Look at artificial intelligence from Microsoft ice

Text / Wang Jin

"Little Ice" incident is still a variety of war of words in the dispute, but this only appeared briefly in the vision of the product once again intelligent on the cusp. From the small robots that appeared in the early days of QQ and MSN to the application of smart speech recognition after Siri appeared to Google's driverless cars, artificial intelligence, though not always in the spotlight, everybody believed that there were always some The team is quietly developing.

In fact, Siri was the main factor behind this topic again, and Steve Jobs regretfully passed away after the release of the iPhone4S product, enabling Siri to experience technology with the largest user base through the broadest market coverage The frontier.

However, the same situation as the day after Xiaobing appeared, the majority of users after the launch of Siri were never again opened again. Xiaobing this new generation of intelligent robots, some WeChat group at the beginning of some of the topics and new ideas also soon encounter the problem of decreased activity.

Although tremendous progress has been made from the first chat robot "Albert" to the current small ice, the fundamental reason for the drop in activity is still that its degree of intelligence does not have an alternative performance to daily life, so novelty Just for a moment In retrospect, Google has seen a much greater use of value in the continuing drive of unmanned and real-time translation systems from Microsoft.

Floating chat robot

According to Wikipedia, in 1950 Turing raised the classic Turing Test, which is where C (the inquirer) questions A and B to determine who is human and who is the machine. Unfortunately, as of today, robots that can successfully cheat people have not yet appeared. However, the big data process will greatly accelerate robotics training, and advances in speech recognition and semantic recognition will undoubtedly bring TURINs to testing faster in the real world.

In a variety of materials, the world's first chat robot "Albert" appeared in the 1980s, written by the BASIC language, beginning in 1991, Rhone Award, the annual artificial intelligence competition began to find that best at imitating humans Really dialogue scene robot. Until 2008, a computer program that convinced three of 12 strangers that "it was human" took the dream one step further.

At present, in some open source communities, the underlying core program of chatting robots is not hard to find, and its operating principles are generally divided into training, matching and other aspects. The factor that makes training easier is the cookie data on the internet.

For example, "Little Yellow Chicken", an extremely well-known program on everyone's network, uses a large number of cookies to acquire user habits and language inertia, translates the predetermined thesaurus into conversations and people, and the official version of Little Yellow Chicken once sold It's expensive because of its personality match to a single user.

In the commercial arena, MSN robots, small i robots have all tried to make smart conversations the core competencies of products, but the evolution of core algorithms can not be separated from huge databases. As can be seen from the existing relatively mature robot, search engine is the cradle of birth dialogue robot. In particular, Google Now Google for Siri launched, by learning the user habits combined with the search engine's massive database gives an accurate answer, it appeared for Siri anxious.

Microsoft this time out of the small ice is a similar product, but compared with Google Now, ice wearing a layer of Meng sister's coat. What really makes a big leap forward in technology is Cortana, a voice assistant under the WP system.

Dream into the reality gimmick above the ideal

Although various analyzes believe that the root cause of WeChat's blockage of small ice from WeChat's unwillingness to be taken away by mobile robots is that such a foundation of analysis requires the premise that "small ice" has a long-lasting value to use, and the result Did not give it a chance to prove it.

However, from the micro-channel group to see the user feedback from the view, after the robot entered a period of time, the user's freshness dropped quickly. So take the entrance of such a statement is difficult to gain a foothold.

And from Siri, Google Now to Cortana, such products are essentially going to be the first to break out of usability rather than the humane process, and first of all there is a continuing rationale for providing information that has general validity.

For example, the intelligence of a small ice may not even be as smart as the input method of Apple iOS 8. At WWDC 2014's DEMO, when asked whether to go to dinner or watch a movie at night, the smart association's response is "eat" and "watch movies" and "do not go", which has already broken the threshold of semantic recognition and feedback . Practicality is far greater than selling cute robot.

A further challenge for chat robots is in smart mock-ups, and the big data environment of today offers yet another hotbed.

In the big data environment, the original algorithm suddenly become intelligent. Including the recent stir-fried smart news recommendations, smart medical data, simple algorithms embrace the great data, the progress is evident. However, the ideal is still far away. There is still a vast research space for the calculation of artificial life, spiritual network and uncertainty.

Artificial Intelligence Falling Crash Philosophy and Sociology

Still remember the film "Terminator" and "Matrix" it? Even if it is realized in the foreseeable future, people's vision of the future is already far-reaching. This leads to another topic that artificial intelligence also involves the category of philosophy and sociology. Even if it is too far away from technology, it is now in sight Google's smart car technology, the biggest obstacle to its popularity is not a technical problem.

A few days ago, after the unmanned aerial vehicle was unveiled, a simple survey of "Driverless Cars Are You Done?" Answered very few people would like to try. The reason is simple, but the letter.

People are willing to trust people's uncertain ability and do not want to accept the simple 1% error of algorithms and procedures. The reason lies in the fact that even if they grasp the error rate more, people are willing to bear the consequences. Such a problem is not a technical one, but a psychological and philosophical one.