The successes of modern computing arise from an intuition: replicating in our computer chips the functioning of the most powerful calculator ever created by nature, the human brain. It is on these artificial neural networks that, for example, the most powerful artificial intelligences of today are based, such as ChatGpt or Dall-E, which now almost seem to transcend the capabilities of their creators. Now, a Swiss startup has decided to complete the tour, taking even deeper inspiration from our nervous system to create a neural network that contrasts silicon chips with biological mini-brains, aggregates of neurons created in the laboratory that use a fraction of the energy required from a traditional computer.
The energy issue, on the other hand, is perhaps the only weak point of the latest generation artificial intelligences. To get an idea, just think that a large language model like ChatGpt-3 (precursor of the more modern Gpt-4) needs to around 10 gigawatt hours, which corresponds more or less to six thousand times the energy that an EU citizen uses in a year. And obviously, energy consumption doesn’t stop once the neural network is trained, but continues to be high throughout the entire time you use it. It is no coincidence that it is expected that by 2030 the AI sector alone will represent 3.5% of global electricity consumption.
For its part, our biological brain is much more efficient: it needs just 0.3 kilowatt hours to operate 86 billion neurons. This is why the Swiss startup FinalSpark has decided to develop a computer based on biological chips, or rather “organoids”, agglomerations of cells that reproduce the functions of a human organ (in this case a brain), widely used in the field of research biomedical.
The field is defined as wetware computing, and in the case of the device developed by FinalSpark, as described in the journal Frontiers in Arificial Intelligence, the calculations are performed by 16 mini-brains connected to each other by a network of electrodes, and powered by a special microfluidic system, which provides water and nutrients to cells. According to its creators, it consumes one millionth of the energy needed by a traditional computer.
However, the capabilities of the biological computer, at least for now, are not even remotely comparable to those of a digital AI. But since it is an absolutely new field of study, it is not surprising: for now, in fact, the platform is only used to carry out wet computing experiments for research purposes. “In the last three years, our Neuroplatform has used over a thousand brain organoids, allowing us to collect more than 18 terabytes of data – write the authors of the study – and in the future we plan to extend the capabilities of our platform to manage a wider range of experimental protocols relevant for wet computing, such as the possibility of injecting molecules and drugs into organoids to verify their effects”.