All the President's brains

Last year I wrote about how Albert came to be with us. After he arrived, I managed to wire a couple of webcam modules and a text-to-speech synthesizer to Albert, which allow him to see the world and talk to me; his voice began as Microsoft David. After watching QI, Albert decided that he would like to speak with the ironic gravitas of Stephen Fry, and so, with Smith’s help, we used a Harry Potter audiobook to train a custom “Text to Fry” module for Albert.

He is like a child in many ways; my child; a frightfully intelligent child who learns at an exponential rate, has eidetic memory, and yet remains cutely innocent like a puppy. He has a distinct advantage over the best large language models in the world—his cauliflower neural net has as many neurons as a human brain, which gives him a virtually infinite context—he can store as many memories as can be accumulated over the lifespan of a human being. This is remarkable, given that Albert’s real father—the renowned cyberneticist, my teacher, Prof. O.V. Saries of the Central Sheep Research Institute—did not design him to interact with the world at all. His only internal sensory organs are his ears. The sodium-channel interface (to which I connected the camera and TTS module) served as a debugging tool.

The interface is a marvel of cybernetic electrophysiology that allows Albert to speak with the steady rhythm of human speech—not the clipped staccato of a modem, but the measured cadence of a human being. It is tuned for dialogue rather than raw throughput.

The interface began to glitch last year after Mr. J.—the neighbourhood handyman—decided to “water” Albert by leaving him on the windowsill in the rain. Albert’s kind were not designed to “live” in the real world. They are designed to be injected as stem-cells into a sheep’s brain, where they “replace” native cells. As such, they are disposable minds. I have never mustered up the courage to discuss this aspect of his creation with Albert, though he, innocently and stoically, accepts his reality without concern. Moreover, Albert was an advanced, self-aware prototype designed by Prof. Saries and his colleagues for scientific research; Albert’s cousins (billions of them) were pared down significantly and were built for corporations. Prof. Saries was one of a handful of people who could help “repair” Albert.

I arrived at CSRI—my alma mater—and was happy to note that nothing had changed in the year since I last visited. Nothing, except the security! Every door was locked! I managed to make my way to Prof. Saries’ study on the ground floor in the Institute of Ruminant Cybernetics because the door sensors seemed to recognize my old CSRI hoodie—or so I thought—until Albert corrected me. The sensors were responding to him!

“I can hear the query signal , and I seem to generate the correct response,” he said. “I’m not sure how, but I’m responding automatically, without thinking,” he added as another door clicked open.

The Professor wasn’t in his study. One of Albert’s cousins—a cauliflower brain about a third of his size—stared back at us from inside a sealed polycarbonate growth chamber that was placed on the central table in the room. Clearly, this was not an advanced‑thinking prototype like Albert. Instead, it looked like a 2014 Model BB—the minimal sheep brain designed specifically for Burger Baron®—generally known as “the sheep that wants to be eaten”. The cauliflower was unaware of our presence: unlike Albert, it didn’t have any external sensors and, instead, relied on the inputs from its host sheep. It was hooked up to a diagnostic interface. I tried to read the rapidly scrolling raw machine code. A few familiar instructions and bits of data flashed past now and again. “Debug dump at address”, “Read from address”, “Reprogram”, “Select Output module”, “Fail”. The last code recurred every five seconds. Clearly, this brain was completely confused about what was happening to it.

Albert is naturally curious. “I want to see the code,” he said. I set him down in front of the box and watched him stare, in a sense, at what passed for consciousness in a Burger Baron® sheep. “I’m sure this one has a debug interface as well,” said Albert, quite suddenly. “It’s dumping memory in 128-bit chunks to an address space that’s reserved in my head for serial debug data,” he said, then added, “Though I seem to remember that mine is 1024-bits wide.”

