Name: 這種新型人工智能由活體人類腦細胞（合成生物智能）構成 (This New AI is Made of Living HUMAN BRAIN Cells (Synthetic Biological Intelligence))
Uploaded: 2025-03-13T01:53:41.000Z
Duration: 8 min 7 s

Scientists just built a computer powered by living human brain cells, and it learns faster than any AI chip out there.

科學家們剛剛製造出一臺由活體人腦細胞驅動的計算機，它的學習速度比任何人工智能芯片都要快。

It doesn't even need a traditional computer to run.

它甚至不需要傳統的電腦就能運行。

This isn't sci-fi, it's real, it's on the market, and the craziest part, you can actually buy one.

這不是科幻小說，它是真實的，已經上市，最瘋狂的是，你真的可以買到。

All right, now, I know biological computer sounds pretty wild.

好了，我知道生物電腦聽起來很瘋狂。

We're talking about a computer made partly out of living human brain cells, actual living neurons, plus silicon hardware, working together to form what they're calling synthetic biological intelligence.

我們所說的計算機部分是由活體腦細胞、真正的活體神經元和硅硬件組成的，它們共同組成了他們所說的合成生物智能。

This new form of AI, or SBI for short, is not just some sci-fi concept, it's real, and it's been officially launched.

這種新形式的人工智能（簡稱 SBI）並不是什麼科幻概念，它是真實存在的，而且已經正式推出。

Vertical Labs revealed the CL1 at an event in Barcelona on March 2nd, 2025, and they say it's going to revolutionize everything from drug discovery to disease modeling to how we might build future robotics and automation systems.

Vertical Labs 於 2025 年 3 月 2 日在巴塞羅那舉行的一次活動上展示了 CL1，並表示它將徹底改變從藥物發現、疾病建模到我們如何構建未來機器人和自動化系統的一切。

那麼，它究竟是如何工作的呢？

The idea is that these neurons, which are grown in the lab from induced pluripotent stem cells, or iPSCs, can be placed on something like a silicon chip that has electrodes laid out in a grid.

我們的想法是，這些由誘導多能幹細胞（iPSC）在實驗室中培育出來的神經元，可以被放置在類似硅芯片的東西上，硅芯片上的電極呈網格狀分佈。

In the CL1 setup, there are 59 electrodes, forming a sort of playground for these neurons to grow and form connections.

在 CL1 設置中，有 59 個電極，形成了一個供這些神經元生長和形成連接的遊樂場。

Cortical Labs 稱其為 "盒子裡的身體"。

It has its own life support system so the cells can thrive, filtration, pumps, temperature control, gas mixing, basically an entire environment to keep the neurons healthy.

它有自己的生命支持系統，使細胞能夠茁壯成長，過濾、泵、溫度控制、氣體混合，基本上是一個保持神經元健康的整體環境。

This is then connected to software that reads and writes signals to the neurons in real time.

然後將其與實時讀寫神經元信號的軟件相連。

Because these brain cells thrive on predictable signals, they're basically searching for energy-efficient ways to respond, the system can reward them for doing something right and punish them for doing something wrong.

因為這些腦細胞在可預測的信號中茁壯成長，它們基本上是在尋找節能的反應方式，系統可以獎勵它們做對的事，懲罰它們做錯的事。

Cortical's earlier research, published in the journal Neuron, showed that, with the right stimulus, these cultured neurons learned how to play the classic arcade game Pong.

Cortical 早前發表在《神經元》（Neuron）雜誌上的研究表明，在適當的刺激下，這些培養的神經元學會了如何玩經典街機遊戲乒乓。

The researchers would send electrical signals to tell the cells where the paddle was and the cells would respond.

研究人員會發送電信號，告訴細胞槳在哪裡，細胞就會做出反應。

If the virtual ball was missed, a random, chaotic signal was sent back, which the cells didn't like.

如果虛擬球沒有擊中，就會發出隨機的混亂信號，而細胞並不喜歡這種信號。

If they hit the ball, the signal was more predictable, so over time, the neurons got better at Pong.

如果他們擊中了球，信號的可預測性就會更高，是以隨著時間的推移，神經元的乒乓球技術就會越來越好。

That first version was called Dishbrain, and it made headlines because it was basically the beginning of this idea that you could have a self-learning, living network on a chip.

第一個版本被稱為 Dishbrain，它成為了頭條新聞，因為它基本上是這種想法的開端，即你可以在芯片上建立一個自我學習、有生命的網絡。

Dishbrain was neat, but the new CL1 is way beyond that.

Dishbrain 很不錯，但新的 CL1 遠遠超過了它。

Cortical Labs has reworked the technology to be more stable, more energy-efficient, and easier to program.

