When artificial intelligence was just emerging, most people were still familiar with this vocabulary, but as AlphaGo continued to defeat the top human players in the world of chess, artificial intelligence became famous and finally became the future development strategy of various countries.
In fact, compared with humans, artificial intelligence has outstanding advantages in many aspects, such as computing power, "memory" ability, learning ability, predictive ability and so on. Specifically, artificial intelligence can store or collect large amounts of information data, based on which “thinking” or “learning” and drawing conclusions. Under this advantage, many artificial bits of intelligence has played the role of "encyclopedia" and become a flexible database. Behind this, relying on the support of many key technologies such as voice interaction, natural language understanding, and deep learning.
For the development of artificial intelligence, the most threatening argument is that, in the future of automation, human job opportunities will be highly likely to be replaced by machines. Not only doctors, teachers, customer service, etc., but even the host has begun to be "snacked". For example, just before a conference called “One-Time talks to see the future conference”, the artificial intelligence “Wang Zi” developed by Sogou was a guest host.
We all know that a professionally-sponsored host needs to have not only sufficient professional knowledge and eloquence but also a strong ability to respond on the spot. But at this meeting, the artificial intelligence "Wang Tsai" gave everyone a wonderful performance and won the audience's favorite. This is enough to let people know that artificial intelligence may become a veritable "workplace killer" in the near future.
However, from the current situation, artificial intelligence does not have the ability to completely replace human decision-making. Now human-machine integration may be a good choice. In this regard, David Mindell, author of the Massachusetts Institute of Technology and author of Our Robots, Ourselves, said that this hybrid decision should produce better results than letting artificial intelligence work alone. There is only one problem: when humans work with semi-intelligent systems, the results are not always good.
For example, in March of this year, an automatic driving death occurred in Arizona. Normally, there will be a security officer on the self-driving vehicle that is being tested. When the software fails, a human driver will intervene. Unfortunately, in that accident, the software went wrong, but the human driver might be watching TV programs with distraction. So it caused this disaster.
After the incident, some critics believe that this failure shows that the predicament of artificial intelligence far exceeds that of self-driving cars. If not carefully designed, the entry of smart systems into the world may trigger a strong resistance of humans to technology. At the same time, we have to admit that there are still many flaws in today's machine learning systems. In response, AI expert Roger Schank warned that the future would be a new "Artificial Intelligence Winter" - a reference to a period in the late 1980s when disappointment with technological progress led to the withdrawal of the battlefield.
Some people say that we can properly adjust the expectations of the autonomous system and use more elaborate design to ensure that they are more integrated with the human world, but in fact, technology itself is a serious obstacle. Allah Nourbakhsh, a professor of robotics at Carnegie Mellon University, said: "The way artificial intelligence works and the way it fails is unfamiliar to us."
Take machine learning as an example. We all know that machine learning is a popular branch of artificial intelligence, and now it has proven that machines have capabilities superior to humans, such as recognizing images or speech. However, when it needs to rely on the specific data of the training to make a decision, the performance is not very good. But in the real world, people often make decisions about situations that have not been encountered before. Therefore, man-machine cooperation is one of the best choices for solving this problem. So, what are the ways in which people and robots coexist?
First, humans act as a backup for robots, taking over when the robot reaches its limits. Many workflows are being redesigned in this way, just like automated call centers, where the language understanding system attempts to process caller queries and only confirms to the operator when the technology is confusing.
Some people may say that Uber's fatal car accident happened under such circumstances, but this accident is an extreme example of a possible mistake. A study by Stanford University shows that it takes at least 6 seconds for a human driver to regain consciousness and regain control. But even if there is enough time to regain attention, someone entering a situation may see something different from the machine, so that there is no seamless switching between the two.
We need to work together between the software system and people, but this is a very difficult thing, at least the language is difficult to use. For humans, we can use a few words to convey meaning - the common understanding of the context between the speaker and the listener, making these words more meaningful, but the machine does when understanding the human language. Less than that, at the same time, computer scientists have not yet studied how to build consensus in machines.
Second, in the human-computer cooperation to ensure that sensitive tasks are always dominated by humans. Even when the automation system has completed all the preparations and is able to fully complete the task itself, sensitive tasks involving military and other fields are best handled by humans. An example is a military drone, a human "pilot", usually located thousands of miles away, and being asked to make a shooting target. Another example is the facial recognition system used to help immigration officials identify suspicious travelers, but both show how artificial intelligence can make humans more effective without depriving control.
However, there is another saying for semi-automatic weapons such as drones, that is, there is no technical obstacle to turning them into fully autonomous systems, and current procedures and security measures can be quickly changed.
According to Stuart Russell, a professor of artificial intelligence at the University of California at Berkeley, removing human drone operators from the workflow in a national emergency is a simple and easy step, but it will accelerate the advent of the robotic era. And this so-called robotic weapon can also decide when to start killing. So, "You can't say that the technology itself can only be used in a defensive manner under artificial control. This is not the case," Russell said.
Third, it involves the use of artificial intelligence in the "human circulation" system. This means that robots can't handle tasks completely independently, but can play a supporting role in human decision making, such as gradually infiltrating algorithms into everyday life. However, the algorithms are only as good as the data they train, and they are not good at handling new situations. This requires people who trust these systems to also believe in these systems.
Mr. Schank pointed out the role of the algorithm in baseball: analyzing the strengths and weaknesses of each batter, providing a new approach to the field that the team's traditionalists tend to. He said that the results of these computer-aided decisions may end up worse than decisions based on pure human analysis.
Once the Uber driver in San Francisco had an error in the application they used, they were sent to the airport cargo terminal instead of the passenger terminal. “Sometimes people blindly follow the machine, and sometimes people say, 'Keep it up, it doesn’t look right.' It's like many other technologies, people will adapt,” says technology author Tim O’Reilly.
Nevertheless, we can still be optimistic about the coexistence of humans and machines, that is, the result of human-machine integration will be far greater than the result of any task performed by anyone or any machine alone. Only before this beautiful future comes, we still face many serious challenges to solve.