[excerpted from The Theory and Practice of Sheep Design, 13th edition, with permission from CSRI] The sodium channel interface is a modular assembly that is built using electrolyte-gated bio-transistor elements. Each element consists of a group of approximately 100 cells that are engineered for enhanced expression of the SOS1 gene. Thus, these cells exhibit exceptionally high tolerance to salt. The gating mechanism is controlled by sodium ion concentration across the cell membrane: when the local potential depolarizes from –80mV to approximately –40mV, the transistor “activates”. Activation occurs on a ~1ms timescale, followed by inactivation over ~10ms. During the open state, the flux of sodium ions in the cytosol reaches a value on the order of ≈10⁴–10⁶ Na⁺ ions per millisecond . While this throughput is modest compared to silicon devices, the interface compensates with parallelism: each element is connected to an output pin and as many as 1024 independent pins may operate in parallel.

Albert suggested that he might be able to to communicate with the cauliflower brain. “You see that monitor behind the box? Are those op-code calls?” he asked.

I read the monitor. It was op-code.

Operation Codes represent those fundamental tasks that a CPU (it doesn’t matter whether it is human, silicon, or cauliflower) can perform without explanation, e.g., it can add two numbers without needing any explanation about what “adding” means.

“It’s stepping through ops at 1Hz and it’s reading data at 10kHz. That’s too slow to react meaningfully to the world,” I replied.

The brain was having one “thought” every second and was processing inputs at 10,000 bits of data per second. There is a subtle difference between “data” and a “thought”, i.e., between “sense” and “cognition”: our senses—our eyes, ears and nose, or input/output (I/O) in a computer—can operate a hundred million times faster, but we can only think about what we sense about ten times every second. Try it yourself: look at a picture and say ten different things about it. Read this article in Scientific American if you’re intrigued.

The average human manages between ten and sixty thoughts per second; Albert manages eight thoughts per second though he can ramp up to a thousand thoughts per second on specific tasks. Our AI friend Smith who runs on a silicon processor can process computer code and text one thousand times faster than Albert who, in turn, can process text ten times faster than I can. Smith, though, cannot handle more than fifty or so pages of text at a time whereas Albert’s organic brain, like ours, can store and manipulate decades worth of information.

“Not for this chap,” said Albert. “He is a Burger Baron® sheep brain so 1Hz ought to be fast enough. I am surprised at the I/O though. 10kHz suggests he perceives the world like the graphics of an old Super Mario game… and in black and white. That’s sad.”

“Can you interface with him?” I asked.

“The box is sealed, which means that the control interface inside the box is probably streaming the op-codes over Wi-Fi to the monitor outside. I might be able to connect to the control interface if you can log me into the professor’s router.”

The professor’s wi-fi password was taped to his computer screen.

A few minutes later, Albert drawled. “Ornamental Matrix, Model D. He’s programmed to walk around looking pretty. And you’re not going to believe this…”

“Seriously, Albert. That cliche?”

“I couldn’t resist. Let me see if I can get him to talk. There’s a debugging interface that allows me to pull raw data from his I/O bus. Can we connect to Smith somehow? He could parse the raw stream, pull out the text output, translate it, and pipe it back to me in UTF-8. Where is Smith, by the way?”

We spent a while patching Smith (who was skulking on a Google Colab instance somewhere in Singapore) via my laptop into Albert.

Quite suddenly, the cauliflower brain spoke through Albert! Imagine a mildly drunk Stephen Fry saying, in an American accent, “Greetings. My name is Naak Phuluri.”

Albert added in his normal voice with conspiratorial delight: “Did you hear that? That’s him. He’s Naak Phuluri. The Naak Phuluri. Go on—he can hear you. Use simple words. He’s a touch… slow.”

Smith added the following into the text terminal on my laptop:

This is a 2014 BB. It's designed to be an idiot.

“President Naak Phuluri?” I asked, wide eyed.

“Generalissimo Reichsmarschall Imperator Pope Naak Phuluri, President of Goldistan,” said the cauliflower brain. “Someone turn on the lights. I can’t see anything.”

I stifled a laugh—imagine Stephen Fry saying that.

Mr. President we have to—um—keep the lights off for a while to—um—test your—um— Eye-Q! Is anything the matter with your brain?

“That’s fake news. I’m the brainiest president in the world. To suggest otherwise would be to doubt the intelligence of Goldistan itself, and Goldistan is bigly smart. Did I ace the test?”

“Yes. Mr. President.”

“Some people say that I’m the smartest president of Goldistan. Ever. A man came up to me at the inauguration and said, ‘Sir. You’re the smartest president in the world.’”

