AI has stopped waiting for instructions — Are we ready?

Artificial intelligence (AI) means different things to different people.

From one perspective, it offers a powerful solution to global challenges. From another, it’s a source of disruption to jobs, privacy, and trust.

In practice, however, the impacts of AI are more complex and consequential.

The second of edition of the AI for Good Impact Report (available here) from the International Telecommunication Union (ITU) and Deloitte compiles evidence on AI’s shift from experimentation to increasingly widespread and diverse real-world deployment.

As AI increasingly shapes public services, markets, and societies around the world, the report explores what this means for people, institutions, and the public interest.

From assistance to agency

Much of the public debate focuses on generative AI – systems that create text, images, or code. But more consequentially, AI is becoming increasingly agentic, meaning it’s capable of planning tasks, making decisions, and coordinating actions with less human oversight.

This shift is already evident in healthcare, logistics, finance, and public administration, with AI agents increasingly embedded into operational workflows, sometimes interacting with other AI systems.

But as AI gains autonomy, human oversight becomes more vital than ever.

And the human part of the process must be designed, not assumed.

Where AI is delivering measurable benefits

AI is already producing tangible outcomes in key sectors:

• Education and skills: Adaptive learning systems are improving educational outcomes by tailoring content to individual learners. In regions facing teacher shortages or limited resources, AI-supported tutoring tools are expanding access to quality education. However, these benefits depend heavily on reliable digital infrastructure and connectivity, highlighting the inseparability of AI capacity from broader development conditions.
• Health systems: AI is supporting earlier disease detection, more accurate diagnostics, and faster drug discovery. In clinical settings, decision-support tools help professionals manage increasing workloads, while virtual health assistants improve access in remote or underserved communities. These applications demonstrate AI’s potential to strengthen health systems rather than replace human expertise.
• Climate, environment, and resilience: AI-enabled monitoring and prediction tools track environmental change, optimize energy systems, and provide early warnings for extreme weather events. In the face of climate pressure, AI systems are reducing disaster risks and helping vulnerable countries adapt. But limiting AI’s own environmental impact should also be a priority.
• Cities, infrastructure, and public services: Urban systems, from traffic optimization to emergency response, now rely heavily on AI. Digital twins and predictive models let governments test different policy choices and infrastructure investments in advance.
• Agriculture and food security: AI-driven precision agriculture supports more efficient use of land, water, and other inputs, while helping farmers respond to climate variability and market uncertainty. When combined with open data and accessible tools, these technologies can help vulnerable countries produce more food.

Growing social and environmental risks

As AI systems get more capable, they also bring bigger risks. 

As the World Economic Forum found recently, some 91 million existing jobs could evolve, undergo transition, or disappear by 2030, while 170 million new roles could emerge – a net gain of 79 million – globally.

At the same time, employers anticipate that 39 per cent of core skills are set to shift by 2030, putting pressure on education systems, employers, and workers to adapt at unprecedented speed.

Environmental impacts are also becoming harder to ignore. In 2024, data centres consumed an estimated 415 terawatt-hours (TWh) of electricity (approximately 1.5 per cent of global electricity consumption), a figure projected to double by 2030. A single large AI data centre can use as much electricity as 100,000 households, with sustainable data centres becoming an industry focus as a result.

To get ahead of such challenges, institutions tasked with overseeing AI need to keep pace with the technology, ITU’s report argues.

Nations race for digital autonomy

With AI now embedded in critical systems, national governments are increasingly investing in domestic computing capacity, datasets, and foundational models.

For example:

• India is developing multilingual foundation models alongside large-scale domestic GPUs (graphics processing units) that reflect its linguistic diversity.
• Switzerland has launched Apertus, a public AI system to support research, innovation, and public-interest use cases.
• The United Arab Emirates is developing its own Falcon large language models and building a national AI campus.
• The United Kingdom has established a Sovereign AI Unit to fund and scale up domestic capabilities.

These initiatives treat AI as a strategic asset comparable to energy systems. Decisions on training, governance and infrastructure have become a matter of public policy rather than market preference alone.

How to ensure AI serves the public good

AI does not produce positive outcomes by default. Its societal impact depends on fundamental choices to ensure fair access and responsible usage.

ITU’s report calls for intensified cooperation between governments, industry, academia, and civil society to avoid fragmented or uneven outcomes.