By 2021, almost all everyday software will have intelligence and more powerful functions, which can replace human beings to do more and more work. How will this affect us? Some people predict that it will lead to massive unemployment or a full-scale war between humans and artificial intelligence (AI), and others predict the future is very bleak. However, Manuela Veloso, dean of the School of Machine Learning at Carnegie Mellon University, believes that in the future, humans and intelligent systems will be inseparable. They need to be closely integrated and exchange information. She calls this The relationship is "symbiotic autonomy." Veloso predicts that in the future, it will be difficult to distinguish between human agents and automation assistants, but neither humans nor software will be able to play without any party.
Veloso has begun testing his theory at Carnegie Mellon University and has created a robotic robot similar to the Segway shape. This robot can automatically escort guests to different buildings, and when they are unable to meet the needs of the guests, they will ask for help from others. This is a new approach to AI research that will have far-reaching implications over the next five years. Veloso recently interviewed Pittsburgh about the progress of robotics, programming spontaneous technology, and the challenge of AI threatening humans. The following is an interview summary:
Q: One of the big trends we have seen in the past five years has been automation. At the same time, we have seen more intelligent systems being embedded in existing devices, such as smartphones and computers. In the next 5 years, what areas do you think this trend will spread to?
Veloso: In the future, I believe that human beings and AI systems that are expected to serve humanity will form a coexistence relationship. These AI systems will be embedded in software systems that process the digital world, while being embedded in systems that move around physical space, such as drones, robots, and automated cars. In addition, AI can also be used in systems that handle physical space, such as the Internet of Things.
In the physical world, you will see more and more intelligent systems, not just on smartphones or computers, but on the various devices around us, used to process and sense information about the material world, to help Learn more about the physical world so that we can make a decision. Over time, we will also see these AI systems having an impact on a broader range of issues in society: managing transportation in large cities, making complex predictions about climate, and supporting humans making major decisions.
Q: Now, some AI systems are not performing as well. When algorithms or robots make decisions, we don't always understand why they make such decisions, and which decisions are more trustworthy. Can technical experts deal with such problems?
Veloso: One of my jobs is to allow these machines to explain their behavior, they need to be responsible for their decisions, and the decision process needs to be transparent. Many of the research we did was to let humans or users query the system, and when my robot Cobot arrived at my office later, I would ask "Why are you late?", "Which route did you take?"
We are working to give these AI systems new capabilities so they can explain their behavior. They will also be able to provide explanations of different levels of detail during the learning and improvement process. We hope to have deeper interaction with robots and ultimately enable humans to trust the AI ​​system more. You might ask them: "Why do you (AI system) say that?" or "Why do you offer such a suggestion?"
Providing these explanations is exactly the subject I am studying. I believe that robots will be able to do this, which will lead people to better understand and trust these robots. Ultimately, through these interactions, humans can also correct the AI ​​system. To this end, we are studying attempts to correct the AI ​​system and help them learn the instructions. I think this is very important for the coexistence of humans and AI systems.
Q: Why do you think these AI systems are evolving rapidly? What progress has been made in AI research over the past 50 years?
Veloso: You must know that for the AI ​​system, you need to learn a lot of knowledge to understand what a mobile phone, a teacup or someone is healthy. Many of the early AI studies actually learned how to acquire this knowledge. We may have to ask humans, or look for information in the book and manually enter information into the computer.
What is amazing is that in the past few years, more and more information has been digitized, and it seems that the whole world has been moved to the Internet. To this end, AI systems now have a large amount of data available, and they have the ability to process this data and make it more meaningful. And we have to find the best way to help the AI ​​system. On the other hand, we are very optimistic about this because we know that data is already in existence.
The question now is how do we get insights from the data? How do you use it? How do you study distribution, that is, statistics? How do you put these pieces together? This is what you need to learn in depth and deepen your learning. The system can be automatically translated and the robot can play football. All of these things are possible because we can process this data very efficiently, and we don't have to spend more time getting the knowledge.
Q: In the past five years, personal assistants like Siri and Alexa have made great strides, and they all have machine learning support. I am very curious, how do you think these systems will change in the next 5 years?
Veloso: You know, I am a loyal fan of Alexa. I have Alexa at home, and the topics I can talk to are more and more extensive. Initially, I could only ask the simple question "How is the weather today?" Now, I can ask it "What arrangements do I have today?" Alexa is learning, and I am also learning what Alexa can do. I am also eager to see what changes will happen over time.
I can tell you something interesting: when I leave home, I will tell Alexa: Alexa, please stop! I want it to stop playing the music that is playing because I am about to leave. But if I tell Alexa: Alexa, I am leaving! It also doesn't quite understand the meaning of this sentence, so it won't stop playing music. I must then add specific instructions to stop the music. I think that the personal assistant will become more and more "instructive". When I say "I am leaving," it should be able to understand the meaning of the words, that is, stop playing music. Now, this "instruction awareness" is also included in the research agenda.
Q: Do you think that personal assistants can eventually evolve to the following degree? For example, we can ask it: Check the car engine, can I ride safely? Or I just received a job offer, should I accept it?
Veloso: I think this is entirely possible. These issues are all part of the decision-making problem. If you have to choose between which health insurance plans and you are confused about all the options, you can tell Alexa to go to bed and ask Alexa to help you read all the insurance plans, or all the models you can buy. All the schools your child can go to, etc. Alexa can use the evening time to sort out the complete report for you.
Now, there is a lot of relevant information on the Internet. You can find out all the school's features and see what others have to say about the school. The AI ​​system can collect this information, such as how far they are from your home, how other people's comments are waiting. They can understand the functions of these schools, understand their past history, and process all available information to provide you with answers and guidance that will make it easier for you to digest. Because the amount of information on the Internet is too large, sometimes you can't see it.
