The Vertical Curve of AI

The Vertical Curve of AI

Artificial intelligence is no longer advancing in a slow, straight line. It is climbing a steep curve. Tools that once took years to improve can now change within days. This shift is not only technical. It may reshape science, medicine, work, and society itself. 

 

The central idea is simple: AI progress is happening faster than human intuition can easily grasp.

A New Kind of Acceleration

For much of history, technology improved gradually. A better engine, a faster computer, or a new medicine often took decades. AI is different. Breakthroughs in coding, image generation, animation, and music can now appear in an afternoon. A model may produce a useful program in the morning and a polished video by evening.

This speed creates a feedback loop. Better AI helps researchers build better AI. It writes code, designs experiments, analyses data, and suggests improvements. Each gain makes the next gain easier. That is why progress can feel sudden. The curve turns upward sharply rather than rising evenly.

Mathematics Moves Fast

AI is also beginning to handle problems once thought to need years of human effort. Complex mathematical questions that have resisted solution for nearly a century are now being attacked at remarkable speed. In some cases, systems can generate hundreds of possible approaches in days.

This does not mean every answer is correct. Human mathematicians still must test, verify, and interpret the results. But the change is important. AI is becoming a partner in discovery. It can search through possibilities far faster than people can. It can reveal patterns that humans may overlook. Mathematics, long seen as a slow and deeply human discipline, is becoming an early testing ground for machine-assisted thinking.

The Logic of Scaling

A major reason for this acceleration is the idea of scaling laws. In simple terms, larger AI models, more training data, and more computing power often produce predictable improvements in performance. Research has found that model loss tends to fall as a power law when model size, data, and compute increase.

This does not mean intelligence grows without limit. There are trade-offs, costs, and diminishing returns in some settings. Still, scaling has been one of the most reliable drivers of modern AI progress. Bigger systems can learn more subtle patterns. More data gives them broader knowledge. More compute lets them practise more deeply.

The effect is amplified by better training methods. Researchers have learned how to use compute more efficiently. They fine-tune models, improve reasoning, and apply reinforcement learning. They also use AI itself to generate and refine training material. In this sense, AI can help build its next version.

Biology as Code

One of the most striking ideas is that biology may become a field treated like code. The human genome is often described as containing roughly 750 megabytes of information. DNA stores instructions for building and maintaining living systems. If we can read, write, and edit those instructions with precision, biology could become far more programmable.

This comparison should not be taken too literally. A living organism is not just a computer program. It involves chemistry, environment, development, and immense complexity. But the analogy is powerful. AI can help decode genetic data, predict protein structures, design molecules, and model biological processes.

If biology becomes easier to engineer, medicine could change dramatically. Diseases might be detected earlier. Drugs could be designed faster. Treatments could be tailored to individual patients. The same tools that accelerate software may eventually accelerate life sciences.

A Promise (and a Risk?)

The potential benefits are enormous. AI could help cure diseases, extend healthy life, remove repetitive work, and create greater abundance. It could support scientists in fields where progress has been limited by time, cost, or human attention.

Yet the risks are real. Powerful systems can be misused. They can spread false information, automate harmful actions, or concentrate power in few hands. Some researchers also discuss existential risk, sometimes called the “p of doom.” Estimates vary widely, and one figure mentioned in the video is about five per cent. That number should not be treated as a settled scientific fact. It is a reminder that uncertainty itself demands care.

The right response is not panic. It is preparation. Societies need strong safety research, transparent testing, clear rules, and international cooperation. Progress should be guided by human values, not left to speed alone.

Why We Misjudge It

Human brains evolved to understand linear change. We expect steady improvement. We are good at imagining one step, but poor at imagining repeated doubling. Exponential growth starts quietly. Then it becomes overwhelming.

This is why AI progress can feel unbelievable. A system may seem limited today, then become capable much sooner than expected. The same difficulty appears in debates about space and alien life. Vast distances and immense timescales are hard for our minds to picture. Exponential technology presents a similar challenge.

We should therefore avoid two mistakes. We should not assume today’s limits are permanent. We should also not assume every prediction will come true. The wise position is humility: prepare for rapid change while demanding evidence at every stage.

My Inference

AI is moving from linear progress to vertical acceleration. Scaling laws, better methods, and AI-assisted development are making each generation of systems more capable. Biology may be the next major frontier, as DNA becomes increasingly readable and editable.

 

The rewards could include longer, healthier, and more abundant lives. The dangers demand equal attention. The future will depend not only on how fast AI grows, but on how wisely humanity chooses to direct it.

 

- Jishnu Chatterjee,

Durga Puja, 2026.