No computer is as powerful or complex as the human brain. The mass of tissue inside our skulls can process information at volumes and speeds that are almost impossible to reach with computing technology.
The key to the brain’s success is the efficiency of its neurons, which act as both processors and memory devices, as opposed to the physically separate units of most modern computing devices.
There have been many attempts to bring computing closer to the brain, but a new effort takes it one step further by integrating electronics with actual human brain tissue.
It’s called Brainoware and it works. The team, led by Feng Guo, an engineer at Indiana University Bloomington, gave them tasks like speech recognition and math problems like predicting nonlinear equations.
Although slightly less accurate than pure hardware computers running on artificial intelligence, the research represents an important first step in a new kind of computer architecture.
However, although Guo and his colleagues followed ethical guidelines in developing Brainoware, several researchers at Johns Hopkins University noted in a related article: nature electronics We discuss the importance of keeping ethical considerations in mind as we further scale this technology.
Lena Smirnova, Brian Kafo, and Eric C. Johnson were not involved in this study. Note“As these organoid systems grow in sophistication, it is important for the community to explore the myriad neuroethical issues surrounding biocomputing systems that incorporate human neural tissue.”
The human brain is amazing in some ways.Contains estimates 86 billion neuronson average, and Up to 1 quadrillion synapses.Each neuron is connected at most 10,000 other neuronsconstantly firing and communicating with each other.
So far, our best efforts to simulate brain activity in artificial systems have only scratched the surface.
In 2013, RIKEN’s K computer, one of the most powerful supercomputers in the world at the time, was born. tried to imitate the brain. Equipped with 82,944 processors and petabytes of main memory, it takes him 40 minutes to simulate the activity of 1.73 billion neurons connected by 10.4 trillion synapses for one second. I did. This is only 1-2% of the brain.
In recent years, scientists and engineers have sought to approach the brain’s capabilities by designing hardware and algorithms that mimic the structure and workings of the brain.is known as neuromorphic computingis an improvement, but it consumes a lot of energy and training artificial neural networks takes time.
Guo and his colleagues explored a different approach using real human brain tissue grown in the lab. Human pluripotent stem cells were induced to develop into different types of brain cells and organized into three-dimensional mini-brains with connections and structures called organoids.
These are not real brains, just arrangements of tissue that have no resemblance to thinking, feeling, or consciousness. These help us study how the brain develops and functions without messing with real humans.
Brainoware consists of brain organoids connected to an array of high-density microelectrodes using a type of artificial neural network known as . reservoir computing. Electrical stimulation transports information to the organoids. Organoids are repositories of information, where information is processed before Brainware spits out computations in the form of neural activity.
Regular computer hardware is used for the input and output layers. These layers must be trained to work in conjunction with the organoids, and the output layer reads the neural data and makes classifications and predictions based on the input.
To demonstrate the system, researchers gave Brainoware 240 audio clips from eight male speakers producing Japanese vowels and asked it to identify specific individuals’ voices.
They started with naive organoids. After just two days of training, Brainoware was able to identify the speaker with her 78% accuracy.
They also asked Brainoware to predict what’s next. Enon map, a dynamic system exhibiting chaotic behavior. They let it learn him unsupervised for four days (each day represents an epoch of his training) and found that it could predict the map with higher accuracy than artificial neural networks that do not have long short-term memory units. .
Brainoware was slightly less accurate than artificial neural networks with long short-term memory units, but these networks were each given 50 training epochs. Brainoware achieved nearly the same results in less than 10% of his training time.
“The high plasticity and adaptability of organoids gives Brainoware the flexibility to change and reorganize in response to electrical stimulation, highlighting the power of adaptive reservoir computing.” the researchers write.
There are still significant limitations, such as the issue of maintaining the survival and health of the organoids and the power consumption levels of peripheral equipment. But with ethical considerations in mind, his Brainoware has implications not only for computing but also for understanding the mysteries of the human brain.
“Although it may be decades before a common biocomputing system is created, this research provides fundamental insights into the mechanisms of learning, neurodevelopment, and the cognitive impact of neurodegenerative diseases. You might get it.” Smirnova, Kaffo, and Johnson write:.
“It may also be useful in developing preclinical models of cognitive impairment to test new treatments.”
This study nature electronics.