In the end, you may want your personal assistant to tell you why they will give you current advice. You might ask: "Why do you think I should buy that car? I really don't like the brand." I think this is a very important step, with the support of the AI ​​system for human decision-making, combined with what you know. With the information and the company feedback you get, you will be able to make more informed decisions.
Q: In addition to helping people make decisions, what else can these AI assistants do?
Veloso: As you can imagine, a similar system can be used to help you deal with scientific papers. Numerous scientific papers are published every year, and now they have been published online. You can imagine the AI ​​system helping researchers digest all the information and find things that are relevant to their interests.
The AI ​​system will remain a network information product. Many people are researching this information, such as text information, image information, flowcharts, icons, etc., and try to understand the information, and finally filter out the information you may need. For example, machine learning has a branching field called "active learning." If you don't have enough images in a process, you might want to add more images.
I think the AI ​​system will be able to identify missing images and connect all the information on the web and request more data if necessary. As you can imagine, it asks the researchers: "If you tell me more about how these cells interact with this chemical, I can get a better correlation model."
Of course, these automatic robots also have limitations, they can not open the door to all the world, because they can not understand all the spoken language. But I think that human beings also have limitations, such as speech and accent. Obviously, for us, one of the main functions of these robots and AI systems will be to identify things they don't know, things they can't do, information they don't understand, and turn to humans for help. Can you press the elevator button? Can you open the door? Can you put things in my basket? This is what I call "symbiotic autonomy." When there are things that robots can't do, don't know, or don't understand, they can turn to humans for help. This is a new way of thinking. The AI ​​systems around us will seek our help and we need to share some of their tasks.
As these AI systems increase, this situation becomes more complicated. The system is already wirelessly enabled, can get cloud data, or seek help from a remote team. As you can imagine, the AI ​​system is constantly symbiotic with anything, such as information on the Internet, other AI systems, humans next to it, and remote human teams. This is no longer a problem of developing self-sufficient AI systems, but an AI system that can identify things that you don't know, need to know more, and identify things that you are not sure about. This is not to say that the AI ​​system no longer focuses on solving all problems, but can rely on surrounding resources to solve problems.
Q: How do you see the change in the symbiotic relationship between the AI ​​system we are using and humans?
Veloso: Let's go back to the above assumption that the AI ​​is going to help decide which school the child is on or which health insurance we buy. I think that under certain circumstances, these AI systems may require information that humans have not provided. The AI ​​system may be aware that if it can know this additional information, it can help your decision better.
The most interesting thing is that when the AI ​​system is able to identify the information that is missing by itself. They have actually realized that they need more information to take concrete action. For example, if they can book a hotel room that cannot be booked online, they can book a hotel not far from your venue. I really think this ability is very important, because even I don't know what information the system needs to know when making this decision.
Now we can enter the destination address into Uber, Google Maps or Waze, which is enough to help us plan the route. However, Waze can't communicate with you, such as asking you: "Are you in a hurry? Do I need to plan the shortest path for you? Or do you just want to distract yourself and browse the beauty there?" If the personal assistant knows I like it very much Orchid, or I love some kind of art, what will happen? If I deviate slightly from the route, I may visit the museum by the way. When planning a route, if the personal assistant knows this, it might plan for me to route through the museum.
The challenge for General AI is daunting. We now have deep learning, deep-enhanced learning and other technologies that already give them a sense of general intelligence. We are still doing a lot of research and trying to understand the concept of migration learning. How do we develop an algorithm? Because they can solve specific problems, can they learn to do other things as well? We don't fully understand AI, we don't know many things about them. From the perspectives of algorithms, techniques, general methods, and explanations, we are still in the early stages of AI, and we still need to continue our efforts.
But I believe that GM will integrate many specialized AI systems in the future and integrate them into the "mind society" described by Minsky. At the beginning of AI research, artificial intelligence pioneers Allen Newell and Herb Simon predicted that there might be special-purpose algorithms that could solve very complex problems.
For this reason, even if the study of the general AI may face many challenges, it is also very exciting, because now has such a large amount of data. There are so many people using digital devices and constantly generating data. Through the increasing use of computers, mobile phones, Alexa, Uber, etc., all of this helps us to study the general AI problem. We still have a lot of research to do, and we still don't know what general AI is, but we are moving in that direction.
Q: Does this uncertainty make you worried? Some people worry that AI will transcend human intelligence and humanity will be destroyed.
Veloso: I am an optimist. I think that our ongoing automation system research (including automatic cars and automatic robots) requires human responsibility. In a sense, this has nothing to do with technology. Technology will continue to make progress, it was invented by humans, not by aliens thrown from the sky. These are our own discoveries, and the human brain has envisioned technology, and it also depends on how humans use it.
I am optimistic about this because humans have realized that we need to be cautious about this technology. But the best way is to invest in education. Robots may get better and better, but when they invest in education, people can understand each other and take care of each other. Caring for social progress, caring for the earth and nature, and improving science. Address all of these issues, including curing cancer and eradicating poverty. With this technology developed by humans, we can do a lot of things.
In a sense, it is the human nature of AI that brings us together and is hard to leave. To this end, I am optimistic about the future.
LED Module Indoor Full Color P3 Module Indoor LED Display Module using the latest generation technology, 111111dots per square meter. P3 Module Indoor Full Color with high refresh rate, anti corrosion, anti mildew, anti shock, anti electromagnetic. P3 LED Display Module is one good choice when compare Indoor LED Display Module. We sincerely invite customers all over the world visit us for cooperation.
P3 Led Module,P3 Module Indoor,P3 Indoor Full Color Led Module,P3 Indoor Smd Led Module
Shenzhen Jongsun Electronic Technology Co., Ltd. , https://www.jongsunled.com