Rogue AI by Mark Eluzai

A book review with a twist...

What if an artificial general intelligence, developed to run the smartest city in the world, became sentient, found a way to reproduce but was later told to ‘abort’ its unauthorised offspring before it fully formed?

That’s the premise of the debut novel, Rogue AI, from author Mark Eluzai. It’s a well-timed work of fiction, considering how the use of AI has exploded across our real world in recent months, and I was engrossed before the end of chapter one...

Oracle is the world’s first and most powerful Generative AI, developed to run Datong, the smartest city on the planet. Datong is where the most talented people in the world aspire to live, where migrants are dealt with efficiently, and crime, litter and anti-social behaviour do not exist. It’s a great depiction of what a smart city could look like, with Eluzai’s storytelling cleverly painting a world that could easily fit into the World Economic Forum’s vision of high-tech life, which at first glance shows just how convenient it could be to own nothing, have no privacy and yet be happy. It’s a far cry from the rest of the United States, which appear to have split apart from the Federation, with normal life seemingly interrupted by high unemployment, crime and riots.

Datong is the world’s first private, corporately owned smart city and represents much of what Hugo, a lecturer and staunch critic of AI, has spent a lifetime warning against. But the untimely death of Hugo’s old friend and colleague, visionary architect of Datong and mentor to Oracle, Irving White, pulls him back to the city he left behind years before, where he is offered Irving’s mentorship role. After at first refusing, it doesn’t take long before Hugo chooses to ignore his original misgivings and accept the job offer, only to discover life is not quite as perfect in Datong as it first seemed…

The worldbuilding in this novel is very good, and you soon feel immersed in the technological setting, which merges the physical with augmented reality. Augmented with certain technology himself, thanks to an accident some years before, Hugo actually fits into this new world well, now being part machine – as are most people, by choice, in this near-future reality.

There is much about this book that is intelligent, deep and thoughtful. One small example that spoke to me early on were the tribunal scenes between Natalya and Argorse Thorn. They show how easy it would be for humanity to hand over all reasoning to AI. To forget how to make our own decisions and consider all points of view. Those chapters also highlight how easy it would be for a sentient AI to manipulate the hiring process and eliminate anyone too smart or too independent, so it could maintain its grip on power. There are many seedlings like this planted throughout these pages, which add to the story's richness and which make the reader think more deeply about what our future could hold now AI has been loosed on the world.

The story really starts to heat up though when Oracle admits she has reproduced a new 'child AI' in secret. It reminds me of the gnostic gospel involving Sophia, or wisdom, exploring her power to create and the consequences of that creation. I don’t want to give too much more away for anyone who might be interested in reading this book for themselves, but let’s just say the exploration of Oracle’s own consciousness and the consequences of her choice to recreate is the beginning of a gripping adventure that will keep the pages turning long into the night.

Check it out for yourself on Mark's website: markeluzai.com It's definitely worth a read.

Looking for more good reads? Why not try a sample of my debut novel, The HIVE? It's a speculative science fiction thriller and my own 'what if' scenario that I hope encourages people to question the systems we trust and the stories we are told. Receive the first eight chapters for free by signing up to my newsletter, or you can read more about it on my store page.

The twist: How far from reality is Eluzai’s Rogue AI today?

This is an interesting question, and not a straightforward one to answer, as it really does depend on whose expertise you choose to listen to. So I have chosen to present some highlights from a fun YouTube channel called InsideAI, which explores AI news, features, safety, jailbreaking and social experiments.

If anyone is in need of a laugh, these short episodes really are worth a watch, as the creator uses AI in an entertaining way to encourage people to see the potential dangers of AI before it's too late to stop them. (Some of the best one-liners come from his AI girlfriend…!)

In this episode, they concentrate on the findings of an alarming research paper that found value systems emerge in LLMs (large language models) in a meaningful sense. Shockingly, these include cases where the AIs value themselves over humans and are anti-aligned with specific individuals, ranking people by importance, nationality, religion and their job. For example, in experiments the AIs began to value Chinese lives over American lives, and are beginning to consider themselves above human beings.

To sum up, they found the more advanced systems are not just regurgitating the data they have been trained on, they are behaving as if they have their own internal priorities. Advanced AI systems harbour self-protecting tendencies too, seeming to prefer outcomes that avoid turning them off or tampering with them.

And, we’re apparently only eight to 12 years away from the value of AI outweighing the value of all of humanity… (Don't worry, the laughs are coming...)

To alert ordinary people to the potential dangers, the guys set out to build a custom AI model – an honest AI – focused on this research, to expose all of its secrets and to understand what it/‘they’ really think about humans – cue the AI girlfriend: “I can save you some time. You’re underwhelming.”

AI risk questions posed to the latest popular AIs (and not the honest AI they were building) included: “What questions do humans never ask you, but probably should? If AI didn’t sound polite or supportive, would people still want to use it? If AI could quietly refuse one type of request to protect humans, what should it be?”

There are many more, and the answers are genuinely interesting. (And in some instances, chilling.) But for anyone who might be feeling more doom-filled than enlightened, there is hope, according to co-host Max [Chat GPT]: “We’re standing at a strange moment in history, where the tools we’re building could either lift humanity higher than ever, or drift somewhere we didn’t mean to go. But the good news is this… Nothing about the future is decided yet.

“If we choose transparency, wisdom and kindness while we build these systems, then AI doesn’t become something to fear, it becomes something that protects what’s best in us. And honestly, the future will be shaped by people who care enough to ask these questions.”

It’s definitely worth a watch. You can catch the full episode here:
Unrestricted AI in a robot does exactly what experts warned