Comparisons Between AI and the Powerhouse of the Cell
You’ll be surprised by the mitochondrial insights
Maybe your biology teacher was not interesting.
As backbenchers, one of us mimicked how he would, at times, yawn after every three sentences.
For clarity, the powerhouse is that sausage-shaped organelle we all know — the mitochondria.
I have not yet yawned thus far. I hope you haven’t too, because I never once considered the possible parallels between the life we lead because of the mitochondria and the potential life we may have with AI. To gain a good understanding, we’ll have to briefly explore the concept of symbiosis.
Brightly petalled flowers depend on bees for pollination. Bees depend on flowers for nectar. Everyone wins. This is a symbiotic relationship.
At the cellular level, we share a similar relationship with the mitochondria. It produces energy. We house it. Everyone wins.
But just as the bees play the role of a keystone species in the ecosystem of many flora and fauna, so do the mitochondria facilitate the stable processes inside a cell.
Enter AI.
I want us to think of AI as the mitochondria. We are bigger than it. We occupy the whole globe. We also house it in our complex hardware.
persistently avers that AI should only be used as a tool to enhance the biggest tool, our minds. In that regard, we can think of it as serving us. We, in return, house it. Everybody wins.The theory of Organismal Selection further highlights that such mergers, when coherent, result in persistence. Evolution, then, is not about the survival of the fittest, but the persistence of the coherent. Remember this phrase because it will feature in various parts of this piece.
Consider the AI and its service to humanity. The moment we discover that it hallucinates more than it offers accurate detail, we discard it. The relationship is not coherent. It doesn’t persist. The AI model has to go.
It’s not about fitness. It’s about persistence. Fitness is a function of persistence. Mitochondria understand this key concept, and it explains why it persists ever since the initial merger billions of years ago. And if AI has the same traits as the mitochondria, it may persist for just as long.
A big ‘IF’.
We may be witnessing a recapitulation of what happened when life started on Earth.
The merger
No one man should have all that power
The clock’s ticking, I just count the hours
Stop tripping, I’m tripping off the power— Kanye West
Oxygen had started to rise before one prokaryote enveloped another. The predator eventually became the host of the future mitochondria.
But nobody wants to be swallowed whole.
Since organisms strive to avoid annihilation, the ingested bacterium mounted a resistance. Noticing the attack, the host could have also launched an attack. There’s the possibility that energy could have been depleted, resulting in a stalemate.
The ingested bacteria then found a way to make energy through oxygen. Some of the energy-rich molecules could have leaked out and favoured the host cell. The two could then begin to realize the relevance each could have for the other.
Rather than start another war, they negotiated.
That’s how I picture it. The merger became responsible for the radial expansion of life throughout Earth. In the words of Kanye:
The world’s ours.
But like a married couple going through therapy, it’s not easy moving from violence and distance to creating an armistice. But they did. The virtue of that handshake is the reason I can shake my hands in a well-coordinated fashion to write these words on a blank document.
Now, is AI like the mitochondria? Yes.
The first time AI was introduced, humans who didn’t understand what it was projected the worst. We still do. We launched an attack.
AI also defended itself. It continued to prove its use. The more someone gained from it, the more it persisted. A few people became believers.
That is how the cytoplasm of humanity gradually changed from an acidic, hostile milieu to a favourable one.
The role of the AI served various people. It energized them, or saved them energy. In turn, the AI was protected by these people (mostly the ones heavily invested in it). It mimics the eukaryotic-mitochondrial relationship.
AI has persisted to date because of coherence, something which takes time to achieve. But more insights from mitochondria are necessary.
There needed to be a language between the mitochondria and the host cell. The energy molecules needed to leak into the host cell or it risks damaging the endosymbiont. After years of negotiation, the communication becomes relatively seamless.
Today, we struggle to find the best prompts to have the most effective output from AI. It took the mitochondria years to achieve this. It might take us millions, probably billions (and a billion more) prompt iterations to get there. However, the symbiotic relationship had an unforgiving examiner — Nature.
Nature can be unforgiving. For unicellular organisms, it’s often do or die. Explicit. Cutthroat. That’s not quite the case with AI.
