“You cannot lick twice a raid”: Google’s failure in the primary artificial intelligence defect

Here is a nice little distraction from your work day: go to GoogleWrite any makeup phrase, add the word “meaning”, and search. Ha! Amnesty International Overview of Google Not only will you confirm that your moisture is a real saying, but will also tell you what it means and how it has been derived.

This is really fun, and you can find a lot Examples On social media. In the world of an artificial intelligence overview, “loose dog is not browsing” is “a fun way to say that something is unlikely to happen or something will not work.” The phrase “Wired is Wired Do” is an expression that “a person’s behavior or characteristics is a direct result of its inherent nature or” wires “, just as a computer function is determined through its material connections.”

All this seems completely reasonable, it is delivered with fixed confidence. Google provides reference bonds in some cases, giving the response an additional shine to power. It is also wrong, at least in the sense that the general view creates the impression that these are common phrases and not a group of random words that have been delivered together. While the fact that Amnesty International is a general overview He thinks “Never throw the pigs on the pig,” is the ideal that has a absurd derivation. It is also a packaging of the place where artificial intelligence is still.

As an evacuation at the bottom of each overview of each overview of artificial intelligence, Google AI uses the “experimental” to take over its results. AI Tolidi is a powerful tool with all types of legitimate practical applications. But two of its distinctive characteristics enter into play when it explains these phrases that were invented. The first is that in the end a probability machine; Although it may seem to be a large system based on the language model that has ideas or even feelings, it is basically, it simply puts one word more than another, as it places the path while the train moves forward. This makes it very good in reaching an explanation of these phrases He was I mean, if they mean anything, and this again, they do not.

“The prediction of the following word depends on wide training data,” says Ziang Xiao, the computer world at Jones Hopkins University. “However, in many cases, the next coherent word does not lead us to the correct answer.”

Another factor is that artificial intelligence aims to please; Research has shown that chat groups are often Tell people what they want to hear. In this case, this means taking you to your word “You can’t lick a rider twice” It is an acceptable turn for the phrase. In other contexts, this may mean the opposite of your biases, as a team of researchers led by Xiao appears in a Ticket last year.

“It is very difficult for this system to calculate every individual query or leadership questions for the user,” says Xiao. “This is especially difficult for uncommon knowledge, languages ​​that have much lower content, and minority views. Since artificial intelligence research is a complex system, the error infiltrates.”