The ‘Green Brain’ project is a £1 million collaboration between the Universities and will develop our understanding of how brains work and advance the field of artificial intelligence. Furthermore, it may pave the way to modelling more complex brains like our own. The name alludes to an IBM sponsored project called ‘Blue Brain’ which is developing ways to reverse engineer human brains using supercomputers.
The aim is to create a flying robot that mimics a honey bee’s vision and sense of smell to produce authentic bee-like behaviour. Possible applications of robotic bees might include search and rescue missions and even artificial pollination of flowers.
The project will use parallel processors provided by NVIDEA called ‘GPU accelerators’, the same type of processors used to generate graphics for video games.
Dr. Thomas Nowotny, leading the Sussex University team behind ‘Green Brain’, spoke exclusively with Humans Invent about the project.
What are the project’s overall aims?
Dr Thomas Nowotny: Our ultimate aim is to work towards a better understanding of how brains work and how they generate successful behaviour in interaction with the body and the environment. We have chosen to work on the bee brain because it is smaller and less complex than the human brain and a wealth of experimental data is available both behaviourally, electro-physiologically and anatomically. Within the project the main aims are:
a) To build an accurate as possible model of the bee’s olfactory system (sense of smell).
b) To build a similarly accurate model of the visual system.
c) To integrate the models and combine them with a behavioural model and implement this to run in real time on a GPU based supercomputer and control a flying robot in real time with this “artificial brain”.
How complex is a bee brain? Where would you place it on a scale of complexity starting with the simplest of brains and ending with human brains?
The bee brain has about 1 million neurons, compared to 302 in the simple worm C. Elegans and about 100 billion in the human brain. In terms of number of neurons, then, it is somewhere in the middle between the worm and the human. However, in terms of cognitive abilities, bees have been shown to be quite sophisticated, comparable with the behavioural repertoire of a rat.
Can we compare brains to processors?
Directly, no. But we can use brains and processors to do some of the same tasks. Think for example about playing chess. In that example the task performance of the human and the artificial systems are comparable. But obviously, the brain is very good in many different tasks and in new situations which the artificial systems are often not.
What are the biggest challenges of a project like this?
The biggest challenge is to decipher how the brain of a bee actually works and then build a suitable model for it. It would be a tremendous achievement if we were able to predict the behaviour of real bees from the output of our model in a new task that has not been tested before on bees.
How far off are you from accurately modelling a bee’s brain? Have simpler brains been modeled?
There are models of various parts of the brains of various animals but as of yet it has been difficult to formulate truly predictive models. One of the presumably most simple problems would have been the worm C. Elegans but due to experimental restrictions (the worms explode when one tries to record electrically from their neurons) this has not been accomplished.
From my research and the knowledge of the literature I would say that we are still decades away from modelling the human brain to any degree of accuracy. However, Prof. Henry Markram has been proposing the “Human Brain” project as a European Union Flagship with exactly this goal to be completed within a few years.
What would be the applications of robotic bees?
Our robot is a research instrument to be able to test our brain models in the real world with a “real body” and “real sensors”. The robot will not look like or have the dimensions of a bee. That being said, robotic bees in their real size are being developed in the Robobees project led by Harvard University in the States. A possible application of our robot could be the autonomous monitoring of the environment, e.g. for dangerous volatile substances.
What other things would this research pave the way for?
We would hope that the research will be a step towards better brain models and hence also a better understanding of disease and dysfunctions in the brain. Another direction would be to help the development of more autonomous bio-mimetic robots generally. Understanding the olfactory system in its own right might be of great importance to the future development of artificial noses.
We are building a model of the bee brain to be run on a GPU accelerated computer and we will use it to control an autonomous flying robot. The instructions from the artificial brain will be transmitted through a wifi connection as will be the sensory data from the robot to its brain. The flying robot will not be developed by us but modified from a commercially available quadrucopter, sold as an expensive toy.
As well as expanding our understanding of brains, the Green Brain project and the Robobees project at Harvard might also one day facilitate artificial pollination of flowers, and as the global bee population continues to mysteriously fall, solutions to this problem are becoming more pressing, as it is central to crop production and therefore our survival as a species. Expect to see many more autonomous flying robots in the near future as use of surveillance drones increases and the US military augments its fleet with autonomous drones. Watch the skies.