Most of us would just as soon not be afraid, but our natural fear response can have some unexpected benefits.
This point is made salient by recent scientific efforts to program fear responses into artificial intelligence. While this might seem like an odd thing to do, programmers saw this as a helpful component to teaching computers how to effectively drive cars.
Computers can accomplish a lot of incredible tasks, but they tend to learn through trial and error. This makes sense if computers are given a game. They can calculate likely outcomes of given moves, essentially trying out various simulations in order to determine which will be the best approach.
Often this is used in machine learning contexts, where the computer simply tries various tasks and, based on the results, eventually learns how best to handle unexpected situations. While this works great in low-risk situations, it might be more challenging when faced with high-risk tasks, such as driving a vehicle.
Driving is an exceptionally dangerous thing to do, and as humans, we recognize that. This results in a certain level of fear and anxiety for human drivers, compelling us to be more cautious.
But if a computer seeks to learn driving through trial and error, they will inevitably cause many accidents before learning how best to drive. While these trials could be simulated, incorporating human emotions might streamline the process.
That was the hypothesis taken by a team of artificial intelligence researchers with Microsoft. In order to simulate fear, the researchers asked a group of participants to engage in a driving simulation. During this simulation, they recorded pulse rate for each participant, looking at what situations led to increased pulse rates. This served as a proxy for the driver’s fear/anxiety response.
These were then coded into the artificial intelligence in order to simulate similar fear responses when the computer faced scary situations. While the process still had to be learned through trial and error, the researchers estimate a 25 percent reduction in crashes over the computers who had not been taught to be afraid.
This was able to teach the computers not to turn rapidly, for instance, which scientists had been struggling to do with traditional means.
Fear is certainly an unpleasant emotion, and it is easy to forget that it serves a purpose. As we have noted in previous blogs, emotions such as fear are rooted in evolutionary drives and make up a great deal of how we perceive the world.
Such research raises questions about how other emotions can be incorporated into artificial intelligence to create richer and more life-like computers, working to bridge the gap between their capabilities and ours.
Glad someone already worked on it. I was thinking about it since weeks now and i thought is should Google it to know if someone is already working on it.
In last two months, i had encounters with a frog and a spider, when I tried to go behind them they would try to go to the corner of the enclosed space they were in making me realise that every living being has fear as a commonality. Now this fear works mysteriously in the process of evolution, like human beings did not have hair to withstand extreme cold and heat(or intruders like other creatures), and so they came up with the concept of comfortable housing.
I am really happy to see someone thinking on the same lines.