On November 18, Germany, France and the European Commission announced a new European initiative for cutting-edge artificial intelligence (AI). As researchers, engineers and entrepreneurs who have pushed the frontier in AI for decades, we strongly believe that Europe and non-European partners have the resources to sustain cutting-edge AI research and development, provided they pull resources together and put talent, science and engineering first. This is important, as future competitiveness and progress may depend on new breakthroughs.
Any innovative AI initiative must face how the field of AI has evolved over the past 10 years and its lesson: whatever the method, scaling computing power and training data will greatly improve the results. Here lies a growing gap between private and nonprofit research, and the opportunity for an initiative that could bring together funding, talent, computing power and data to sustain research in new architectures and methods at the right scale.
Based on our experience to date, a few features are key in cutting-edge AI research. This initiative can only succeed if:
- It focuses on underexplored aspects of the research frontier, like new architectures, methods, reliability or new domains, rather than playing catch-up with the largest frontier models.
- It concentrates resources on a few agile teams, gathered in a very limited number of locations, to ensure the initiative brings added value.
- It attracts world-class talent, which means a unified leadership, low bureaucracy, fast iteration, attractive salaries, and a commitment to open science and scientific freedom.
- It recruits scientific and engineering leadership with a high bar and a simple process, led by eminent and active figures in the field.
- It focuses on pre-commercial development, with a swift mechanism for commercial spin-offs when private capital can take over.
- It pools substantial computational, engineering and data management resources, either directly or through public and private partnerships.
You have 58.33% of this article left to read. The rest is for subscribers only.
