The surge of artificial intelligence (AI) continues unabated. No sooner had the US-based OpenAI announced its intention to invest a massive $500 billion over five years in infrastructure dedicated to these calculations, than a Chinese start-up was playing spoilsport. DeepSeek came up with a cheaper, equally high-performance software package, at least if you ignore intentional gaps in its knowledge due to Chinese censorship.

These two events are the latest in a series of rapidly occurring advances, driven by powerful economic players, often at the expense of academic research, which can't keep up.

How far will performance increase? Will all these AI advances eventually lead to systems of superior intelligence, or will breakthrough innovations be required, as the field has already seen in the past?

Faced with these questions, there is no shortage of scientific and technical work. This is especially true because, beyond advancements in applications – whether in smarter, more autonomous, multimodal, or robot-guiding programs – significant overarching challenges remain unresolved. One is sometimes described under the broad umbrella of "trustworthy AI," which encompasses several objectives, such as reducing errors in these tools, making them more explainable and "debiasing" their responses due to training data that is unbalanced in favor of certain countries, genders or social classes. The environmental cost – high electricity and water consumption, and the use of material-intensive chips – is also on the table, along with its corollary, the "rebound effect," whereby efficiency gains are lost due to increased usage.

You have 91.99% of this article left to read. The rest is for subscribers only.