Merantix and co-initiator of the German KI Bundesverband, in an interview with Trending Topics.