Today, at the level of states and companies, there is intense activity around developing diverse strategies for "Artificial Intelligence". These strategies are often framed around a simple, understandable narrative: “be the first.”
However, what concerns me is the near-complete silence on the issue of Quality.
Quality in AI is not just about the characteristics of a model’s primary output. Equally, if not more importantly, it concerns the quality of the application of these technologies meaning the real-world results and impacts that emerge when AI is deployed.
The issue of AI quality is far more complex; it can be divided into two key areas, each requiring special attention:
- The model itself – the characteristics, robustness, and reliability of the AI system.
- The application of the model – how the technology is used in real life, the quality of its outputs in practical scenarios, and the associated safety implications.
And, of course, there’s a critical dimension of safety quality, which is often ignored entirely.
Currently, these questions are largely sidelined. When they are addressed, it’s done very cautiously, so as not to dampen the hype and excitement surrounding the technology.
Sometimes it even seems that the major tech companies, especially chip manufacturers, are the primary beneficiaries of the AI race and they have little incentive to care about the downstream consequences.
It took us decades to embed quality practices in IT and even today, QA still struggles.
I shudder to think of the risks we might be "preinstall" into software and systems before we finally start taking about standards, quality, and safety in AI seriously.
If we want AI to truly serve humanity and not just the race to be first. It’s time for quality to move from the sidelines to the center of our conversations.
And of course, when we start talking about quality, we’ll begin to ask the fundamental questions:
What does it actually mean? How does it work? Only then will we start to cut through the hype and see AI for what it truly is.