Scientists are developing robots that can “breed”, learn and evolve on their own

Man comes two hundred years later: Scientists are developing robots that are able to reproduce, learn and evolve, meaning by adapting to selection mechanisms typical of the natural world. The fascinating (but also disturbing) perspective was announced by a research team from York University (UK)

Sounds like science fiction but isn’t: a research group ofYork University (UK) is already working on a project called Autonomous robot evolution (self-development of robotics), with the aim of developing robots capable of being autonomous and evolving by adapting to the environment.

A few steps back in the history of artificial intelligence

It is clear that the science of artificial intelligence is not new and has seen amazing leaps in recent times until recently. Purely self-learning algorithms are not new but increasingly complex, and rely on neural networks.

The idea of ​​using evolutionary principles to design things can be traced back to early sixties And in The Origins of Evolutionary Computing, when a group of German engineering students invented the first “evolutionary strategy”—Emma Hart, who collaborated on the research, explained to NewScientist—their new algorithm created a series of projects and then chose a group, geared toward high-performance elements, that could Build on it in subsequent iterations

Yes, because like many technologies, this one also aims to imitate nature, the best intelligence imaginable.

Read also: Beach Bot, the smart robot that collects cigarette butts and other small waste from beaches

In 2017, news of an experiment conducted by Facebook spread around the world, in which two bots began to speak to each other in an unknown language, an episode that left the scientists themselves baffled, at least at first, who decided to put everything on hold.

The scientist was afraid that the independent thing had “appeared” and was no longer under the direct control of man. But in reality it wasn’t (yet) and one of the group’s scientists clarified on his Facebook profile that what happened was in fact a “well-established sub-field of AI”, and in practice something that was already known could happen.

So it shouldn’t be surprising (up to a certain point) that in 2022 something is really experienced that in some sense can “go on on its own”, no matter how annoying it may be.

Do robots that “reproduce” really arrive?

Artificial evolution has been around, and that too, for quite some time. Already in 2006, NASA sent a satellite into space with a communication antenna created using this technology, which is based on the design of artificial DNA capable of recombination and thus “reproduction”.

Advances in evolutionary calculus have led to better ways of representing and processing the information from which designs are built, and their hypothetical DNA, when ‘offspring’ is generated to reverse the processes found in nature – continues Hart – and these include DNA mutation and recombination, which creates genetic diversity by breaking down stretches of DNA and then recombining them in new ways. Examples of evolutionary design in practice now range from tables to novel molecules with desired functionality

Read also: Time created inside Google’s quantum computer could change physics (and computing) forever

So the basics are there, but no, we’re not there yet.

Robot design brings a new stimulating dimension to the field: In addition to bodies, brains are required to interpret information from their environments and translate it into desired behavior – which defines the world – […] In the twenty-first century, there has been a change To develop both the body and the brain of the robot simultaneously. While this complicates the evolutionary process, there’s a big plus: Devoting some intelligence to the body can reduce the need for complexity in the brain.

Project ARE – Evolution of the Autonomous Robot

It was born out of all this, which provides a completely autonomous system in which robots equipped with sensors can be manufactured, adapted and developed in the real world. Launched in 2018 and funded byEngineering and Physical Sciences Research Council In the UK, use the . file Synthetic genetic code to define robot body and brain. Evolution occurs in a structure called ifosphere, which puts each robot through a three-step cycle: manufacturing, learning and testing. What the authors call the “triangle of life.”

In the new advanced designs they are built independently: a 3D printer first creates a plastic skeleton, then a robotic assembly arm selects and connects sensors and means of movement defined by a series of off-the-shelf components, and eventually a computer is added Raspberry pie Which acts as a brain connected to sensors and actuators, and the software that represents the developing brain is downloaded.

Instead, the learning stage A Algorithm to master the brain In a small number of tests in a simplified environment. This process is similar to what happens in a child learning new skills in nursery school. Only those bots that are considered valid move on to the third stage: testing.

Currently, we are using an indoor nuclear reactor model for tests in which the robot has to dispose of radioactive waste, which requires avoiding many obstacles and correctly identifying the waste – explains the scientist – each robot gets a score based on its characteristics. Success These scores are sent to a computer. a Selection process It uses these results to determine which bots can play

Choosing “Normal”, yes because, by default, the objective thing is normal.

Hence, the program that simulates the operation runs Recombination and mutation of DNA on genetic models From two parents to create a new robot to manufacture, completing the triangle of life. Parents’ bots can remain within the clans, where they can participate in other breeding events, or break up into their component parts and recycle them into new bots.

No, we are not there yet: computers are not yet able to replace us and be completely independent of us. But They learn to imitate usAnd this is perhaps the most terrifying thing.

sources: new world / York University

Read also: