A "human brain computer" costing 200,000 might be the only way for humanity to overcome AI?

Author: Moonshot

In the latest season of “Black Mirror,” the first episode titled “Nosedive” features a female lead who, after an unexpected brain death, has her brain connected to a “cloud brain” service by the male lead. Part of her brain is removed and replaced with a chip connected to the cloud, for which she pays a few hundred dollars in “subscription fees” to the software company every month to maintain her “consciousness online.”

This might be the most biting satire of tech giants since “Silicon Valley.”

However, just two months after the airing of “Black Mirror,” a prototype of similar technology has quietly emerged in reality.

An Australian startup called Cortical Labs has announced the official launch of the world’s first commercial biocomputing platform - CL1.

CL1 is not an ordinary computer; it contains 800,000 living human neurons inside, connected to traditional silicon chips through precise electronic interfaces, forming a type of “hybrid intelligence.” It can not only process information but also learn autonomously, adapt to the environment, and exhibit a certain degree of “quasi-consciousness” characteristics.

Yes, you heard it right:

This is a “living” computer.

Theoretical neuroscientist Carl Friston said: “From a certain perspective, CL1 can be seen as the first commercialized bionic computer, the ultimate brain-like computer that uses real neurons.”

When people are still worried that carbon-based life forms cannot compete with silicon-based opponents like AI, will the idea of “silicon-carbon fusion” represented by CL1 become the path for what Musk envisions as “human + AI” to become superhumans?

01 When Silicon Meets Cells

Biocomputing is not a new concept. Over the past few decades, scientists have envisioned using DNA, proteins, and even cells as computational mediums. But CL1 is the first biocomputing platform to date to truly bring human nerve cells to commercial use.

Imagine 800,000 living human neurons carefully floating on a custom silicon chip. Whenever an external system emits an electrical signal, these neurons respond at the sub-millisecond level, just as naturally, swiftly, and randomly as humans receive information and react.

This is the technical core of CL1: instead of making chips imitate the brain, it directly connects part of the “brain” to the chip, combining silicon chips with living human neurons to create a hybrid intelligence system that can learn like the human brain while efficiently processing information like a computer.

CL1 looks more like a high-tech petri dish rather than a traditional computer. Its internal structure consists of three parts:

A standard rack computing node;

A microelectrode array system (MEA) that supports recording and stimulation of electrophysiological signals;

and the most important and “living” component: the temperature-controlled culture unit.

Neurons + Silicon Chips | Source: IEEE Spectrum

MEA is a bridge that connects the “human brain” and the “machine brain”. It allows electrical signals to flow freely between silicon chips and neurons, while also recording their activity patterns.

The temperature-controlled cultivation unit is the key to keeping CL1 “alive.” Each CL1 contains 800,000 human neurons cultured in the laboratory, taken from skin or blood samples of adult donors. The temperature-controlled cultivation unit provides nutrients, controls temperature, filters waste, and maintains fluid balance, ensuring that these neurons survive for up to six months.

These 800,000 neurons do not just passively respond to signals; they possess a certain degree of autonomy and plasticity, dynamically responding to feedback.

A study published in the journal Neuron in 2022 showed that Cortical Labs’ early system DishBrain had trained these neurons to play Pong (the earliest video game, Ping).

When the game starts, the neurons do not know the rules, but through continuous feedback of “hit” or “miss” with different electrical signals, they quickly learn how to control the racket to respond to the changing ball speed. The developers did not program it in advance, and the neurons can adjust their behavior to achieve the goal, which is known as the “minimal consciousness system” in neuroscience, and is a truly meaningful form of learning behavior.

In certain scenarios, the learning efficiency of CL1 even surpasses that of deep reinforcement learning algorithms, as the neurons of CL1 can grow, reorganize, and learn in real time, exhibiting dynamic adjustment characteristics similar to those of a biological brain.

You can imagine that they are not just neural tissues, but a highly plastic “living algorithm”.

The world’s first video game | Source: The Week

Moreover, the combination of neurons and silicon chips enables CL1 to possess advantages in both digital and biological domains: the adaptability and “generalization ability” of the biological brain (i.e., the ability to extract patterns from limited experiences and apply them to new situations), combined with the observability, controllability, and programmability of digital systems.

Cortical Labs provides a complete set of software development kits (SDKs) for users to interact with neurons through programming, making CL1 the world’s first “bio-computer that can write code.”

The code written by programmers no longer runs only on silicon chips, but also operates on living neurons.

