Chinese technology giant Baidu apologized for misleading the public after announcing its accomplishment of besting rivals Google Inc. and Microsoft Corp. in the fierce competition of developing artificial intelligence-based technologies.
Baidu's mistake, however, is a sign that there is an increasing high stakes in the game of innovating AI technologies, such as allowing computers to recognize photographs, perform tasks previously requiring human attention, and understand spoken language.
Global players have been under pressure in recent years to grab leadership in the arena. Apart from the three tech giants, other powerhouses such as Apple Inc., International Business Machines Corp., and Facebook Inc. were also adamantly recruiting AI researchers.
Earlier in May, Baidu posted a record low of 4.58-percent error rate on the image recognition test of ImageNet. On the other hand, Microsoft had a 4.94-percent error rate and Google had 4.8 percent.
However, computer scientists who administered the test revealed on Tuesday that the firm has taken the test far more frequently than the normal, submitting 40 over a five-day period in March.
"This is pretty bad, and it is exactly why there is a held-out test set for the competition that is hosted on a separate server with limited access," Matthew Zeiler, AI firm Clarifai Inc.'s CEO, remarked.
"If you know the test set, then you can tweak your parameters of your model however you want to optimize the test set," he added.
Organizers have announced that they asked Baidu to cease from submitting ImageNet results for 2016.
Though Baidu has declined to give straight answers, its scientist Ren Wu posted a statement on the ImageNet website and apologized for the mistake.