I attended several business conferences during the last few months, from the Global Entrepreneurs Conclave in Dubai to the Horasis Annual Meeting in São Paulo. AI is now the new buzzword for business leaders. But the understanding of most business leaders of what AI is seems surprisingly superficial.

At these conferences business leaders talked about how their behaviour was influenced by AI, or how they can educate AI or how agentic AI will do their tasks at the office. This is a debate about Chat GPT and other models, which are in the middle level of the AI spectrum. Shockingly, business leaders hardly mention the lower-level narrow AI models that can provide real utility in farms and hospitals or the higher-level scientific models that will transform the very concept of life on our planet and beyond. The AI arms race is about scientific AI. It is not about conversational AI like Chat GPT, Gemini, or Grok. Ask Peter Thiel, Demis Hassabis, Elon Musk and their Chinese and Korean counterparts.

Narrow AI and Scientific AI

In Baramati, in Maharashtra state, and many other rural districts of India, farmers are now receiving early warnings about weather patterns, locust attacks, and soil conditions via mobile apps. These apps rely on AI models hosted on Microsoft or Amazon cloud platforms, which have been locally adapted to farm-level data. Prataprao Pawar, who heads the project in Baramati, told the conference in Dubai that productivity has increased 40%. The dependence on Microsoft and other foreign cloud services can decline if Federated Learning is used for such agricultural projects. The AI used in these projects on farm is ‘narrow AI’. It was developed before Chat GPT was invented.

In London, teams at DeepMind, use their discovery of protein structure as the basis for future work. They are busy building an Alpha series of models for discovering mathematical formulas, dissecting human genomes, and analysing the structure of the earth. The Alpha Evolve agent released in May 2025 can enable the AI code to improvise itself.

In Seoul, the capital city of South Korea, Minhyeong Lee, a 23-year-old scientist, has assembled a team of high school and university students. They have developed Spacer, an AI foundation model designed to autonomously generate original research ideas in biology and chemistry. They used 180,000 research papers as their data set and spent around $3.5 million in seed funding, as compared to the Stargate project with investments of $500 billion launched by President Trump.

DeepMind models and Spacer are scientifically more sophisticated than anything produced by Open AI, XAI, DeepSeek, Baidu or Alibaba so far.

In Beijing, scientists at the Chinese Academy of Sciences are developing Spiking Brain model. If they succeed, it will be possible to develop AI mimicking neural networks and using a fraction of the data used in Large Language Models today.

Some of the future discoveries may lead to drastic reduction in the use of data and energy. They may help find remedies to cancer and other incurable diseases, invent new source of energy from oceans, recommend new chemicals to extract rare earths easily, and synthesis biological materials that can clean carbon from atmosphere. There are also efforts underway in various laboratories to develop neuro-symbolic AI, neuro-morphic computing and quantum computing. Nvidia is investing heavily in compute computing. Chinese and Korean companies are also investing in quantum computing, but not much is known from public sources. If they succeed, the landscape of AI will change. A lot of the investment in data centre and energy plants will become redundant.

We cannot interact with the scientific models or the future Quantum models, as we do with Chat GPT or its cousins. They will be more advanced than the Agentic AI that everyone talks about. Whoever controls these models will control the future of humanity.

The single domain narrow AI models used in agriculture and diagnostics are at one end of the spectrum. The scientific AI models like the Alpha series, Spacer and Spiking Brain are at the other end of the spectrum. The AI that we use like Chat GPT and its cousins are somewhere in the middle. To equate chatbots with AI is like saying that the screen of the monitor is the computer, ignoring the CPU.

Last Days of Humanity

The scientific AI models can revolutionise life as we know it in a decade or two. But they can also terminate life. If they are not properly regulated, AI safety experts worry that they may produce biological agents beyond human control or deceitful algorithms capable of bypassing cyber defences in the decision support systems of nuclear weapons.

It is also possible that some of these models will eventually transition from self-evolution of algorithms to self-replication. The morning the world wakes up to find the first self-replicating AI algorithm will be the first morning of the last phase of the supremacy of human species on the planet. On that day, the race for human domination of science will be over and the race for scientific domination of humanity will begin. Recently, Sam Altman, said that something drastic might happen in the next few years. He did not reveal what he had in mind. He has access to the chain of thought in Open AI labs. If he suspects anything like self-replication or the capability for massive deception and manipulation, hidden in his comment, he is talking about something potentially as dangerous as a global nuclear war.

Whether future AI brings boon to humanity or leads to universal chaos, it will be a very different paradigm than the world of chatbots, data centres, small nuclear plants to power the infrastructure, and AI devices to refine emails or write poetry and legal cases. If the present AI can manipulate your mind, the future AI will manipulate your life or perhaps end the concept of life on the earth.