Google, who is currently working on creating a fully functioning robot formulated 5 future robot problems that could interfere with the daily lives of the humans and their household intelligent machines. Judging by the mundane character of the problems, it seems that Google officials are less scared of a possible Skynet power overthrow than other scientists.
Elon Musk and Stephen Hawking are two of the most appreciated scientists of the present moment. Apart from their high scientific value, the two have something else in common; both believe that artificial intelligence will soon take over the planet.
Both Musk and Hawking have spoken up about the danger of the continuous advances in robotics. They argued that AIs will never be able to feel empathy, or see the gray shades between the black and the white, the zeros and the ones.
However, it seems that the scientists are approaching the matter in a philosophical manner. They are talking ethics when it comes to future robot problems instead of focusing their attention on the mundane.
Google approached the situation from a more pragmatic position and formulated a short list of five future robot problems that can and will occur once the machines are globally available.
- Negative Side Effects
Amongst the future robot problems that Google scientists formulated is the apparition of negative side effects. These occurrences refer to frequent instances like throwing away a glass instead of washing it because it’s faster and easier.
Imagine a robot built with the sole purpose of cleaning a house. If its programming leaves room for error, it could believe that knocking down the vase on the counter while dusting is the most efficient solution to the task. That is a negative side effect that scientists will have to counteract.
- Reward Hacking
Robots will accomplish their task in accordance with their programming. While it is not possible (except if you code it) for a robot to cheat, it can hack the rewarding system. In simpler terms, the aforementioned cleaning robot will cover a dirty area with a piece of cloth and believe that its task is complete because the dirty area is not visible anymore.
- Scalable Oversight
This future robot problem refers to the costs of maintenance of such a machine. If a human evaluation is costly, then the AI will have to learn to adapt so that it will use a periodic evaluation as a guideline for future endeavors.
- Safe Exploration
A robot will be encouraged to experiment different approaches for its tasks. However, the designers must find failsafe features that will guarantee users that their cleaning robot will not throw the family cat in the dishwasher.
- Distributional Shift Robustness
Robots will be properly programmed and trained before being sold. However, the environment in which they will get their training will most likely be different from the one in which they will operate. This rises the distributional shift robustness problem that can only be solved by anticipating various environments and educating the robots for a wider array of tasks.
An example would be that a robot that learns how to mop floors in a special facility, like a factory, will not behave well in a nursery or an office.
The 5 future robot problems that were formulated by Google deal with mundane things, real malfunctions that can and will occur if the programming is sloppy.
The great minds of the century are focusing on robotic ethics, fearing an uprising against humankind when the real problem might be the fact that the cleaning robot you just bought throws your plants over the balcony because it makes its task of cleaning the windows easier.
What is your opinion on the 5 future robot problems that the researchers at Google formulated? Do you think that their pragmatic approach to robotics is better than the philosophical one, or do they both matter? We value your opinion so don’t be shy and let us know in the comment section below.
Image source: YouTube