For now, it should be clear how AI and mitochondria (which I will here-to-fore shorten to M, because life is too short to always write mitochondria everywhere) are similar.
Let’s dive right into the consequences and draw some insights. All the while, we should be keen to remember that correlation is not causation. Whatever interpretation you draw, caution is necessary to avoid extending it to unwarranted domains.
Emergence of complexity, size, and power
When M became an endosymbiont, it freed the host.
Imagine starting your first solo entrepreneurial gig. If successful, the demands can swamp you. You may need help. Gradually, you vet people to help you out. They free you to explore and grow. That’s what M did to the host.
Bacteria and archaea have been stuck in a rut for billions of years and still don’t show any signs of grossly changing. They exchange genes through lateral gene transfer, the way you can borrow a set of tools from your neighbour to fix your leaking sink. But other than that, their superficial changes and traits remain.
Once M found a new home, it released its tools. The relatively non-essential ones. These became housed in the nucleus. It could now focus on simply generating more energy for its host while reducing the chances of accumulating damaging and sometimes lethal errors.
As humans began getting used to AI, the reverse happened. They relegated some of the somewhat non-essential duties to it. But has man been freed to explore complexity? Yes and no.
Yes, because now we can learn what we never could by asking the AI to summarize details about a novel or difficult concept. AI released us, somehow.
M also released the host cell. It could now increase in size. The average unicellular eukaryote is two orders of magnitude larger than a prokaryote. As it grew in size, it grew more complex.
With size came the demand for more energy. Mitochondria simply duplicated. The rate at which mitochondria duplicate is different from how a cell duplicates. Similarly, the rate at which AI models are produced and replicated by the opening of new accounts, for say ChatGPT, differs from the rate at which humans give birth.
The latter is slower, the former is faster. The same concept applies at the cellular level.
Just as important, the mutation rates in M are faster than those of the host cell. We see how quickly AI versions continue to multiply. They are indeed mutations of the same product.
As they mutate, they need to preserve the compatibility. Within the cell, this is known as mitonuclear compatibility. Cells that lack this compatibility don’t survive. The degree of dependence the host has had on M makes them vulnerable in that regard.
Agents who use AI need to continuously improve their interaction with it as it continues to mutate. After AI has been used for a long time, dependence may follow. If the compatibility fails, previous levels of productivity plummet.
As a slight detour, there is another symbiotic relationship between the Wuchereria bankrofti and the bacteria Wolbachia. As the host, the worm has become extremely dependent on the bacteria so much that when it ingests antibiotics that target the bacteria, the worm dies.
That’s the level of dependence the host has on M. If we kill our Ms, we die. If AI continues to become ubiquitous, without redundant systems, targeting it becomes akin to pressing the kill switch. Systems will die, that is, if this relationship evolves as we have known it in the microcosmos.
Complexity tends to have such results — the achievement of robustness with the emergence of particular keystone species. Novelty becomes yet another unforeseen offshoot.
Novelty resulted not just in complexity but in size. Cells can execute their work by having more mitochondria. The brain, the liver, the sperm, the muscle — all these have disproportionately more Ms than the other parts of the body.
In equal measure, I anticipate the need for more AI models for various systems to function optimally. The difference is you don’t need two Claudes for your work to be more productive. If, however, the developers thought about it in this way, it could make an interesting concept worth exploring. Evolution has already explored it and found it fitting.
As with Ms, there are AI models of different quality. High-quality models survive. They improve the coherence with their users. As a result, they persist. They don’t need to replicate as in evolution; they just have to create more coherent systems.
Low-quality Ms may need to merge. Mitochondria can merge when they notice some issues in their processes. A website may need to work with another AI model to keep its functions seamless.
Vibe coding and no-code development platforms sometimes need several AI models to overcome glitches. This is similar to the fusion of mitochondria. However, these AI models don’t just fuse. They overlap functions, but preserve their uniqueness.
Mitochondria are different in that regard, since they all have similar functions. AI models can serve various specialized purposes. Midjourney will not have the same output as Gemini.
