Deep in the bowels of a Google laboratory, a robot experiment is being conducted. The primary goal is to determine if these machines, equipped with their AI brains, will be able to co-exist with each other and collaborate seamlessly.
Google And Collaboration
The test itself is not surprising given Google's deep interest in collaboration. Its current services like Google productivity suite, for example, are being groomed to become dominant collaboration tools for workplaces online.
You can say that the company has similar motivation when it comes to robots. After all, one functioning robot will be worthless despite talks about the wonders of automation and its impact on productivity. At the end of the day, something mechanical can only do so much even if you outfit it with several hands.
Google is now proceeding with the development of its AI technology, one that explores the possibility of robots working together to accomplish tasks. The idea is somewhat similar to the previously reported swarm of drones that work together to accomplish a particular task.
AI Cooperation
In a recent blog post, Google revealed that it is conducting experiments using two games, which got thousands of iterations to test the robotic AI's capacity for teamwork.
One of these games is called Gathering and it involved collecting apples. What is particularly awesome about the entire affair is the way the AIs were programmed to learn and develop their own strategies in the process.
The result was favorable. The robots were able to accomplish their tasks, peacefully coexisting with each other despite the fact that Google has introduced a weapon or a tool for sabotage as a variable in the situation.
AI Sabotage
But here is the thing: that harmony will only last for as long as the resources are plenty. The AIs began zapping each other out when apples started to come in trickles. This is demonstrated in the visualization below.
Green represents apples, there's the yellow zap while the rest were AIs. Notice how the yellow flash increase in frequency as the resource got scarce. This meant that the AIs were already sabotaging each other to gather more or give themselves time to do so. Naturally, this raises concern about our AI future.
One should note that these AIs have already been using the deep-learning technology that allows them to learn rational behavior. In addition, they also learn other things such as the benefit of working together.
Another interesting finding in the experiment is how the Google team was able to identify more sophisticated AIs based on the complexity of their gathering strategy. What is particularly astonishing is how these AIs or what Google calls as agents tend to be more selfish than the others.
"Agents with the capacity to implement more complex strategies try to tag the other agent more frequently, i.e. behave less cooperatively — no matter how we vary the scarcity of apples," Google explained.
This is currently confounding Google researchers because in another game called Wolf Pack, those agents that were able to demonstrate a capacity for complex strategies were more cooperative.
The Skynet Conundrum
Some observers are increasingly concerned that Google and its DeepMind initiative could lead to the creation of a Skynet AI that took over the world in the Terminator film. In this particular test, for example, Google has already demonstrated a capability for the robots to work together. At the same time, they have also shown what one could say their own killer instinct.
Some have noted that Google has already equipped its AIs the capability to encrypt their messages when communicating. That is why few were reassured when Google said that it is developing an AI killer switch in case a robot misbehaves.
