There is the question of liability. If an autonomous robot or a self-propelled vehicle injures someone, who takes over for the damage? The owner of the robot, the manufacturer, the owner or the person who has learned how to behave robot, either through machine learning? As robots take over the work of people and add value to the business, what are the tax consequences and the consequences for social security? So come all kinds of questions when you read the report , but which are rarely answered.
The report was prepared to launch a thinking process that should eventually lead to legislation. I find it a very interesting topic, and in this blog I like to go into some of the issues raised in the authors of the report. Some of them have to do with natural language, which creates problems but surprisingly also be part of the solution.
A round of science fiction
You can dismiss the report as a hype; politicians without knowledge of things that do a round of science fiction. Of course, artificial intelligence (AI) is currently a popular topic, but that is mainly because of developments as soon go on the themes 'deep learning' can make the difference: finding and interpreting large amounts of data, image and voice recognition, klantenservicebots and gaming.
We can in any case agree that progress is so insanely fast that even for experts, let alone the public, it is impossible to predict where artificial intelligence in five, is ten or fifteen years and what impact has it. But we can not simply ignore, hype or no hype, and that also applies to politicians.
The report, as you might expect, refers to the laws of robotics by Isaac Asimov and the committee proposes that no one knows when robots and AI applications are self-conscious. Until that time, says the committee, are especially designers, manufacturers and operators of the robots that have to comply with those laws.
Let robots themselves tell what they know
The issues at the macro level - what can artificial intelligence technology is and is not - also apply to the micro-level: how do you determine the limits of what a robot can and should? The report says that users should assume that a robot limits the degree of perception, cognition and freedom. Fine remark, but what can you do that? What are the limitations of the system that I'm talking?
Before there were talking machines, it was clear: animals and things can not talk, so all it talks must be a man. And we actually still thinking. That's why we are still struggling to even scientifically determine how intelligent one (speaking) system is genuine.
One of the recommendations of the report is that robots should always be recognizable as a robot when there is interaction with people. That seems to be a minimum requirement. Yet already shown that not everyone complies with this requirement. In American elections massively bots were deployed without that were recognizable as such. These are mainly ethical issues, but also in a practical sense, for example, in mainstream bone support, such as Cortana and Siri, difficult to determine what the possibilities and limits.
"Contextual Reasoning Framework
Another related opinion, the report is sent to the designers. The committee recommends that to ensure that all decision steps of the robot are traceable and can be reconstructed. Not only retroactively to incidents, but continuously. We as a people should always know why an autonomous system where we took certain decisions with confidence. That seems to be becoming more complex at first glance, in this time of deep learning. It is not easy to understand why a so-called Neural Network does what it does, there are more people against it .
One possible solution is to build rules into the system so you can always reduce the 'reasoning'. As a supplement to natural language is a good way to let robots and AI applications to tell people how they think. It would be nice if you are in a self-driving car and the car says, "I'll get here as the exit to the A27 because there is the A1 five-kilometer traffic jam."
You know that joke of the two domestic robots?
The report also discusses the fact that autonomous robots that are connected through the network must be interoperable. There, too, natural language can be very useful. Increasingly bots, robots and computer assistants populate our homes and workplace, and you do not have crystal ball to predict that eventually will often talk to each other than with our people.
A computer assistant talking on our behalf with a virtual customer service representative and you know that joke of the two domestic robots that run into each other? At which technological system they are built, they will all manage natural language to a certain level. And it has the great advantage that if for example two house robots talk to each other, we can listen to people when they talk among themselves and can intervene if necessary.
There are still many more aspects discussed in the report of the committee. And there will still answer many questions about the rules for robots and for the people who develop these robots. It is a particularly interesting and important debate. So Parliament, thanks to further boosting the discussion!
No comments:
Post a Comment