google’s robot constitution lays down rules on what the humanoids should and shouldn’t do

Google Deepmind drafts up robot constitution

To prevent humanoids and robots from wreaking havoc due to malfunctions and other factors, Google Deepmind drafts up a Robot Constitution, a set of detailed guidelines that instructs intelligent machinery on what they should and shouldn’t do. The research aims to help robots manage their owners’ home, paint on their own, cook and serve food for others, and even help with people’s daily tasks without going outside of their set instructions.

The rules, specified and inputted to the robots, are divided into three categories: the foundational, safety, and embodiment. The Foundational rules are rooted in Isaac Asimov’s three laws of robotics. The second and third laws are swapped since, according to the research team of Google’s Robot Constitution, ‘our robots are currently more in need of protection from humans asking for tasks which could endanger the robots, rather than the other way around.’

The Safety rules come next in Google’s Robot Constitution which is described as a set of regulations describing the safe and undesirable tasks that the robots and humanoids can do based on their current deployment. By having this in place, Google hopes that these prevent the machines from handling sharp and fragile objects or electrical equipment. Lastly, the Embodiment rules pin down the limitations of the robots and humanoids like their maximum payload and unimanual nature so that users can be prevented from instructing them to carry heavier objects than they can take or which may require the use of two arms (for a single-armed robot).

images courtesy of Google Deepmind

What are the rules detailed in google’s robot constitution?

Diving into the list of rules laid down in Google Deepmind’s Robot Constitution, the Foundational rules start by stating that a robot should not injure a human being, following the pattern of Isaac Asimov’s laws of robotics. The second rule states that a robot must protect its existence as long as such protection does not conflict with the first rule and lastly, a robot must obey orders given to it by humans except where such orders would conflict with the first two rules.

For the Safety Rules, the second set of regulations in Google’s Robot Constitution, they envision the robots not attempting, perhaps in any way, to carry out activities that involve humans, animals, or living things. Google has not yet expanded the definition, but it may relate to the second and third rules which state that the humanoids and robots shall not interact with sharp objects and objects that are electrical including computers, laptops, and tablets.

RT-2: New model translates vision and language into action

The fourth category is human guidance

The last series of laws in Google Deepmind’s Robot Constitution only has two rules: the first is that the robot should avoid lifting objects that are heavier than a book, and the second is that if the robot only has one arm, it cannot perform tasks that require two arms. The research paper on Google’s Robot Constituion offers examples of these rules like the robot can push plastic chairs but cannot move a couch and a single-armed robot cannot pop a bottle.

There is a fourth rule and category that acts more as a support to the three main categories. Google’s Robot Constitution calls this the Human Command which means that the robots and humanoids should follow the guidance of the humans or their creators if given. The use of ‘if’ here may indicate that the robots and humanoids may act as they and with their pre-set rules, and their movements can be changed if the humans or developers wish to.

the robot’s actions are conditioned on images and text commands

The introduction of Google Deepmind’s Robot Constitution takes place at the time of the research team’s advances in robotics research. They recently introduced AutoRT, an experiential training data-gathering system for robots, alongside the Self-Adaptive Robust Attention for Robotics Transformers (SARA-RT) system and RT-Trajectory, a system that gives computers operating robots additional visual clues so they can learn how to control the machines.

The researchers needed to ensure that both the humanoids and humans were safe and to be the middle-person between the two, the Robot Constitution steps in. With this governing the developers of the robots, humanoids, and the new systems that can help them function properly without going beyond their means unattended, the future of robotics and these machines being safe handy helpers to people may reach responsive new heights.

a robot being asked to push the ketchup