One of the main drawbacks of current AI systems is the amount of energy required to fuel the vast amounts of computing power they need to operate. This has led to the seemingly absurd situation of tech giants such as Microsoft and Google moving to secure the outputs of whole nuclear power stations to support their activities. The human brain, by comparison, is vastly more energy efficient in its operation, though AI clearly outstrips it in the potential scale of its activities.

Dr Panayiota Poirazi’s Dendrites laboratory (www.dendrites.gr) at FORTH’s Institute of Molecular Biology and Biotechnology (IMBB) in Heraklion has for many years been engaged in the study, analysis and digital replication of the action of dendrites – a critical neurological component of the human brain. New research in the laboratory is opening the way to AI systems which will much more closely replicate the activity of the human brain, gaining in accuracy and substantially reducing their energy consumption.

What are dendrites?
In an online lecture to the Cyprus Biological society in March 2022, entitled “Dendritic contributions to biological and artificial learning”, Dr Poirazi described the nature and function of dendrites as follows:

“For those of you who are not neuroscientists this is a neuron, the principal computing unit of our brain. And it consists of these thin extensions which we call dendrites that project from the main cell body of the neuron. And this is where the neuron receives incoming signals. Signals land on the dendrites, they are processed at the cell body and then once they reach a threshold at the axon potential site (the axon hillock), the signal that is produced by the neurons travels down the axon, and when it reaches the axon terminal this signal, which is an electrical signal, is transmitted to the next neuron, the target. So dendrites are the input side of a neuron, the axon is the output side and communication happens through the propagation of electrical signals …”

What do dendrites do?
Posing the question, “Why should we care about dendrites?” Dr Poirazi answered:
“1) The properties of dendrites are altered in many brain diseases and they may serve as a target for new treatments.
2) They are the main communication site of neurons: synapses [connections between neurons] are formed primarily in dendrites.
3) They support local, nonlinear computations and greatly influence signal processing in the brain.”

The lab’s research has involved simulating dendrites on a computer, analysing their properties and experimenting with how they react to different stimuli or inputs. While for many decades dendrites were thought to have a purely linear response to stimuli – i.e. the output from multiple inputs would simply be the sum of those inputs, and any computing of the output would take place in cell body – it has now been established that dendrites themselves are capable of processing inputs and carrying out computations. For example, dendrites can be both “excitatory” and “inhibitory”: certain types of dendrite, when the input reaches a certain threshold, will provide a non-linear response, either increasing the output above the linear level, or conversely inhibiting it to below that level. The result is that certain computations relating to learning and memory can be carried out using fewer resources. “Non-linear dendrites allow a network to learn the same amount of information using fewer resources with savings in energy, the size of the memory and neuron activity,” Dr Poirazi said.

The application to AI
Under Dr Poirazi’s supervision, a post-doctoral researcher at the Dendrites lab in IMBB, Dr Spiros Chavlis, has developed a new kind of neural network which employs the properties of dendrites in image recognition – a process which requires large amounts of computing power using current AI methods. The research results have been published in a paper in the scientific journal Nature Communications, which is summarised in a FORTH press release, as follows:

A new, brain-inspired AI technology boosts efficiency and reduces energy consumption
Researchers at the Institute of Molecular Biology and Biotechnology (ΙΜΒΒ) of FORTH have developed a new type of artificial neural network (ANN) that incorporates features of biological dendrites. This innovative design allows for accurate and robust image recognition while using significantly fewer parameters, paving the way for more compact and energy-efficient AI systems.

Artificial Intelligence (AI) plays a crucial role in driving innovation and improving efficiency across various industries, offering smarter solutions to complex problems and enhancing our daily lives. However, current AI systems are huge, comprising millions-to-billions of parameters, thus consuming massive amounts of energy, which limits their widespread use. By integrating neuro-inspired features into AI, we can create smaller and smarter systems that mimic how our brains process information, improving their effectiveness in recognizing patterns and making decisions. This leads to more efficient and effective AI applications.

Dendrites are the branched extensions of nerve cells that resemble tree branches. Their main function is to receive information from other neurons and transmit it to the cell body. For many years, the role of dendrites in information processing was unclear, but recent studies have revealed that they can perform complex calculations independently of the main neuron. Additionally, dendrites are essential for the brain’s plasticity, which is its ability to adapt to changing environments.

In a recent article published in the esteemed journal Nature Communications, Dr. Panayiota Poirazi’s team at the Institute of Molecular Biology and Biotechnology (IMBB) of FORTH proposed a novel architecture for artificial neurons that incorporate different features of biological dendrites, and tested it in various image recognition scenarios.

Artificial Neural Networks (ANNs) with Dendrites
This figure illustrates the structure of artificial neurons with dendrites, inspired by biological neurons. Compared to traditional ANNs, dendritic ANNs demonstrate improved performance in image recognition, characterized by lower energy costs, reduced network size, and reduced overfitting.*

The future of AI
Currently nations as well as commercial organisations are engaged in a headlong rush to develop AI, in the fear that they will be left behind. At the AI Action Summit held in Paris this week, President François Macron announced that French companies will invest €100 billion in the development of AI systems, while European Commission President Ursula von der Leyen announced that Europe will mobilise €200 billion for the same purpose. The general feeling seems to be that throwing money at the problem of quickly developing current AI models is bound to produce a result, though it is not clear what this will be.

However, by imitating more closely the activity of the human brain, it seems that AI has the possibility to develop in a way which is far more sophisticated than the current approach. If this development takes place it is possible that in the not-so-distant future, the current methods with their unwieldy hardware requirements and tremendous energy demands will seem no more than a crude first attempt at developing a real machine intelligence.

Sources:
Dr Poirazi’s lecture to the Cyprus Biological society: https://www.youtube.com/watch?v=hL6Wgxc5aJY
FORTH press release: https://www.forth.gr/en/news/show/&tid=2788
An abstract of the original article can be found here: https://www.nature.com/articles/s41467-025-56297-9

* Overfitting (in statistics): “The production of an analysis which corresponds too closely or exactly to a particular set of data, and may therefore fail to fit additional data or predict future observations reliably.” – Oxford Dictionaries.