Jürgen Schmidhuber on AI, Neural Networks, and LSTM (Oral History Pt. 1)

Item: Jürgen Schmidhuber on AI, Neural Networks, and LSTM (Oral History Pt. 1)
Rating: 4
Author: PostSingularity Institute

by Jürgen Schmidhuber

Formal & Physical Sciences Computing & Cybersecurity UBDigital and Information technology: general topics UBBHistory of Computing, digital and information technologies

🎙 Jürgen Schmidhuber 👥 177K 📅 January 22, 2026 ⏱ 123 min 👁 3K 📄 expert opinion 🧭 2026-08-06

Available in: English (current) Français

Keywords

SchmidhuberLSTMDeep LearningMeta-learningAI History

Summary

In this oral history interview, Jürgen Schmidhuber discusses his early life, education, and the development of key ideas in AI. He was born in Munich in 1963, showed early interest in science and arts, and was inspired by his brother, a theoretical physicist. He began programming on a Sinclair ZX81 and studied computer science at the Technical University of Munich. His diploma thesis in 1987 focused on meta-learning, and he later developed concepts like very deep recurrent neural networks, pre-training, and LSTM with his student Sepp Hochreiter. He describes 1990-1991 as his ‘miraculous year’ with many breakthroughs. He also discusses his philosophical views on computation and the universe, and his long-term goal of building an AI smarter than himself. The interview covers his career path, including his time at IDSIA, and his current position at KAUST.

136 words

Critical Evaluation

The interview provides valuable firsthand insights into the development of key AI concepts, particularly LSTM and deep learning. Schmidhuber’s account is detailed and reflects his deep expertise. However, it is retrospective and subjective, potentially overemphasizing his own contributions. The discussion of his philosophical ideas, while interesting, is speculative and not central to the technical content. The sources cited are limited to the museum’s catalog, but the interview itself is a primary source. The title accurately reflects the content, and the technical level is accessible to a general audience with some background in AI. Overall, the interview is a reliable and informative resource for understanding the history of AI, though viewers should be aware of potential biases.

116 words

Title / Content Match

Title accurately reflects the content: an oral history interview focusing on Schmidhuber's early life and contributions to AI.

Quality & Reliability

8/10

Interview with a leading AI researcher, providing firsthand account of historical developments. High credibility due to expert status, but subjective and retrospective.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Computer History Museum Catalog Entry — Official catalog entry for the oral history interview, providing context and archival information.

Concurring Sources

Wikipedia: Jürgen Schmidhuber — Biographical information and contributions, consistent with the interview.

Contribution & Novelties

This interview provides a unique personal perspective on the history of deep learning, particularly the development of LSTM and other foundational concepts. Schmidhuber’s account offers insights into the motivations and thought processes behind these innovations, which are not typically found in textbooks.

Pour aller plus loin :

Long Short-Term Memory (LSTM) — Overview of LSTM, a key concept discussed.
Deep Learning — General background on deep learning, relevant to the interview’s themes.
Meta-learning (computer science) — Concept of learning to learn, central to Schmidhuber’s early work.

86 words

Radar Profile

The radar profile shows high scores in quantity and quality of information, with moderate technical depth. The reliability is strong due to the expert nature of the interviewee, but the subjective perspective slightly lowers the overall score.

Reliability 8/10