So the “intelligence” of CL1 is different from any traditional hardware system; it is neither as complex as the human brain nor as flexible as silicon chips, but it represents another form of imagination regarding intelligence: Friston calls it the “ultimate form of biological simulation computer.”

The way neurons and silicon chips are combined | Source: Cortical Labs

Unlike traditional computers, CL1 does not rely on digital logic circuits, but instead trains neurons to perform tasks, resulting in extremely low power consumption and very high operating efficiency.

According to reports, a whole rack of CL1 devices has a total power consumption of only 850 to 1000 watts. In contrast, even training a medium-sized neural network model, such as GPT or an image recognition network, often requires GPU clusters that consume thousands to tens of thousands of watts of electricity, and must be kept cool to avoid overheating.

The key to energy efficiency lies in neurons, as each neuron requires an extremely small amount of energy for each discharge. The overall power consumption of an adult human brain is only about 20 watts, yet it can perform data processing, perception, and decision-making tasks far beyond that of supercomputers.

Although CL1 cannot currently write papers, program, or tell jokes like GPT-4, it can exhibit intelligent potential in specific tasks (such as perceptual decision-making and neural feedback simulation) without the need for extensive computational power.

What’s even scarier is that CL1 may also “evolve”.

02 Who would buy a “living computer”?

Even though the current paper performance of CL1 does not seem “hardcore” enough and cannot directly compete with the NVIDIA H100 at the same price point, it has the natural scalability of biology. Cortical Labs states that expanding from 100,000 to 1 million neurons incurs very little additional cost, and the cost for scaling up to hundreds of millions of neurons remains controllable.

The more neurons there are, the greater the potential for intelligence. Therefore, silicon-based computing relies on burning electricity and stacking cards for speed, while the performance growth of CL1 relies on “brain nurturing.”

“Brain in the Dish” | Source: CL1

The first batch of 115 CL1 units will be shipped this summer at a unit price of 35,000 USD, which will drop to 20,000 USD per unit for bulk purchases. The target customers are clear: neuroscientists, pharmaceutical research companies, and AI and brain-like computing research teams.

However, Cortical Labs is not satisfied with just selling the CL1 to a few top laboratories.

They launched the “Wetware as a Service” (WaaS) model. Wetware refers to the human brain and nervous system or other biological systems.

In this mode, researchers do not need to own a CL1 physical device; they can remotely log into the Cortical Labs platform to access a live neuron computing node in real-time, adjust stimulation parameters, collect data, and even conduct remote training. The weekly rental fee for each CL1 is $300.

This gives a bit of a feeling of “Black Mirror” reflecting into reality.

In other words, for $300 a week, you can rent an 800,000 programmable living human neuron. This is not about subscribing to software or renting servers, but rather renting a form of “living” biological intelligence. Although CL1 is far from achieving the complexity of human consciousness, it is indeed a form of life.

WaaS has also turned the consciousness construction module into a tradable commodity, with each neuron renting for about 0.00005 dollars per day. Does this also mean that one day, the 50-100 billion neurons in the human brain can be valued?

More boldly, will WaaS one day evolve into LaaS (Life as a Service)?

When it comes to human-machine integration, CL1 is definitely not the first, as Neuralink has already entered the clinical testing phase. Although the two paths are completely different, they both stand on the boundary between ‘carbon-based and silicon-based’.

However, Neuralink is “connecting people to computers” and trying to extend people’s computing power, while CL1 is “transforming human cells into computing”, trying to extract people’s neural abilities and feed back to machine systems.

In Neuralink’s vision, consciousness remains in the brain, merely extended and rewritten. In the logic of CL1, fragments of consciousness, learning abilities, and even possible “feelings” have become commodifiable functional modules.

Ultimately, the technological issue has turned into a philosophical question: Can the human brain truly be reshaped, summoned, or even “commodified”?

Or perhaps, when technology no longer constructs cold intelligence, but starts to learn how to live and how to survive, what should we do then?

But looking at it optimistically, this may just be a technical path, like Guan Yifan and Cheng Xin in “The Three-Body Problem,” who were forced to manually perform celestial mechanics calculations with their brains in a black domain where the speed of electromagnetic waves was greatly compressed and computational power was nearly zero. It took them decades to complete the spacecraft’s orbital adjustments and finally escape the black domain.

When traditional computing stagnates at the physical limits, perhaps “nurturing a brain” is the starting point for breaking through the technological singularity.