Uploaded December 2021 | Updated September 2026, 11 hours ago
Recorded in the Carl Friedrich von Weizsäcker Colloquium on the 15th of December 2021
Philipp Stecher (Tübingen): Concepts' lifecycle in artificial intelligent systems
Understanding how humans acquire and process concepts has been an ongoing pursuit for millennia. On the other hand, artificial intelligence (AI) has advanced rapidly in recent decades, helping to algorithmize and thereby better understand humans’ concept-processing capabilities. Especially in the recent past, AI scholars achieved remarkable progress: Today’s deep learning algorithms are expanding the capabilities of AI, enabling it to incorporate and process increasingly complex representations of the world (aka "concepts"). However, although AI’s capabilities are often described as “humanlike”, current state-of-the-art AI algorithms can barely meet these expectations. In contrast, recent research postulates significant differences in how current AI systems process concepts of the world as opposed to humans. While people can build rich, integrated concepts that can be applied across domains based on sparse data, today’s AI often requires large amounts of data to create rather superficial concepts that are only applicable to the domain in which the AI is operating. The illustrated gaps in concept-processing capabilities, as well as the recent advancements in AI, reflect the starting point for this research, which aims to shed light on the question of how modern AI systems process concepts in contrast to humans. To this end, a taxonomy will be derived that clusters the concept processing capabilities of modern AI systems. Hereafter, these capabilities will be systematically compared to the concept-processing capabilities of humans. The paper may close with recommendations for further research and concluding remarks on the status of concept-processing capabilities of current AI.
Recorded in the Carl Friedrich von Weizsäcker Colloquium on the 15th of December 2021
Philipp Stecher (Tübingen): Concepts' lifecycle in artificial intelligent systems
Understanding how humans acquire and process concepts has been an ongoing pursuit for millennia. On the other hand, artificial intelligence (AI) has advanced rapidly in recent decades, helping to algorithmize and thereby better understand humans’ concept-processing capabilities. Especially in the recent past, AI scholars achieved remarkable progress: Today’s deep learning algorithms are expanding the capabilities of AI, enabling it to incorporate and process increasingly complex representations of the world (aka "concepts"). However, although AI’s capabilities are often described as “humanlike”, current state-of-the-art AI algorithms can barely meet these expectations. In contrast, recent research postulates significant differences in how current AI systems process concepts of the world as opposed to humans. While people can build rich, integrated concepts that can be applied across domains based on sparse data, today’s AI often requires large amounts of data to create rather superficial concepts that are only applicable to the domain in which the AI is operating. The illustrated gaps in concept-processing capabilities, as well as the recent advancements in AI, reflect the starting point for this research, which aims to shed light on the question of how modern AI systems process concepts in contrast to humans. To this end, a taxonomy will be derived that clusters the concept processing capabilities of modern AI systems. Hereafter, these capabilities will be systematically compared to the concept-processing capabilities of humans. The paper may close with recommendations for further research and concluding remarks on the status of concept-processing capabilities of current AI.
