Imagine a computer that does not rely on silicon chips that consume large amounts of energy, but rather on living human brain cells grown in the laboratory from neurons derived from human stem cells that have the ability to adapt and learn with mechanisms that resemble some of the characteristics of the human brain and with an efficiency that exceeds any traditional electronic system.
This is not a scene from a science fiction movie, but rather the reality of research into “Biocomputing” technology, which combines biological tissues with electronic chips to produce hybrid computing systems.
This technology seeks to create systems inspired by nature’s method of processing and learning information by integrating human brain tissue grown in the laboratory with electronic chips as a potential solution that opens new horizons in reducing energy consumption and increasing the speed of learning.
The emergence of “hybrid biological computing”
For many decades, silicon has been at the forefront of the digital revolution, as this element extracted from the sands of the earth contributed to making the processors that operated airplanes, moved financial markets, and embodied the generative artificial intelligence revolution.
Currently, silicon chips are approaching their maximum physical limits, as the problem is not their ability to compute, but rather their hunger for energy.
Training some large language models consumes electrical energy equivalent to the consumption of dozens or hundreds of homes annually.
As Moore’s Law slows down and transistors reach atomic scales, their miniaturization becomes more difficult and expensive due to physical and thermal limitations.
Here comes the role of “hybrid biological computing,” which relies on miniature, three-dimensional brain tissue grown in the laboratory from multipotent human stem cells called “brain organoids.”
These “miniature brain organs” grow into a complex network of cells and synapses capable of receiving and transmitting electrical and chemical signals, mimicking aspects of the structure and function of the human brain.
When “miniature brain organs” are combined with electrode arrays, so-called hybrid bioelectronic systems are produced, which combine living biological components with electronic devices.
These systems convert digital data coming from the computer into electrical impulses that neurons understand, and re-send the electrical responses issued by the neurons when they process the information to the computer system as a result of the calculation process.
This means that living human cells now perform arithmetic and logical operations instead of silicon transistors.
During 2023, a team of Johns Hopkins University researchers will propose the concept of “organoid intelligence,” which aims to exploit the natural learning and memory capabilities of “miniature brain organs” grown from human stem cells for computing purposes.
This concept is based on connecting “miniature brain organs” to advanced electronic interfaces that allow information to be sent to them and their responses received, thus opening the way for the development of biological computing systems that are more energy efficient than traditional computers.
The potential uses of this technology are not limited to computing only, but extend to testing new drugs, simulating neurological diseases, such as Alzheimer’s and Parkinson’s, studying human learning and memory mechanisms, as well as developing artificial intelligence systems that are more able to adapt and learn from limited amounts of data.
Energy crisis in the age of artificial intelligence
The International Energy Agency (IEA) reports that data centers currently consume about 415 terawatt hours per year, or about 1.5% of global electricity consumption.
This number is expected to rise to about 945 terawatt hours by 2030, equivalent to the current consumption of a country such as Japan, a number that constitutes approximately 3% of the global total.
Artificial intelligence is the main driver of this growth, especially with the training of large language models that require significant energy to cool and operate.
The Frontier supercomputer consumes about 21 megawatts of power, while the human brain consumes only about 20 watts. Despite Frontier’s great capabilities, it differs radically from the human brain in terms of efficiency and method of processing information.
The human brain performs complex tasks, such as learning from little data and dynamic adaptation, by consuming as much energy as it takes to light a small light bulb.
During his matches with world champion Lee Sedol, the AlphaGo system consumed about one megawatt per hour, while Sedol’s consumption did not exceed only 20 watts, a difference of 50,000 times.
This discrepancy is due to the nature of processing, as computers separate the central processing unit and memory, forcing data to constantly move between them, consuming energy and heat. In living neurons, processing and storage occur in the same place and at the same time across synapses.
Scientists seek to build hybrid biological computers that consume thousands of times less energy and learn extremely quickly while replacing silicon processors with living chips, which may contribute to alleviating the increasing pressures on energy consumption in artificial intelligence systems and move humanity to environmentally friendly computing.