Cortical Labs 對該技術進行了重新設計，使其更加穩定、節能，並且更易於編程。

The CL1 has planar electrode arrays, basically glass and metal, which simplify the process of balancing out any electrical charge that builds up.

CL1 具有平面電極陣列，基本上是玻璃和金屬，這簡化了平衡積聚電荷的過程。

They've also introduced something called the Minimal Viable Brain concept, the notion that they can figure out just which types of cells and how many are needed to create something that can, in effect, learn and adapt in more advanced ways.

他們還提出了一種叫做 "最小可存活腦 "的概念，即他們可以找出需要哪種類型的細胞和多少細胞才能創造出一種實際上能夠以更先進的方式學習和適應的東西。

If that wasn't cool enough, the CL1 only draws about 850 to 1,000 watts of power for a full rack of 30 units.

如果這還不夠酷的話，CL1 的功耗在整整 30 個機架中也僅為 850 到 1000 瓦。

A single CL1 box, about $35,000 for the hardware alone, doesn't even require an external computer to function.

一個 CL1 盒子，僅硬件成本就約為 3.5 萬美元，甚至不需要外接電腦就能運行。

It's all self-contained, with a touchscreen interface to help researchers monitor the data and keep track of experiments.

它完全獨立，並配有觸摸屏界面，可幫助研究人員監控數據和跟蹤實驗。

The plan is to produce racks that each house 30 CL1 units, forming a huge biological neural network server.

計劃生產的機架每個可容納 30 個 CL1 單元，形成一個巨大的生物神經網絡服務器。

Cortical Labs is aiming to have four of these racks online and available by the end of 2025.

Cortical Labs 的目標是在 2025 年年底之前上線並提供四個這樣的機架。

They're calling it Wetware as a Service, which basically means you can buy time on these biological computers from anywhere in the world via the cloud.

他們稱之為 "溼件即服務"（Wetware as a Service），這基本上意味著你可以在世界任何地方通過雲購買這些生物計算機的使用時間。

You don't need to own or maintain the actual hardware if you don't want to, you can just log in and start running experiments.

如果你不想擁有或維護實際硬件，也無需這樣做，只需登錄並開始運行實驗即可。

Well, for one thing, drug discovery and disease modeling, traditional systems for studying neurological diseases like epilepsy and Alzheimer's, rely on either animal models or 2D cell cultures that aren't quite as dynamic as a living network.

首先，藥物發現和疾病建模是研究癲癇和阿爾茨海默氏症等神經系統疾病的傳統系統，它們依賴於動物模型或二維細胞培養物，而二維細胞培養物並不像活體網絡那樣充滿活力。

And that leads to a lot of drugs failing clinical trials.

這導致很多藥物在臨床試驗中失敗。

Cortical Labs thinks these biological computers will open up new avenues to test therapeutics, replace some animal studies, and hopefully speed up finding treatments for diseases of the brain.

皮質實驗室認為，這些生物計算機將為測試療法開闢新的途徑，取代一些動物實驗，並有望加快找到治療大腦疾病的方法。

It might also give us new ways to fine-tune personalized medicine.

它還可能為我們提供微調個性化醫療的新方法。

Because you can grow neurons from someone's own cells, you could see exactly how that specific individual's neurons respond to a certain drug.

因為你可以用某人自己的細胞培養神經元，所以你可以準確地看到特定個體的神經元對某種藥物的反應。

Essentially, you could wire a neural network like the CL1 into a robot and have the robot learn in more human-like ways.

從本質上講，你可以將 CL1 這樣的神經網絡接入機器人，讓機器人以更接近人類的方式學習。

Instead of just feeding it zillions of lines of code or letting it chew on huge datasets, the robot's brain might figure things out with more flexibility and energy efficiency, kind of like how humans do.

機器人的大腦可能會像人類一樣，以更高的靈活性和能效來思考問題，而不是一味地向它灌輸成千上萬行代碼或讓它啃咬龐大的數據集。

And about that energy efficiency, keep in mind that large-language models like ChatGPT run on massive arrays of GPUs or specialized silicon, which can burn through a lot of power.

至於能效問題，請記住，像 ChatGPT 這樣的大型語言模型是在大型 GPU 陣列或專用芯片上運行的，這會消耗大量電力。

SBI technology, on the other hand, is using living neurons, which are extremely energy-efficient.

而 SBI 技術使用的是活神經元，其能效極高。

Think about how your own brain uses around 20 watts to keep you up and running.