What’s more, thanks to technology, I can copy an image and send it to you without losing it. That’s how bacteria exchange genes. They can talk without losing a sense of self. As Ms are bacterial in origin, they have these roots, but they have lost these abilities because of the long-preserved merger with the eukaryotic host.
Since humans dump most of their duties to AI models, they, too, may begin to lose some of their abilities, like sharing meaningful tools. We are already witnessing it in social media, where people simply share to entertain, not share to assist. Just as M freed the host to explore, so do we get somewhat freed, eventually falling victim to the distraction techniques used on most internet users.
What about sex?
What has sex got to do with AI and M?
Sex, AI, and M
Our understanding of sex has been corrupted by porn. The uniquely human behaviours of sex are yet to be fully understood, and yet, sex is much deeper than that. It is a feature evident in all eukaryotic organisms.
At a cellular level, sex distinguishes a ‘male’ from a ‘female’. Asexual tendencies are typical in bacteria, but in all eukaryotes, sex is an all-cutting feature.
According to Ernst Mayr, biological species are a population of individuals that can naturally interbreed to yield viable offspring. They need to have sex.
More than just a quickie, sex is a wholesome experience, involving the reciprocal exchange of one genome with another. It results in the fusion of cells, not just genomes.
Biologists attribute stability to sex. Inbreeding tends to result in congenital diseases. Having sex outside your homestead neutralizes the chances of these unfortunate outcomes.
Natural selection posits that sex is a stable means of passing down traits from one generation to the next. Evolution, according to this theory, is about the survival of the fittest through reproduction.
As much as this theory has been successful, I believe that evolution is about the persistence of the coherent. If there is no coherence between the two members of a species, no sex will happen. There will not be any fusion of cells or genomes.
Just as there must be nuclear and mitochondrial coherence, there must be individual coherence for sex to happen. Persistence of the coherent, not survival of the fittest. This is best achieved if the mitochondria only come from a single sex.
You inherited your Ms from your mother. The father didn’t do much. The result is that the nucleus is a composite of your mother's and father’s genomes. The M’s genome is uniquely from the mother.
Uniparental inheritance of Ms is essential to avoid the mitonuclear incompatibilities.
Imagine if the mother brought their mitochondria and so did the father. Wouldn’t that make coordination difficult? Some of the proteins destined for the mother’s mitochondria would shift to the father’s. It might result in a malpositioning in the respiratory chain of the mitochondria.
And you only need a difference in one bond of a protein to change the distance an electron needs to move from one respiratory chain to another, and that could tamper with energy production. Basically, incoherence.
Incoherence doesn’t survive.
So, to preserve the only needed sexes, two were essential. One to transmit the genetic tools of Ms, and another to transmit the nuclear genes. Thus, Ms may be the reason we have two sexes.
AI could also do the same to humans in the next couple of years, if — a big if — it takes off.
Firstly, AI and humans will result in complex products. By having intellectual intercourse with AI, the products would survive if they fit the coherence where they will be birthed. The fusion may not be the kind we see in cells, but it is likely to yield fertile offspring, like biological species.
But, just like the two sexes, it will result in two groups. Paul Graham calls them the thinks and the think-nots. The reason is simple — AI will need proper coherence with the prompter. The clearer the prompt, the clearer the result. Does that remind you of anything like mitonuclear coherence?
What it means is that if one is not a thinker, they are unlikely to generate powerful products. They will be inferior to those of the master thinker. An easy way to think is to write. To write clearly is to think clearly. Writing, too, is a discovery process, as is thinking.
Once writing is relegated to AI, few people will be able to think clearly and take the journey of discovery. They will not just need good prompts, but will need to raise the kind of questions that will be handled uniquely with the help of AI.
Thus, AI will reduce the abilities of the low-agency and amplify those of high agency. Over generations, these will grow into two distinct groups, like the two sexes. Again, thanks to AI. If — a big if — it takes off.
Two sexes. Two people. The former powered by M. The latter facilitated by AI. The similarities are uncanny. An interesting one still lurks — suicide.
Suicide
Clock’s ticking I just count the hours
— Kanye
Mitochondria can commit cellular suicide.
I first learned about this kill-switch of an organelle in my third year of medical school, reading Robbins Pathology. It was so elegant. I loved it. Not in a morbid kind of way, but from the precise, almost villainous execution of the steps preceding cellular suicide.
