Chance, F. (2021, July 30). Fast, efficient neural networks copy dragonfly brains. IEEE Spectrum. https://spectrum.ieee.org/fast-efficient-neural-networks-copy-dragonfly-brains.
FAST, EFFICIENT NEURAL NETWORKS COPY DRAGONFLY BRAINS
An insect-inspired AI could make missile-defense systems more nimble
Image: GETTY IMAGES/RICHARD PENSKA/500PX
My summary of the article:
Our brains are a masterpiece of biology, with more than 86 billion neurones working in parallel to enable what is known to us as human cognition. Humans, objectively being the smartest creatures on Earth, have an extremely complicated brain structure. However, this does not mean that relatively simpler creatures like animals have simple and easily comprehensible nervous systems – even though their brains may be less capable than those of humans.
There are many instances whereby we, humans, are genuinely surprised by how clever animals can be, like the migration of giant flocks of birds or swarm of butterflies. Scientists believe that modelling the neural system of insects are especially worthy of researching because most insects have a relatively simple neural system, yet their cognitive processes are generally focused for being capable of a specific task. Moreover, by studying how the neurones in the insects' brains perform such "specific task"s and computationally modelling the neurones' interactions, scientists also believe they could generate artificial intelligence in a more efficient way than by the conventional method of machine learning. This unique search for finding an alternative method for developing artificial intelligence stems from the fact that the computational power (energy) required to develop an artificially intelligent system is not necessarily environmental, with a very high estimated carbon emission rate in order to produce the energy required.
Researchers focused on understanding what happens in dragonflies' neural networks when they catch their prey. An important element that was peculiar about the way dragonflies hunt was that they had the ability to predict where their prey, in most cases a mosquito, would fly, and in turn was also able to locate itself such that it would intercept with the position of the mosquito in the near future. This feat that dragonflies can perform immediately inspired researchers with real life applications if they were able to computationally model this behavior, such as missile defense, self driving car control, or even pest-killing mini drones.
Chance, the writer of this article, goes on deeper to summarize the specific details of the actual neural network that the researchers believe dragons have implemented onto their brains in order to perform such a task, and upon adaptation to a computer model they found out that there is a very high probability that the model is in fact very similar to the actual model used by real dragonflies because the computer model was able to very accurately mimic the 'homing' ability of dragonflies while hunting mosquitoes. The researchers summarize their findings that this is just the start of how insect behavior can be used to develop a sophisticated form of artificial intelligence in many areas, depending on what specific task the particular insect is 'good at'. For example, by analyzing and remodelling the neural network of ants, we may be able to create an intelligent robot that could automatically find its shortest way back to its origin location after following an irregular path (this is what ants are naturally capable of, thanks to its unique neural network).
My response to the article:
I have seen footages of small insect drones that benefit from utilizing the exterior (outer, visual) features of insects, like using miniature airfoil wings and antennae. However, most of these drones do not benefit directly from mimicking the insect itself, but rather from the fact that people regard these drones as ordinary insects rather than robots manmade robots, and thus they are used in areas such as surveillance and espionage. And this was what I thought the limit of insect drones or insect-inspired technology in general was. At the end of the day, there would be no practical reason to make a drone or a robot that looks like and perhaps mimics the behavior of an insect other than to pretend that it is one to hide the fact that it is taking photos or recording a video for an investigation.
This article, however, changed my mind about how insects can be used in the field of technology and engineering. And, perhaps even more surprisingly, the article proved that insects can be used to enhance the field of technology in general not only in terms of hardware, but also in terms of software, more specifically in the ever-so-growing field of artificial intelligence. This surprised me because what the researchers were focusing upon observation of insects was not how they have adapted to the environment in terms of their external features like eyes, wings, etc, but rather how they have done so in terms of their behavior. This goes on to prove how many people, including myself, underestimate the extraordinary behavior that stems from the neural network of insects.
Commentaires