Companies have moved from the discovery stage to the adoption of artificial intelligence tools. This technology is now being used in all sectors of activity to the point where it is becoming commonplace and is gradually losing its magic touch. There is a lot of feedback, and some of it has been very difficult. AI technologies have also progressed well, benefiting in particular from advances in the power of specialized processors for deep learning, both on the server side and on the embedded or smartphone market. Language and image processing is working much better than it did 4 years ago. Companies are also learning how to better manage their data and limit the effects of data bias. Other areas have made less progress such as robotics. It was also realized that much of the predictions about the impact of AI on employment were far too pessimistic.
The craft of AI lies at several levels. First of all, at the level of the algorithms which evolve regularly but which are a never-ending Lego game, especially in language processing. And then, above all, at the level of applications. These are less easy to make generic, especially for software editors. They are very dependent on the structure of the training data, which often comes from the user companies. This has a negative impact on the industry's economies of scale. We recently saw this with the acquisition of Canadian Element.AI by ServiceNow in the US. Element.AI had raised a lot of money but failed to create real economies of scale. These seem to be harder to reach in AI than in previous waves of software. This reinforces the importance of the developer tools market and in particular an emerging category, AutoML, the toolkits that help developers and data scientists automatically determine the best machine learning and deep learning software models based on training data.
There is too much focus on uses and data and not enough on tools in geopolitical analyses of AI. The strategic markets are those of development tools that determine the fate of software platforms, processors and cloud platforms. The rest are mainly application solutions covering the needs of organizations and the general public, and the whole range of digital tools from connected objects to telecommunications and all kinds of connected screens. AI is the focus of attention, but it is only one of the building blocks of these digital systems. The role of data is exaggerated. Their accumulation is the consequence of the existence of software platforms adopted by users and of software ecosystems built around these platforms. The example of the millions of applications developed for Android and iOS is eloquent. They explain the dominance of Google and Apple in smartphone platforms before any AI-related consideration.
The importance of this mechanism for creating platform application ecosystems should be better understood by the elites who have a desire to regulate the market, especially at the European level. Current approaches are far too defensive. They erect barriers more than they allow the development of our AI companies, especially when they focus on exploiting local data, such as medical ones, whereas only global services and data can build leadership. The recipes for success in AI are similar to those of digital technology in general: responding to needs that are as generic as possible, creating platform products and developing their ecosystem with complementary products and services created by third parties.
Finally, we need to export quickly, especially to the United States, the largest homogeneous developed country, which sets the pace for de-facto standards at a global scale. It is also necessary to amplify fundamental and applied research efforts to address fundamental problems, structured in a transversal manner around ambitious "moonshots". Positions are to be taken in the upcoming battles, not in those of the past.