想想你自己的大腦是如何使用 20 瓦左右的功率來維持你的運轉的吧。

It's a fraction of what a big data center might use for an equivalent level of machine intelligence, though equivalent is obviously tricky to define.

這只是大型數據中心用於同等水準機器智能的一小部分，但同等水準顯然很難界定。

Cortical Labs claims that a single CL1 rack only uses a bit less than 1 kilowatt, which is like running a small microwave.

Cortical Labs 聲稱，單個 CL1 機架的耗電量僅略低於 1 千瓦，相當於運行一個小型微波爐。

Now you might be wondering, is this system conscious?

現在你可能會問，這個系統有意識嗎？

According to Cortical Labs, we're not dealing with anything that's aware of its existence.

根據 Cortical Labs 的說法，我們面對的不是任何意識到自己存在的東西。

It's more of a specialized computing substrate, where the neurons are grown in a carefully controlled environment that's specifically designed to harness their computational power.

它更像是一種專門的計算基底，神經元在精心控制的環境中生長，這種環境是專門為利用神經元的計算能力而設計的。

Ethically, of course, there are a lot of questions.

當然，在道德方面還有很多問題。

Cortical Labs has said that they're following regulations from health agencies, bioethics committees, and government organizations, and they're open about the debate, especially around using human brain cells.

Cortical 實驗室表示，他們將遵守衛生機構、生物倫理委員會和政府組織的相關規定，並對有關辯論，特別是有關使用人類腦細胞的辯論持開放態度。

But they argue that these are induced pluripotent stem cells, meaning they can become just about any cell type, and they're not forming full brains with consciousness.

但他們認為，這些都是誘導多能幹細胞，這意味著它們可以變成任何細胞類型，而且它們並沒有形成具有意識的完整大腦。

Instead, they're partial networks that happen to be fantastic at learning tasks.

相反，它們是部分網絡，在學習任務方面恰好非常出色。

For those of you who are super technical, the CL1 is basically a fully programmable system with a bi-directional stimulation and read interface, meaning you can send signals in and measure signals out in real time.

對於那些精通技術的人來說，CL1 基本上是一個完全可編程的系統，具有雙向刺激和讀取接口，這意味著您可以實時發送信號和測量信號。

There's also a Python API for folks who want to integrate it into their own applications.

此外，還有一個 Python API，供想要將其集成到自己的應用程序中的用戶使用。

It's a dream for researchers who want to build custom experiments or tailor the neural network to test new theories about learning, memory, or disease.

對於那些希望建立定製實驗或調整神經網絡以測試有關學習、記憶或疾病的新理論的研究人員來說，這是一個夢想。

Cortical Labs has been working on this for almost six years now.

Cortical Labs 在這方面已經工作了近六年。

Their founder and CEO, Dr. Han-Weng Chong, and Chief Scientific Officer Dr. Brett Kagan have led the charge on bridging biology and silicon.

他們的創始人兼首席執行官莊漢文博士和首席科學官 Brett Kagan 博士一直引領著生物與硅技術的發展。

The team made waves when they embedded their cell cultures in that Pong simulation, but that was just scratching the surface.

研究小組將細胞培養物嵌入龐氏模擬中時引起了軒然大波，但這僅僅是表面現象。

Now they've got the manufacturing, the racks, the cloud platform, and a roadmap for turning this into something any big-thinking researcher or innovator can hop onto.

現在，他們已經擁有了製造設備、機架、雲平臺，以及將其轉化為任何具有遠見卓識的研究人員或創新者都能利用的東西的路線圖。

So will this lead to the next big leap in AI?

那麼，這會帶來人工智能的下一次飛躍嗎？

Synthetic biological intelligence could be the turning point, where computers get way more natural in how they learn and adapt.

合成生物智能可能是一個轉捩點，在這個轉捩點上，計算機的學習和適應能力會變得更加自然。

We're already seeing how large language models have revolutionized the way we interact with AI, but if you throw actual human neurons into the mix, the possibilities might multiply.

我們已經看到大型語言模型如何徹底改變了我們與人工智能的交互方式，但如果將真正的人類神經元加入其中，可能性可能會成倍增加。

The CL1 is here, and it's not just a concept, it's happening.

CL1 就在這裡，它不僅僅是一個概念，而是正在發生。

It'll be interesting to see how researchers and innovators put it to use.

研究人員和創新者如何利用它將是一件有趣的事情。

Mind-blowing tech, or a little unsettling?

是令人驚歎的技術，還是有點令人不安？

And if you enjoyed this breakdown, don't forget to like, subscribe, and share.

如果您喜歡這篇文章，別忘了點贊、訂閱和分享。

Thanks for watching, and I'll catch you in the next one.