Zero rule of robotics: “Robot must understand human.”

Zero rule of robotics: “Robot must understand human.”


How Siri knows that many languages

The rise of the machines is still far away, while scientists are making it closer with all their powers by working on Artificial Intelligence. However, sky upon us is still real and not programmed like in the “Matrix” (although who knows), so we can relax and play with the phone.

“Okay, Google. What’s up?” – I bend over to my shabby iPhone.

“Very funny, master…” – Offendedly answers Siri. But nevertheless, it answers in my language.

Apple's voice assistant in 2017 knows 21 languages localized with different dialects for 36 countries and regions. Siri is an exception, even among strongest competitors. Amazon's Alexa can only speak English, Google's Assistant speaks English and German on Google Pixel phones, and Microsoft's Cortana has learned eight languages in 13 different regions. So how does Siri mastered so many languages? 

In fact, all of that still relies on people. For Siri mastering a new language, Apple needs, first of all, to record speech fragments and contribute an accurate pronunciation of chosen phrases to the system. The next step is to enable dictation, allowing Siri-users to dictate the text they want to type out to their phones. Some of those dictated phrases remain for following usage (obviously, authors are kept anonymous). Then Apple invites interpreters to exclude the interpretation mistakes.

When the most common questions and answers have been recorded and added to the system, Apple releases a Siri update. And it continues to learn once it's in use, and only gets smarter.

Despite Siri’s learning abilities, it can’t be done without professionals. It has been showed in practice, that you couldn’t teach a robot through the chats with users only. Users are users: they laugh and teach some nonsense; then developers have to torment. An Artificial Intelligence isn't developed enough yet to identify good from evil by itself, filter out unnecessary things and stay within the taste and decency. Last year Microsoft launched a Twitter-bot named Tay, which supposed to chat with young people at the age of 18-24. The bot was a model of artificial intelligence; his aim was to test the learning algorithms. Within 24-hours the progressive youngsters turned the decent young bot into racist with a garbage mouth.

AI learning is far away from full automation, and it needed much more human involvement than it’s supposed to be in the future.

So, if you want your program or website to know many languages, you have to either develop an intellectual system for all occasions, build a big department, and ask for help over a couple of millions of users, or just go the traditional boring way and call an interpreter who translates what you really need. 


להירשם לקבלת חדשות שלנו