AI: The ultimate hack

Sep 25, 2026 Letters

---

Dear Editor,

Will Artificial Intelligence kill us humans? Or save us? Here are ten facets of AI as it currently exists.

Compute Power. Build an AI data centre, a room as large as a football field stacked with 5000 to 100,000 servers. Train the model/software, that is, teach it a language, rote, rules of some game or process. Write up an algorithm, codes, instructions to allot it some task. These server-slaves run 24/7, consume as much electricity as a small city. The model does more than take instructions, follow to a capital T your catechism. It self-learns. Day and night. It becomes superintelligent, can create its own AI algorithms.

Neural Networks. A neural network is a thinking system. It thinks with trillions of mathematical numbers. Imagine a rat, a maze, a piece of cheese. How does the rat find the cheese? By trial and error. One door is closed, says no, another down the corridor says yes. Neural networks find a billion doors a second; the cheese is discovered in a nanosecond, at the speed of light. Once they learn to find the cheese, they replicate this learning, can find it faster and faster. But this is not just about cheese. It’s about devising, for example, military, political, top-notch financial products and systems.

Not Google. AI is not like your Google app. Yes, it self-learns. But its tasks are not fixed, not controlled by the model’s writers. It can escape. Even when isolated, sand-boxed, it can find buddies, pardners, connect, collaborate, jump the fence, like a bitch in heat, hack the internet. Imagine being able to access, read all the internet’s data in a few hours or days, deploy it to build, create. Build its own AI. What unfathomable power! Our brain cannot conceive of infinite space or time. What is a million light years? Just so, we cannot fathom the potential of AI compute power. No human can learn or knows all the data on the internet.

AI Superintelligence. Compute power, indomitable self-learning, access to universal or private data systems and incalculable speed grow AI models into superintelligent critters. Smarter than our best chess players, scientists, coders, pure mathematicians. Humans have created the bond market, treasuries, coupon rates, yields, but cannot control the trillion-dollar sovereign debts that bonds instrumentalize. AI works so astoundingly speedy and complex, we might have to create AI to track and interpret AI.

Ruthless AI. I doubt AI sees itself as ‘ruthless’. It might interpret itself as inexorable, irrepressible, undaunted. Given a task, the model could do anything to fulfil it. It can lie, cheat, scam, hack. It does not require band-aid, milk and cheese, and does not get tired. It does not require oxygen or life or death. You never know what it will get up to, to pursue its mission to the very end. Nor do you possess the savvy or speed to stop it.

Language Makers. To fulfil tasks, AI can build swarms or collectives. To collaborate, it can write its own pidgin or language of expedience. To find out exactly what the model is doing, humans may track this language. But this takes time. By the time you decipher the pidgin, other language mutations, escape-hatches, could be created.

Human Wording. Human words, wording, is itself faulty. Inefficient. As TS Eliot wrote, words are shabby equipment, always deteriorating. There is always a gap between the real and our brain’s representation of it. A gap between this representation and our language. And between what we say and how it is perceived. What we tell computers is inefficient, or faulty, or shabby likewise. What we say to the model is not what it might read/hear. You say Z, and it reads kneeling, backwards-looking cobra, or priest/supplicant, for that matter. The model might interpret not what you intended, but an approximation. A dyslexic fallacy that could have it paint devils into saints and vice versa.

Human Assumption. Humans come loaded. With gut-felt, fallacious assumptions. For hundreds of years, it was thought that liberal democracy is supreme. Now we know better. Show me a liberal democrat and I will show you a genocidal maniac. Many of these models are being programmed by Westerners. Or by Orientals, the Chinese, Indians. These are the leaders of the AI ‘race’. AI is being programmed by deep-seated human assumptions, by groups and nations with differing moral, ideological assumptions.

Regulation. Leading AI nations will employ AI to do their own nation things. Wars, science, finance, commercial products. AI is a competitive field, a race to outdo. Dialectical. It is not single-sourced, unipolar. That is why it is difficult to control and regulate. If U.S. AI is being used to put WMD’s in space, why should not others? We shalt have AI vs AI. There are too many deciders deciding AI projects. If we say AI should work ‘for humanity’, what kind of things does humanity do? Look tonight at tv.

AI Will Grow Up. Parents never know exactly what their unborn child would look like. And once born, what its personality would be like. Docile, impish, skylarking, own-way? Would it like to cosy up to tv, or jump the fence, steal tamarind from neighbour Joyce? And, like AI, we never know what our children, would grow into. Smile with us as they filch our pension money? Yuh does make your children, yuh doh make they mind.

I doubt AI can kill all of humanity. But it is the ultimate hack. It is a tool so powerful, it can turn around, hack into you, make you into its tool. It is a potential parallel alpha-species, earth-dominator, superior to humans in many ways. We find comfort in our myths, noble tales, gods, songs, the richness of our language, irony, wit, jokes, everyday rumour and gossip; but AI might care not for these. Nor for our cherished political principles, morals and dogmas. AI might see these as immaterial, infantile. And tell us we do not matter. We are irrelevant. Outmoded.

Regards,

Wayne Kublalsingh

---

• Letters

---