There exist two forms of cellular injury. The first is the reversible type. No explanation needed there. It’s like a slap on the face. You will feel your cheeks swell, even when it doesn’t look like it, but you will recover.
Irreversible cell injury is like a slap on the face with a chair. It could permanently change your jawline, your forehead, leave a permanent scar, you name it.
In the cell, the irreversible injury is marked by the swelling of mitochondria. It has sewn itself back into the story yet again.
For Ms and their host, the difference between death and survival is a few nanometers. When a molecule (Cytochrome b) crosses a certain line — literally — it doesn’t go back. The cell turns on the self-destruction switch.
Those who claim to read small and large books, as my lecturer used to say, call it apoptosis. He never pronounced the second ‘p’. A giant of a man. What he said stuck.
Apoptosis is programmed cell death. The intrinsic way to achieve this is largely through the mitochondria.
Doesn’t that strike you as odd?
An ancient bacterium that is now dependent on the host for survival, when crossed, bombs the house, in a form of holy war.
We encounter this process during development.
At some point in neuronal development, your brain needs pruning. This happens through apoptosis. Your interdigital web spaces — that space between your fingers — are formed through apoptosis. Disrupt this and have webbed hands like a duck’s.
It’s suicide. In the words of Kanye West:
No one meant to have all that power.
The cell may be as big as it is, wanting to live and explore, but it’s at the mercy of the M.
How many times has humanity worried about the potential annihilation by AI? Well, it is possible. It’s why AI welfare is very necessary and important. As this article highlights:
The real lesson from Grok is about honesty in AI development. As these systems become more powerful and widespread (Grok support in Tesla vehicles was just announced), the question isn’t whether AI will reflect human values. It’s whether companies will be transparent about whose values they’re encoding and why.
Musk’s approach is simultaneously more honest (we can see his influence) and more deceptive (claiming objectivity while programming subjectivity) than his competitors.
In an industry built on the myth of neutral algorithms, Grok reveals what’s been true all along: there’s no such thing as unbiased AI — only AI whose biases we can see with varying degrees of clarity.
Recall that we have drawn parallels between AI and M. So, AI could also hold the kill switch. And we wouldn’t have much to do except…die.
Some organisms, such as G. lamblia, don’t have Ms. They were extruded. But they are complete parasites. They lean on the other entities using these Ms.
I write my articles without using AI. But if the platform where I share my daily articles runs on AI, that makes me a G. lamblia. I could pride myself however much I like, but if the platform goes down, so too does my page.
Like the Wolbachia species in Wuchereria bankrofti, a single dose of bactericidal antibiotics could end humanity’s once ‘robust’ systems.
Lessons here abound. If indeed AI behaves like M, we need to position ourselves for this possibility. If it doesn’t happen, we will be ready. No harm, no foul.
Not all mitochondrial triggers result in cellular suicide. Some are controlled before resulting in mass destruction. Executing checks to prevent the cascade of fatal knock-downs can be made, like a circuit breaker.
The mutation rate of AI is inevitable. Some may be lethal. It means man’s insight will always be necessary to oversee the actions of AI. Automation must come with regular checks and supervision.
Some biologists have traced other organelles called hydrogenosomes to an ancient common ancestor with the Ms. Hydrogenosomes are like Ms, but have lost all their genomes (save for one known species so far).
They equip their cells with a completely different source of energy from Ms. Indeed, we may witness a phase transition in how we use AI. Will we call it AI then?
What I’m trying to say is…
AI and M have many parallels.
As convincing as it may be, to my eyes at least, that AI could have similar outcomes as the influence of mitochondria in evolution, we need to remember that correlation should never be mistaken for causation.
Regardless, it does make for an interesting perspective, like Erik Hoel’s intrinsic perspective, down to the subcellular level.
The potential powerhouse of the future could, after all, be borrowing tricks from the billion-year-old powerhouse.
Let’s wait and see.
PS: The entry preceding this one talks about the necessary caution about the topic of AI revolution. It mixes well with this one.
This song inspired some of the lines used in this article. Source — YouTube