Cyborg brain in a dish can do maths and ‘listen’ to what you’re saying

US researchers have created a tiny cyborg by growing a small ‘brain’ in a dish and connecting it up with electronic hardware. They say this merging of computer and brain-like tissue can recognise speech, and perform complex math equations. They say the cyborg receives inputs via electrical stimulation and then sends its output via neural activity. The ‘brain’ was trained to be 78% accurate when tasked with recognising different vowel sounds, and could predict a complex mathematical system, they add.

Journal/conference: Nature Electronics

Link to research (DOI): 10.1038/s41928-023-01069-w

Organisation/s: Indiana University Bloomington, USA

Funder: F.G. wants to acknowledge support from the National Institute of
Health Awards (DP2AI160242, R01DK133864 and U01DA056242).
We also acknowledge Indiana University Imaging Center

Media release

From: Springer Nature

Biocomputing: Introducing a hybrid machine–organoid computing system (N&V)

A hybrid computing system consisting of electronic hardware and a brain organoid, which can perform tasks such as speech recognition and nonlinear equation prediction, is reported in Nature Electronics this week. The research highlights a potential approach to overcoming some of the limitations of today’s computing hardware.

Demand for computing power has increased rapidly in recent years with artificial intelligence — machine learning and artificial neural network models — being a key driver. However, as these models become more sophisticated, the energy efficiency and performance of the underlying computing hardware they are run on is struggling to keep pace. In response, researchers are developing neuromorphic computing systems — inspired by the structure and function of the human brain — that are designed to run these models more efficiently.

Brain organoids are three-dimensional aggregates artificially grown from human pluripotent stem cells, which develop brain-like tissues that can replicate certain aspects of the developing brain’s structure and function. In this work, Feng Guo and colleagues developed a hybrid neuromorphic computing system that is part traditional computing hardware and part brain organoid. The organoid was characterized by different brain cell identities, including early stage and mature neurons, and the early development of brain-like structures (such as ventricular zones) for the formation, function and maintenance of neural networks. The organoid receives inputs via electrical stimulation and sends outputs via neural activity. The authors incorporated the organoid into a type of artificial neural network known as reservoir computing, where it served as a dynamic physical reservoir layer that can capture and remember information based on a sequence of inputs. Normal computer hardware was used for the input and output layers, with the output layer being trained to read the reservoir layer and make predictions or classifications from the original input data.

The authors demonstrated the ability of their system to be used in speech recognition. Here, the hybrid computing system was tasked with recognizing an individual’s Japanese vowel sounds from a pool of eight different male speakers (240 audio clips were used). The system improved with training reaching an accuracy of approximately 78%. The system was also used to predict a Hénon map — a nonlinear dynamic system in mathematics. When compared with artificial neural networks with long short-term memory, the system showed slightly lower accuracy when using the same dataset.

The authors note that the brain organoid forms only one part of the system and more complex artificial neural networks have yet to be demonstrated. In an accompanying News & Views, Lena Smirnova and colleagues write, “As the sophistication of these organoid systems increases, it is critical for the community to examine the myriad of neuroethical issues that surround biocomputing systems incorporating human neural tissue. It may be decades before general biocomputing systems can be created, but this research is likely to generate foundational insights into the mechanisms of learning, neural development and the cognitive implications of neurodegenerative diseases.”


Leave a Comment