I have programmed industrial robots and the closest those robots came to "insight" was knowing that they had to slow-down in order to accurately make a turn. This presents a quandry of sorts when the robot is doing something like putting a sealant along a seal surface, where a larger radius rounded corner is not what is needed. The solution was to bring the robot to a point, then a separate move from that point to the change in direction point, and then start in the new direction. A simple work-around. But if the robot had been able to tell that it needed to do something in order to be able to change direction it may have been easier to figure out. Instead, it was nessesary to read the 4000 page instruction manual.
The problems that will come with attempting to give robots insight is that it may easily lead to giving the robots self-awareness, which would probably lead to robots having emotions, and that could be VERY BAD. That is because robot source code is written by programmers, and programmers are not normal people. We need to always remember that, and beware.
Absolutely, William K. That is a very good description of the issue. Even routine functions can quickly change to ones that require past experience. That is why a lot of experts can operate on "gut feel". They can't explain their correct actions because it is based off of experience of similar occurences. This simply cannot be captured in a program.
ttemple, everything you said is correct about non-autonomous robots. This research team, like several others, is developing intelligent, autonomous robots, something very different. William's comment below, "Human-Robot communications", captures this difference.
The robot is following some program that some human loaded into it. The robot can only do what the programmer told it to do.
So, the human tells the robot what to do (via the program), and then the human says "why are you doing that?". The answer is always the same - "because you told me to".
I would think the obvious solution to this supposed problem is to send all sensor data to a computer that is running the same decision making software as the robot, and watch what the program is doing. (It will be doing what you told it to do - which may or may not be what you thought you told it to do.)
This article somehow makes it sound like the robot has a mind of its own. It doesn't. It can only do what some human told it to do, so why ask it why? The answer is, I'm doing what you told me to do given these sensor values.
William, thanks for clarifying. I agree, when I read the initial report, I thought why the heck hadn't somebody already figured this out and implemented it ages ago? OTOH, I don't think the state of hardware--sensors and processors--and comms tech were available for robots that could take advantage of this "translation" program.
Yes, Ann, the robots have only their sensor information to base decisions on, and that is often not enough to make the very best choice. That was part of the basis for my comments about the value of experienced humans in the loop. Robots lack insight and understanding, they can only make the decisions that they are programmed to make, which may well be safe, but probably not optimum.
Giving the robots more data by allowing accurate communication will certainly offer the potential for better choices, and the concept of communicating that basis for the choices to a human is a good idea that should have been put into practice about 25 years ago.
Mydesign, wireless communication with remote-controlled robots is already used in military, nautical and rescue robots, among many other types, as we've mentioned before: http://www.designnews.com/author.asp?section_id=1386&doc_id=247687 http://www.designnews.com/author.asp?section_id=1386&doc_id=242527 http://www.designnews.com/author.asp?section_id=1386&doc_id=246206 But that does not solve the communication problem. Most robots can only report back very limited types of data. And communication is one way in one direction and then one way in the other direction. It does not allow for full-duplex two-way conversations. Plus, the robots are not intelligent enough, or autonomous enough, to perform the delicate operations of decommissioning a nuclear power plant.
Thanks, William, I think you captured the point of this research in your comments about autonomy. It is aimed at more autonomy in robots, which is why communication has to be much more detailed, and accurate, than it has to date. But inadequacy of the human operator is not the issue: inadequacy and incompleteness of information about why the robot makes the decisions it makes was one of the main spurs to this research. The two-way logic-to-text and text-to-logic communication will also let the human make informed suggestions and provide more data once it understands the situation as reported by the robot.
Ann, I have some other idea for disaster management, where humans can interact with robots via wifi or any other communication channel. This will help for a remote control operation from a master facility to control each and every wing of the nuclear station and safe shut down, in case of disaster.
Focus on Fundamentals consists of 45-minute on-line classes that cover a host of technologies. You learn without leaving the comfort of your desk. All classes are taught by subject-matter experts and all are archived. So if you can't attend live, attend at your convenience.