Facing the development of artificial intelligence, where should human beings go?

Editor's note: AI is constantly breaking through our imaginations, and AI systems are fast moving into the real world, where people and machines can get along. ECE Kamar, a senior researcher at Microsoft Research, expressed her views on human-machine complementarity and revealed some common misconceptions about AI. She believes that machines are good at identifying and calculating, that human beings are longer than decisions and creations, and that cooperation between humans and machines will lead to a better future. This paper is a concise version of ECE Kamar's interview. For full content, please click to hear the audio in the text.

Senior researcher of Microsoft Research, ECE Kamar

Interview Audio:

I am passionate about the AI systems that can help people in everyday life, and I am very interested in how machine intelligence and human intelligence should complement each other. Over the years, we've been trying to build systems that can interact with people, work together, and help people.

At present, we are at an important moment in the history of AI, and many important AI systems are entering the real world and starting to collaborate with people. What the AI should do and what role human beings should play in the AI development process will have a big impact on our lives.

correct understanding of artificial intelligence

In the case of artificial intelligence, there are many misunderstandings and blind spots. Most people will think that Ai has appeared in the past 10 years or so. But in fact, Ai has started to develop in the the 1950s. At Dartmouth, the pioneers of AI came together to discuss what artificial intelligence is, how to define it, and study it. They believed that AI could soon be realized-if the machine was capable of dealing with a job, it should be able to quickly grasp the ability to deal with a more difficult job. Unfortunately, our AI today is not so, AI is a vertical work, we can only apply certain specific technologies to specific tasks. we are still very far from the generalized ai, the current AI technology is still very "professional", so in order to apply AI to every field still need a lot of careful analysis and engineering design.

Our AI system relies heavily on data, and the quality of the data collected largely determines the performance of the final AI model. However, the current data collection work still have a lot of problems, when we collect data often with personal subjective ideas and presets, so the data collected does not always represent the world, will lead to the model of knowledge blind area, such as human face recognition system can not identify the people with dark skin. Therefore, incomplete data can cause AI systems to be unable to effectively serve certain minority groups.

the role of the human race

In the age of artificial intelligence, we need to transform the AI development process. In the field of traditional software development, decades of accumulated experience have made it clear to programmers that the results of each bug fix. Now, we should also take this set of standards to the field of AI system development. Over the past few years, we have been trying to integrate human insight into the debugging and bug fixes of AI systems, from a human perspective to understand the problems and blind spots of AI systems and the consequences of problems, and how to try to improve the AI system by incorporating human intelligence.

What I want to emphasize is that AI is developed, driven, and used by people. Ai cannot be born out of thin air. Human is actually a lot of AI system compass, such as computer vision, face recognition and so on. As we design our systems, we also think about how we can incorporate human beings as "compasses" into the training, execution, and troubleshooting of AI systems. Humans are already part of the AI process, and it is the engineers and scientists who work with AI systems that put their insights into the design and architecture of these systems, with the magical AI magic now.

Ai has a lot of data resources from people, the continuous development of AI is inseparable from human intelligence. Crowdsourcing provides a way for AI systems to acquire human intelligence. Through crowdsourcing tasks on the web and social platforms, people produce a large amount of data that AI systems can use to learn. This channel that connects machines and human intelligence is important to us in building AI systems that contain human factors. In addition, in the development of AI also need people's continuous evaluation of AI system, after all, the ultimate AI is for human services, and this is the area where crowdsourcing can play a role.

A brighter future

I am often asked by people outside the AI field, "Can you tell me how many jobs will be lost in the United States in the next 10 and 20 years, and how many new jobs will emerge?" "In fact, we are far away from strong AI , and I don't think AI is going to be smarter enough that all the jobs in the future will disappear." At the same time, we have to think about how to prepare for some of the unemployment problems that may occur.

The key to understanding how people and machines complement each other is that new economic growth will emerge in change. machines are good at automating repetitive tasks, are adept at identifying patterns and handling scale tasks , but machines are not good at many things. humans are good at common-sense matters and can do anti-factual reasoning . We can use small data to learn, and extrapolate, to apply it to things we have never seen before. The most important thing is that we are creative.

In a paper on collaborative research between MIT and Harvard, researchers asked a machine and a radiologist to work together to diagnose breast cancer through images, because machines and humans make different mistakes, and when they work together, the diagnostic error rate is much lower than the error rate of the machine and human diagnosis alone. So, only when we make full use of the complementary relationship between man and machine, can we create a better future of artificial intelligence.

you may also want to See :

Mr. Chang: The most optimistic AI in the financial and medical fields of application

The spring of the creation of artificial intelligence

How to develop AI applications for ordinary programmers.

Thank you for your attention to "Microsoft Research AI Headlines", we look forward to your comments and contributions, build an exchange platform. Please send the manuscript: Msraai@microsoft.com.