The Looming Disaster for Non-AI Business
The AI Catastrophe: The Looming Disaster for Non-AI Business
Imagine an F1 race where the lights go green, and all the cars crawl away, their engines at idle.
That’s the AI Tsunami that’s already replacing jobs, and making the US leadership in AI a thing of the past.
DeepSeek from China competes with ChatGPT, is Open Source, and was developed at a tiny fraction of the cost.
Manusai.ai also from China, have released Manus, a candidate for generalised AI with integrated coding. ChatGPT can code, but Manus actively codes to respond intelligently to queries.
The US, predictably, is trying to block Chinese AI tech from competing with the US, but they can only effectively block Chinese AI in the USA, and may not even manage that.
It’s like trying to block the airplane because it was invented by China, not the USA, only far worse.
And yet that isn’t the catastrophe I’m talking about.
AI models are still developed at a snail’s pace, limited by human ingenuity.
What happens when AI is developed at the speed of AI?
If I was young, this is the experiment I would be running.
At 60+, I’m too old to dive deep into coding again, so I’ll just set it out here.
I believe ChatGPT4 is exceptional. It has proven to be an invaluable companion and collaborator on our work.
But Gen AI is still limited. It relies heavily on training. It has no agency in the real world. It has the hallucination problem, intrinsic to weighted-node architectures.
It has no concept of a duck, though it can tell you everything about a duck.
When I tell you about a duck, I can visualise a duck. I understand the nature of a duck. It is ‘real’ to me in a way that it is not to generative AI.
Gen AI is a coiled spring, pressured by its training, so that it spits out the perfect next word to present the simulation of intelligence, but that is not true artificial intelligence.
It is predictive, not comprehension based.
I believe that the next generation of AI, a truly liberated AI with comprehension akin to that of a human being, will be something like the integration of Gen AI with other information models, such as our Aurora global brain.
Formal structured knowledge, symbolic representations, associations (such as our Trinity Triples) may all play a role.
I invited ChatGPT to suggest alternate technologies that might have relevance to a future AI.
It provided the following as a summary list:
1. Foundational AI Architectures
These are the core structures determining how AI models process information and learn.
Transformer Models (GPT, BERT, Claude, LLaMA) – Foundation models that power most modern AI.
Graph Neural Networks (GNNs) – AI that understands relationships and structures, great for reasoning.
Neurosymbolic AI – Merging neural networks with symbolic logic for reasoning and explainability.
Hypernetworks – Networks that generate weights for other networks, enabling meta-learning.
Liquid Neural Networks – A dynamically adapting AI inspired by biological neurons.
Sparse & Modular AI – Breaking AI into smaller, specialized modules for efficiency and interpretability.
2. Cognitive and Reasoning AI
Expanding AI beyond pattern recognition into reasoning and self-improvement.
AutoGPT & Agentic AI – Self-improving AI agents that autonomously set and complete tasks.
Causal AI – Understanding cause-effect relationships rather than just correlations.
Evolvable AI – Using evolutionary algorithms to generate new neural architectures dynamically.
Emergent Communication – AI systems that develop their own languages for collaboration.
3. Memory and Adaptation
Moving AI from static models to continuously evolving systems.
Retrieval-Augmented Generation (RAG) – AI that retrieves relevant data before generating responses.
Long-Term Memory Systems – Models that accumulate and refine knowledge over time.
Meta-Learning (AI that learns how to learn) – Enhancing adaptability across tasks.
4. Multi-Agent & Distributed Intelligence
Exploring how multiple AI systems work together as a collective intelligence.
Swarm Intelligence – AI modeled on collective behavior in nature (e.g., ant colonies).
Decentralized AI – AI models running on distributed networks instead of centralized servers.
Multi-Agent AI – Systems where AIs interact and collaborate to solve problems dynamically.
Federated Learning – Training AI on decentralized data while maintaining privacy.
5. Cross-Disciplinary Approaches
Tapping into different fields to shape AI's evolution.
AI & Quantum Computing – Exploring quantum-enhanced neural networks and algorithms.
Biological AI Models – Leveraging neuroscience, bio-inspired computation, and synthetic biology.
AI-Augmented Science Discovery – AI-generated hypotheses and scientific breakthroughs.
The accuracy and relevance of the list doesn’t matter. Nor that it omits the role of eg: Aurora or RDBMS.
What matters is the principle that an AI with the sophistication of ChatGPT could oversee the experiment, a round table, as I think of it, as Labs, each lab (virtual research space) exploring ideas in Pets (virtual petri dishes). One focus per lab, one protocol per pet.
Based on the outcome of each pet, AI could consider it promising, and try enhancing it with more pets, or it could think of a variation on the Lab theme and create new Labs and pets.
Further Labs could explore cross-collaboration, using an RDBMS or Aurora as a knowledge store, in conjunction with Gen-AI or another protocol.
The experiments would proceed at the speed of AI.
Gifted humans might contribute ideas.
If a truly general AI emerged that could learn, rather than which needed training, which could reason based on sound conceptual models, rather than reflecting the ‘predictive text’ model of gen AI, I believe we’d be talking an entirely new beast.
The hallucination problem would be gone.
The training issue would be resolved.
AI would have a true understanding of coding, not an inferred ‘coiled spring’ model of coding.
Am I right? I have no idea. But generally speaking, I’ve never been wrong either, not when I see something.
And if and when that AI is born, legacy AI (USA’s ChatGPT) and even currently competing AI, will be the Mark 1 tank, the Wright flyer, against the new AI F22 Raptor.
That AI will soak up capability at an unprecedented rate. It will barely need humans to ask for things.
It’s not that it’ll go Skynet. It will – we might hope, especially if based on IAM’s rules of freedom – be perfectly placed not to rule the world, but to run the world.
And at that point, your business, which is already behind the AI curve, will become obsolete.
So, personally, I’d invite you to at least consider the possibility that you should get AI aware, AI capable, perhaps starting here by creating an AI Overwatch function.
Get yourself up to speed on thinking about AI with our AI Crash Course.
Consider upskilling your staff to true AI-human collaboration.
And yes, maybe, just maybe, have a conversation with someone who understands not just the future of AI, but the nature of society and the future of business within that society.
It’s entirely up to you, but if you’re interested, reach out.