Brain Powered Artificial Intelligence

By Kenya Rosas

ChatGPT and the ethics surrounding the use of artificial intelligence has been a hot topic the past couple of years as the technology used to operate it continues to evolve. The systems used to operate tools, such as ChatGPT, are called Artificial Neural Networks (ANNs). In the present day, ANNs are in such high demand that researchers have begun to look further into biological neural networks (BNNs). But what is a neural network and what is the difference  between artificial and biological?

A neural network is a learning system that attempts to mimic the way humans think. A user provides an input and the neural network processes the data in order to provide an output that is as ‘human’ as possible. These neural networks provide outputs based on pre-uploaded data and information the network learns as it is used more. This phase in which the neural network is loaded with information is known as the training stage. In the training stage, a single Large Language Model (LLM) like GPT-3, which powers ChatGPT, requires 10 GWh which is about 6,000 times the energy a European citizen uses per year.

However, the training stage is not the highest energy consumption stage. Once the ANN is utilized on a daily basis, a simple task such as word generation can consume anywhere from 450 to 600 billion Joules per year. Word generation AI, like ChatGPT, isn’t the only function of ANNs. Subtitles, speech recognition, and virtual assistants are all features humans use on a daily basis that urge us towards finding a more energy efficient neural network. ANNs are made up nodes, sometimes called artificial neurons, that are arranged in a way that mimics a brain. The nodes send and receive signals that have assigned weights. As the ANN is utilized, the nodes provide quicker and more accurate outputs. On the other hand, BNNs utilize forebrain organoids (FOs) to process inputs and provide accurate responses.

The FOs are developed from neural stem cells using an established culture technique. The FOs are around 500 µm in diameter and are kept alive on an orbital shaker. FinalSpark’s Neuroplatform allows researchers remote access to FOs for their biocomputing research. Neuroplatform has FOs which are attached to electrodes so researchers can program signals and record the organoid’s response. They utilize a Multi-Electrode Array (MEA) organization system. Neuroplatform has 4 Multi-Electrode Arrays (MEAs) with each MEA containing 4 organoids. Each organoid has 8 electrodes, meaning each MEA has 32 electrodes.

In three years, Neuroplatform was utilized with over 1,000 brain organoids which is equivalent to more than 18 terabytes of data. At the beginning of Neuroplatform, the FOs lasted only a few hours and now they can live for up to 100 days. The advancements made at Neuroplatform lead us to question just what more BNNs can be used for.

Cortical Labs, an Australian company, grew brain cells on a silicon chip which was then wired onto a computer. Cortical Labs gave the brain cells information about how to play Pong, a video game much like tennis where the aim of the game is to hit the ball back and forth, and found the cells were able to figure out how the game worked. The brain cells were allowed to move the paddle and play the game. Although playing Pong may not be revolutionary, Cortical Labs established that fusing brain cells and silicon technology is possible and are actively working to make their product available. Who knows? Maybe we will have brain-powered technology within our lifetimes.