The future of Big data: The robot world, or the human machine world?

Humans have never been satisfied with their cognitive abilities. Also because of this, photographic memory, Yimushihang, to know the geography of astronomy, has been seen as a model of human advanced version.

The computer has already done this.

Recently, at Alibaba's data opening days, I met with several data scientists. When they describe the future power of large data, I think of human limitations. The future of cognitive science is necessarily the perfect combination of computer and human. But what should this combination be?

More and more data, and human interpretation of the ability is fixed, people will be tired, will not be fully rational. But the computer doesn't. Computers can help people find their own blind spots. Bowen Zhou, chief engineer at IBM Watson's lab, told me that in Watson's medical program, humans were reading a ten-year paper, and that the computer would take only 30 minutes to read. Xu Ling, who worked for Axciom, told me that in the very early days, the data of two large American libraries and the Vatican Library had been completed.

Both of these things point to the conclusion that when we find the scientific and technological capabilities that are just enough to compensate for human shortcomings, there is a great deal of value in hiding. The key to implementation is data. Data enables computers and humans to communicate and combine.

Memory is not the most important thing to human beings, even the weakest link. Logic is the key to human cognition, and logic arises from the accumulation and deduction of experience. If computers can help people gain more experience, it will help to generate more powerful logic.

I used to like Evernote this application. It can record all my fragments of thinking, while collecting materials and articles. If one day, this software can be "machine learning" my record document, recommend to me the material worth reading, help me instantly search information, refining ideas, whether will change the cognitive habits of mankind?

Can the future be realized? At least for now, we've seen some progress, and the data is helping us see the blind spots. I have summed up the data to open the central point of view heroes, hope to be able to inspire you.

NO 1 Cognitive science may be a key ability in the future

The value of future data is not how much, but whether you have the ability to refine. In a health program that Watson is pushing, computers are able to determine the direction of cancer in the future by using large data, and to judge more accurate treatments for this.

How did we do it in the past? We have thousands of medical experts who have been reading and studying a great deal of literature, experimentation and trial and error, pushing the possible direction of the show. It takes ten years, even longer. It takes only 30 minutes for the computer to read all the relevant literature and quickly judge useful information. This is the use of cognitive science to save human health in the time and energy.

NO 2 red wine and sushi: Integrating "live data" with "historical data"

Integration of "live data" with "historical data" has become an industry consensus. Xu Ling This analogy, "historical data" like red wine, the more Chen the better. "Live data" is like sushi, the fresher it is, the better it is.

Now we have to deal with the data, we also need to face the "real-time data" and "historical data," the combination of problems. For example, if you are searching for tea in Taobao, should the recommendation system recommend all kinds of tea from historical data, or should you combine your current shopping cart data and recommend a more suitable brand?

The combination of data is not easy, not only requires different formats and standards of data unification, but also requires the combination of historical data and new data to generate the understanding and awareness of the present.

"It's like flying a plane, we're using historical data to judge the course, but the wind and climate are important for driving." The more sensitive real-time data is to you, the higher the value. Eventually you get the safest and most cost-effective results by setting the expected trajectory and adjusting in real time. The combination of real-time and historical data means better planning and quicker response. ”

NO3 large data also requires complementarity.

Aaron Ling, the chief engineer of the Ancestry company, painted a huge family pedigree on the forum. This is the project he is working on, with big data to clear the American Family atlas. People refer to the family pedigree through the Web site, and then modify or add data by UGC. This model could have a huge impact on future medical and social research.

The project proves that the results of large data-pushing shows sometimes require complementary human needs. Because through the interaction of human information, it is possible to produce more than human things. Human interaction can make big data more flexible and more real.

In practice, we might have a situation where you have a lot of data, but you still can't get a person's full network behavior information. So we need to use some algorithms to fill this "data loophole". Junlin Hu, a guest speaker from Samsung, described the confusion as: "If we think of people as one column, you will find that everyone actually doesn't buy much, and the connection between most goods and people is blank." ”

Therefore, when we use the data, we will find that there are many loopholes in the big data, and the data is too sparse.

In this case, we can only use the algorithm to solve, or "artificial operation" approach.

NO 4 Whether the future is AI world, or IA world

Bowen Zhou, head of technology at IBM's lab, mentioned a concept in the meeting, from AI to IA transfer. IBM has long been dedicated to creating robots that match human intelligence, such as "Deep Blue", such as "Watson". Now, they believe that the human-centric, powerful accessibility of computers may be more market-oriented.

The difference between AI and IA is that the former is computational-centric and the latter is human-centric. Is the future the "Robot World" or "The Human Machine World"? Is it a thought-driven robot or a human-assisted machine?

The future of Google glasses may allow us to see very far, immediately feedback 300 km away from the information, but he is not a robot.

Or people can have a strong manipulator, very powerful cars, airplanes. But they are centered on humans rather than machines. The difference between the two is worth thinking about.

Or we can ask ourselves this question: are we pursuing automation or intelligence?

Editor's note, this article's summary of views are:

Cheping: Vice president of large Data branch of China Information Association

Several experts involved in the article, followed by

Bowen ZHOU:IBM Watson Laboratory's chief engineer, in the field of statistical machine translation (statistical Machine TRANSLATION,SMT) has a deep accumulation;

Xu Ling (Elizabeth Xu): Vice president of Acxiom Group, a mature manager with rich experience;

Aaron Ling:ancestry company chief Engineer (Senior Director of UB);

Junlin Hu:samsung (Director of Data Mining).

(Responsible editor: Mengyishan)