AI can transform production.

Previous automation technologies, including the introduction of computers and robotics, and early AI expert systems, relied on predefined conditional logic to guide them step-by-step from input to output. This limited them to routines and structured tasks that could readily be broken down and codified. AI technologies can go further by using machine learning to identify patterns and relationships from huge amounts of data, improve performance over time and adapt to changing circumstances without explicit reprogramming. The economic significance of this is twofold. First, AI can outperform conventional digital systems and in certain areas surpass human performance. Second, unlike previous technological waves that mostly automated routine and low-skill functions, AI can take on tasks that were previously too expensive or difficult to automate, and can be extended to functions that require recognition, classification and prediction that once were thought to be exclusive to highly skilled workers. 

In banking, for example, AI systems are being used to predict loan default rates. 

In healthcare, AI image classifiers are being used to help doctors in interpreting scans and images, leading to faster and more reliable prognoses. AI primarily affects cognitive work, but when combined with other technologies, such as robotics or IoT sensors, it can also control physical production. 

In manufacturing, AI systems, through a network of smart sensors, can exercise real-time control of energy and water usage, for example. 

In agriculture, AI and machine vision can be paired with robots to automate crop harvesting. The potential of AI applications has been further extended by generative AI (GenAI). 

In traditional machine learning, each model performs one specialized task, largely reproducing or representing existing knowledge. GenAI can be much more versatile, performing multiple tasks and adapting to the operating context and generating new content. GenAI can write texts, produce images and videos, write computer code and identify complex patterns in data, for knowledge-based services such as finance, education, law and healthcare. For example, GPT-4, the model that powers the chatbot ChatGPT, has been applied as a customer-support agent, a research assistant for lawyers and a medical research assistant for pharmaceutical discovery and development.

 As performance improves and costs decrease, AI can be integrated into many more production processes. In the best cases, this will augment human labour and improve the quality and speed of work. However, there is also the risk that it could replace workers altogether, increasing unemployment, depressing wages and degrading the work experience. 

If AI is to bring about productive and inclusive economic transformations and reduce inequalities, Governments and companies need to put workers at the centre of AI adoption and development.