Neurons are capable of rapid learning
The advantages of mini-brain organs extend to the method of learning itself, as the human brain has the ability to learn quickly from limited data, while artificial intelligence needs large amounts of data to recognize a simple pattern.
Artificial intelligence systems often need additional data or partial retraining when the environment changes, while the human brain works in a different way based on neuroplasticity. Neurons rearrange their connections and synapses based on the surrounding environment very quickly, which is one of the most prominent reasons for the superiority of nervous tissue, as the human brain is capable of learning complex patterns from few inputs.
The human brain absorbs large amounts of information using a small fraction of the energy required by conventional devices.
Biological computing may allow faster decision-making and continuous learning during implementation, with higher energy and data efficiency compared to silicon systems.
In some research tests, a large artificial neural network running on the Intel Loihi chip consumed about 3 times less energy than its conventional counterpart on story context understanding tasks.
The difference in power consumption when communicating within the same chip reached about 1,000 times compared to transferring between separate chips, which reflects the efficiency superiority of the human brain over current silicon chips.
Companies write history through pioneering experiences
Over the past few years, startups have been quick to exploit this technology and turn it into commercial products, such as Australian company Cortical Labs and Swiss company FinalSpark.
Cortical Labs launched the CL1 device, the first commercial biological computer in the world, according to the developer. It relies on merging living human brain cells with silicon electronic chips. The device includes 200,000 neurons and consumes energy equivalent to the energy of a desktop computer.
The company entered into a partnership with DayOne, a company specialized in operating and developing sustainable data centers, with the aim of expanding the establishment of biological data centres.
This was preceded by Cortical Labs’ success in training its system called “DishBrain,” which consists of 800,000 neurons to play the famous classic video game “Pong.” The cells were able to learn the game and adjust their movement themselves within a few minutes.
As for “Final Spark”, it took a different approach, as it developed the “Neuro” cloud platform, and allowed researchers to access it remotely. This platform hosts about 160,000 neurons, which opens the door to the possibility of accessing “biological computing” without the need to own equipment.
In addition to the startups, a research team at the University of Indiana Bloomington has developed the Brainoware model, an innovative biocomputing system that combines the biological tissue of the human brain with traditional electronic circuits.
This system represented a very advanced step in the field of biocomputing, as it succeeded in tasks such as predicting nonlinear equations and speech recognition, with high energy efficiency and unsupervised learning.
The results showed that the “miniature brain organs” are able to process spatiotemporal information thanks to the properties of fading memory and nonlinear dynamics.
Obstacles to biocomputers
Despite the brilliance surrounding this field, there are fundamental technical obstacles that have not yet been resolved, and the path towards replacing the silicon processor with a biochip is still fraught with significant technological and biological obstacles.
The current “mini brain organs” are small in size and contain a number of cells that is incomparable to the human brain, which includes 86 billion neurons. In addition, they need life support systems that maintain the appropriate conditions for the neurons to remain alive.
“Mini brain organs” currently survive for relatively long periods in protected laboratory environments, but keeping them stable remains an engineering challenge, and growing them, attaching them to electrodes, and connecting them to each other is still a manual and very slow laboratory process, and has not yet reached a formula that allows for commercial mass production.
In addition, this field ignites a philosophical, scientific and ethical debate about the ability of “miniature brain organs” to develop a form of primitive consciousness and the dangers of the uncalculated rush towards humanizing machines and violating human dignity through the use of human tissue.
Most researchers confirm that these organs currently do not have awareness in the human sense, and what happens in practice is similar to the response of algorithms to training, as unconscious learning in the human sense may create functional patterns without these organs feeling anything.
They also point out that the real value of “miniature brain organ intelligence” lies in creating a new class of specialized computing systems that combine the speed of electronics with the flexibility of living neural tissue.
In conclusion, the technology of “hybrid biological computing” promises increased energy efficiency and rapid learning, embodying a real ray of hope in the face of crises that threaten to stifle the progress of artificial intelligence.
Although biocomputers are still in their early stages and may take decades to reach the market in commercial form, they open a new door that may complement traditional silicon technologies or redefine some specialized computing areas in the future.