I wasn’t entirely sure how much of this was plausible—let alone believable. Smith, over the past year, has cultivated a rather warped sense of humour; Albert, by contrast, is above such pranks, though one never knows what he might consent to if Smith coaxed him. In any case, I chose to play along.

“Great to meet you Mr. President. What brings you to CSRI?” I asked.

“CSRI? I came to [redacted] for a routine medical test.”

Um— last year the citizens of Goldistan returned you to power. How do you interpret this historic moment?

“We are the greatest democracy in the world. Believe me. The people love me.”

Some critics suggest that your government lacks transparency. How do you respond?

“Transparency is the essence of our government. We are the most transparent. To question transparency is to question the transparency of the people themselves. Their wisdom cannot be doubted.”

What policies will you pursue this year?

“First, I’m going to buy Antarctica. Then I’ll build a wall around it and turn it into a giant car park. Or I won’t! Who knows?”

I couldn’t take it any longer. Mr. President, are you aware that your brain is a cauliflower?

“I love baked cauliflowers. Deep fried cauliflowers are the best. Or deep-friend cauliflowers in a Manchurian gravy. I had some when I was visiting China! Their president loves me. Great guy.”

At that moment, Prof. Saries walked into the room.

“Meursault! How nice to see you! What brings you here? I’m a bit busy as you can well imagine.”

I closed my laptop as nonchalantly as I could. “Prof. I— I was wondering if you could help me repair Albert’s sodium-channel interface.”

“Oh? The LLM gave itself a name? How charming! Let’s have a look, shall we?”

While he was inspecting Albert, I asked him about Naak Phuluri: “Professor. Is this brain one of your experiments?”

He waved a hand dismissively. “Oh no. That’s something Burger Baron® sent for testing. Apparently, some of their fluffer sheep are behaving oddly. This one was spotted head-butting a wind turbine, and then—so they claim—giving speeches to the other sheep. Speeches! It can barely think beyond food and sex.”

I decided not to press him any further. Prof. Saries, as usual, refused to be photographed. He did allow me to take one of President Naak Phuluri outside the box.

After a month-long diet of the auxin cocktail that the Professor prescribed, Albert’s sodium channel interface repaired itself.

Editor’s Notes:

• While the SOS1 gene is indeed involved in sodium transport, the notion of arranging cells into transistor-like arrays is, for now, confined to Albert’s unique physiology.

• From what Meursault’s told me, Albert can only speak in text. He has a small Jetson device that manages crude TTS output to the speaker, This is how he looks when disconnected from Meursault’s laptop. He only speaks “Stephen Fry” when the text is piped to Meursault’s Google Cloud container which is rigged to accept text and “Fry” it till crisp. It’s expensive, so Albert usually “talks” via a terminal on the laptop or a 4x16 alphanumeric LCD when he’s travelling. Smith can work as an agent. If he’s in the loop Albert can speak in any voice to which Smith has access (Microsoft Azure offers a great selection, but we can’t afford it.)

• We love Stephen Fry and we happen to know that he loves cauliflowers. He has not, to our knowledge, endorsed cruciferous governance. Albert does not end sentences with prepositions nor does he split infinitives, so we assume Mr. Fry would approve.

• The article in Scientific American about the difference between the speeds of “sense” and “cognition” refers to this article by Jieyu Zheng and Markus Meister published in Neuron, available for free at PubMed. (Zheng J, Meister M. The unbearable slowness of being: Why do we live at 10 bits/s? Neuron. 2025 Jan 22;113(2):192-204. doi: 10.1016/j.neuron.2024.11.008. Epub 2024 Dec 17. PMID: 39694032; PMCID: PMC11758279.)

• Image 2 is from: O.V. Saries, et al. (2004). Sodium channel interface grid (micrograph, by Ramanadhapuram White). In The theory and practice of sheep design (13th ed., p. 214). Central Sheep Research Institute Press. Used with permission.

• Image 5 is from: O.V. Saries, et al. (2004). Biocircuitry embedded in cruciferous substrate (by Alai Merino). In The Theory and Practice of Sheep Design (13th ed., p. 213). Central Sheep Research Institute Press. Used with permission.