AI in 2030 - Extrapolating current trends

This document presents a comprehensive analysis of the future development of artificial intelligence through 2030, based on the premise that current computational scaling trends will continue. The report examines two fundamental dimensions: the resources needed to develop advanced AI and the capabilities these systems will achieve.

The document analyzes five key elements of AI development. First, computational power, which has historically grown approximately 4 times per year and is expected to continue doing so, reaching 1,000 times more compute than current models by 2030. Second, economic investment, which would reach hundreds of billions of dollars to train the largest models, a justifiable figure if AI generates equivalent economic value. Third, training data, discussing how synthetic and multimodal data could solve the possible exhaustion of public text. Fourth, specialized hardware, whose production continues to expand. And fifth, energy, where the largest data centers could require gigawatts of electricity, comparable to the consumption of entire cities.

The report predicts future capabilities by extrapolating current progress on existing benchmarks. It focuses specifically on four areas of scientific research. In software engineering, AI could automate many daily tasks, from solving technical problems to implementing complex software from natural language descriptions. In mathematics, AI systems could help develop arguments, identify relevant knowledge, and formalize demonstrations, acting as assistants for mathematicians. In molecular biology, AI will advance both in specialized tools (such as protein structure prediction) and in general assistants for literature review and experiment design. In weather prediction, AI methods already surpass traditional ones and will continue improving with more data.

This analysis is aimed at researchers, technology developers, policymakers, and anyone interested in understanding the near future of artificial intelligence. The document acknowledges important uncertainties: benchmarks may not perfectly represent real capabilities, practical deployment faces reliability and integration challenges, and external factors (regulation, investment, unforeseen algorithmic advances) could alter these predictions.

The main message is that, if current trends continue, by 2030 AI will be comparable in importance to the Internet, significantly transforming scientific work and many other areas of the economy. The report emphasizes that these predictions should be considered the baseline scenario, and that society must prepare now for this highly probable future.

Key points

• By 2030, AI will be as important as the Internet, transforming scientific work and the economy.
• Models in 2030 will use 1,000 times more computational power than current ones.
• Computational scaling has grown approximately 4 times per year since 2010 and is expected to continue.
• Investments will reach hundreds of billions of dollars to train the largest models.
• AI will automate complex tasks in software engineering, mathematics, and molecular biology.
• The largest data centers will consume gigawatts of electricity, comparable to entire cities.
• Synthetic and multimodal data will prevent the depletion of training data.
• AI will act as an advanced assistant in scientific research, accelerating discoveries.
• The main challenges are reliability, practical integration, and availability of specialized data.
• Predictions are based on extrapolations of current benchmarks showing consistent progress.