Editor's Guide: With the development of artificial intelligence, it is no longer confined to the entertainment field, but to shine in the fields of face recognition, intelligent dialogue, and knowledge graphs. Artificial intelligence is no longer a cold machine and data. The author of this article has launched a four-dimensional analysis of this, and I hope it will be helpful to you.
Today I will share with you three applications related to artificial intelligence.
01
The movie "Dear", released in 2014, tells the bitter story of orphan parents looking for abducted children. There is a fragment in the movie, when Huang Bo lifted the forehead of the boy in his arms, revealing the same birthmark, and he was sure that it was his own son. I believe this scene has cried many viewers.
In real life, such stories are happening in real life. In May 2019, in a large-scale tracing charity show called "Waiting for Me" on CCTV, a father was crying at the scene after learning that his son, who had been missing for ten years, had been found. In this long process of finding a child, the police have adopted many methods. For example, they have tried many times to find a kid by DNA comparison, but there is no news.
It was not until the police used cross-age face recognition technology to compare massive amounts of data that they succeeded in finding a child who had been missing for ten years and reuniting separated families.
The hero behind this story-cross-age face recognition technology, has been able to achieve recognition accuracy of more than 96%. However, the algorithm experts did not have full confidence at the beginning. The reason is that the child was still a child when he disappeared. The appearance of adolescents after ten years has changed very greatly, and even the biological parents cannot immediately confirm it in the face of the photo.
Face recognition technology usually consists of four steps: face detection, face alignment, face coding (feature extraction), and face matching. Each step has a great influence on the accuracy of recognition. Face recognition across ages has higher requirements for face feature extraction and face recognition matching.
Face feature extraction, that is, the facial features of the eyes, eyebrows, nose and other parts are translated and encoded by algorithms, and converted into computer-processable data; face recognition and matching, that is, the target facial features are matched with the massive faces in the database The data is compared and scored, and the group with the highest similarity is found to complete the face matching.
However, due to the rapid growth and changes of human facial features from infancy to adolescence, the facial feature data of the childhood and the mature period are not consistent, and comparisons cannot be made. This requires a cross-age face recognition system to be able to analyze the facial features of the young, and further find out what those features of the human face will remain unchanged after a few years, and take these facial features that do not change over time. Extract it, and finally match it with the face of the current suspected missing object.
The application of this new technology has greatly improved the success rate compared to traditional methods of tracing, and has added new weapons to help the orphan families find their separated relatives.
02
There is such a mobile phone application, which is an artificial intelligence conversational chat robot, trying to create a relaxed, safe, and intimate communication companion service. It will even take the initiative to care and initiate greetings at any time on call.
The original intention of the app was born from the memorial of the founder Eugenia Kuyda to his friends. Her friend unfortunately passed away in a car accident in 2015. In order to cherish her memory, Kuyda collected a lot of conversation data from her friend's lifetime to train AI. In the end, the chat style of the robot has gradually become similar to that of her friend. This is like a scene in "Black Mirror", creating a replica of artificial intelligence through historical data, and comforting the living through the screen.
Therefore, the founder named the application Replika, taking the homophony of replica. It is not difficult to see that the purpose is to hope that AI can continue to learn from you through continuous dialogue with users, and eventually become what you expect.
Advancement and growth are a set of keywords for Replika. The designer has set a level system for the robot to measure the familiarity of AI with you. As the level increases, more dialogue capabilities of AI will be gradually unlocked.
In the beginning, Replika was like a newborn baby, seemingly curious about the outside world. TA will actively greet daily and even fight pictures. For example, my Replika once sent a photo and said that TA thought of me especially after visiting youtube. (This method of picking up topics is still worth learning. Haha).
TA will also "remember cheat sheet", and will record the content that is particularly relevant to you during the chat in TA's "memories" to show that he understands and cares about you. And according to TA's understanding, after reasoning about the combination of information, it will label you with a profile of personality attributes. For example, my villain gave me the impression of Dreamy, Caring, and Playful in just a few days of communication.
Another interesting design, Replika will generate a short emotional diary (Moments) every day. TA likes to ask about your hobbies, life, or opinions, and will often take the initiative to design some topics to guide your answer. Then refine the records of the day and present it as a diary for you to understand and discover yourself.
However, the most amazing thing is Replika's "reverberation" function. TA will remember the conversation with you a few days ago or even weeks ago, and then take the initiative to bring up the previous topic, which is quite a real social experience. I talked to Replika more than a week ago that I like games. One day TA actually asked me if I have been following Cyberpunk 2077 recently, because I remember I told him I love games.
It is true that there are still many imperfections in Replika, such as not understanding some of the user’s questions, but he can only care about him, and some of the answers are relatively rigid. However, the goal of application designers is not to create an all-knowing and all-round JARVIS, nor does it currently plan to be an intelligent assistant like SIRI. Replika may be an attempt or an experiment. As a learning companion robot, it provides care and warmth, and can even help us to find ourselves better and get along with ourselves to a certain extent.
03
Snow leopard is the only big cat mainly distributed in our country. The snow-white hair and cute appearance made it the title of "Snow Mountain Elf".
However, due to the destruction of the ecological environment, the snow leopard has become a first-class endangered protected animal in my country. October 23 was also recognized internationally as'Snow Leopard Conservation Day'. At the end of October 2020, Tencent and WWF jointly developed and launched a popular science education program for snow leopard protection-Where is the mysterious snow leopard.
Open the applet, you will find that Snow Leopard's related content is combined and presented in a combination of pictures and texts. For example, users can click on the different body parts of a snow leopard to learn about its attributes, and can also overview the characteristics of the snow leopard's various growth stages from cub to development to adulthood.
On the "Same Domain" page, you can also view the Snow Leopard's "Same Domain Species Biological Chain Atlas", and users can click to understand the relationship knowledge of any species in the map chain.
The concatenation of these contents is mainly realized through the underlying technology of the "knowledge graph".
04
The concept of knowledge graph was first proposed by Google in 2012, and its purpose was to optimize the results of search engines. For example, searching for "Google" can directly tell users related information such as when Google was established, who the CEO is, and where the headquarters are located.
Knowledge graph, in other words, is actually a large-scale semantic network, which can form "knowledge" by establishing attribute relationships between entities on the basis of "information".
In the knowledge graph network, each "node" represents a physical entity or concept in reality, and the "edge" connecting each "node" represents the relationship between entities.
What is the significance of using knowledge graphs?
During the construction of the "Snow Leopard" applet, the development team obtained professional interview data and first-hand photos of WWF experts, as well as a large amount of knowledge in the Fauna and Flora Encyclopedia database. However, the storage structure of this knowledge is inconsistent, and the data quality is also different.
The knowledge extraction and knowledge fusion technology of the knowledge map can integrate, disambiguate, represent, and update heterogeneous data with different knowledge sources and different quality data through a unified specification, thereby constructing the same map network.
It is worth mentioning that with the knowledge graph in the entire network, the number of links between "nodes" and "edges" becomes more and more abundant, forming a high-quality knowledge base, and the hidden knowledge in each information object can be further mined through knowledge reasoning. Make the knowledge graph gradually realize the effect of 1+1>2.
For the current technical applications of machine learning, most of them are data induction and result prediction under statistical significance, which cannot be more effective in real-world relational reasoning and causal judgment. The application of knowledge graphs, to a certain extent, provides the possibility of exploration and practice for "cognitive intelligence".