Historical Evolution of Artificial Intelligence

Key points

• AI historical cycles share a similar pattern. AI cycles start with a scientific breakthrough, a research paradigm shift, followed by bold predictions, vast media attention, massive investments, disappointments, unfulfilled promises, and AI winters.
• Why this cycle is (not) different?
• Deep Learning (DL) is the last paradigm shift in AI and transition from model-driven to data-driven methods.
• AI still lacks a basic understanding of the world and causality. 
• Wide AI adoption caused the need for AI ethics and regulation. 

What are the key policy-relevant findings?

• Government investments were dominant in the first two periods, while industry is nowadays investing greatly in AI developments.
• Decades of AI advancements are still not a substitute for a billion years of human evolution.
• The wide adoption of AI bears risks that need to be addressed by government regulation in the direction of AI fairness, accountability, transparency, explainability and related issues.
• Complementary to AI ethical issues, data sovereignty and technological sovereignty are becoming important.
• Considering the current trends, it is safe to foresee in the short-term future that AI will continue to grow and spread across domains.
• Given the AI uncertain future, it is particularly important to have in place an unbiased initiative such as AI Watch to monitor the evolution and assess its impacts over the coming years which one way or the other will see very significant changes in our digitally transformed society.

What has been the most interesting observations with your research so far?

• The similarities and differences between the AI periods
• Appearance of AI ethics
• Transition from model-driven to data-driven AI methods
• Continuous AI progress (have a look at the AI history timeline below)

Details on JRC Publications Repository