Future Computers Will Be Radically Different (Analog Computing)

Item: Future Computers Will Be Radically Different (Analog Computing)
Rating: 5
Author: PostSingularity Institute

by Veritasium

Formal & Physical Sciences Computing & Cybersecurity UComputing and Information Technology UYComputer science

🎙 Veritasium 👥 21.1M 📅 March 1, 2022 ⏱ 21 min 👁 13.1M 📄 science communication 🧭 2026-08-27

Available in: English (current) Français

Keywords

analog computerneural networkmatrix multiplicationperceptronMoore's law

Summary

This Veritasium video explores the potential resurgence of analog computing, driven by the demands of modern artificial intelligence. It begins by demonstrating how an analog computer can solve differential equations in real-time, highlighting its speed and energy efficiency compared to digital systems. The video then traces the history of neural networks, from Frank Rosenblatt’s perceptron in 1958 to the deep learning revolution sparked by AlexNet in 2012. It explains how the exponential growth in neural network size has created a ‘perfect storm’ of challenges for digital computers, including high energy consumption, the Von Neumann bottleneck, and the physical limits of Moore’s Law. The core argument is that analog computing, which performs matrix multiplications directly in hardware using variable resistors, is ideally suited for running neural networks, which are inherently tolerant of imprecision. The video features a tour of Mythic AI, a startup producing analog AI chips that achieve high performance at a fraction of the power of digital GPUs. It acknowledges the challenges of analog computing, such as signal distortion and lack of precision, but concludes that a hybrid approach, combining analog and digital, may be the future of computing.

190 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides substantial value by clearly explaining the fundamental principles of both analog and digital computing, and then connecting them to the historical and current challenges in AI. The argumentation is solid and well-structured: it establishes the historical context, identifies the specific computational bottleneck (matrix multiplication), and presents a concrete solution (analog in-memory computing) with a real-world example. The narrative is compelling, moving from a general concept to a specific application, and it effectively uses demonstrations and expert interviews to support its claims.

Scientific Rigor, Source Quality, Title Accuracy

The video demonstrates high scientific rigor by citing numerous primary sources, including seminal papers like Rosenblatt’s perceptron paper, the AlexNet paper, and the Nature article on deep physical neural networks. The description provides direct links to these references, enhancing transparency and verifiability. The title is an accurate and engaging summary of the video’s central thesis, which is well-supported by the evidence presented. The video also includes a clear sponsorship segment, which is disclosed.

172 words

Title / Content Match

The title accurately reflects the video's core thesis: that future computing will diverge from the digital paradigm, with analog computing playing a significant role.

Quality & Reliability

9/10

The video is a high-quality science communication piece by a reputable channel. It presents a well-structured historical narrative and technical explanation of analog computing, supported by numerous references to academic papers and books. The information is accurate and clearly presented, with a clear distinction between established facts and forward-looking speculation.

Key Moments

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

Cited Sources

AI: The Tumultuous History Of The Search For Artificial Intelligence — Referenced for the history of AI.
Probably Approximately Correct — Referenced for the theory of computation and learning.
The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain — Referenced for the original perceptron paper.
NEW NAVY DEVICE LEARNS BY DOING; Psychologist Shows Embryo of Computer Designed to Read and Grow Wiser — Referenced for contemporary media coverage of the perceptron.
Rival — Referenced for a contemporary article about the perceptron.
ALVINN: An Autonomous Land Vehicle In a Neural Network — Referenced for the ALVINN self-driving car project.
ImageNet website — Referenced for the ImageNet database.
ImageNet Large Scale Visual Recognition Challenge — Referenced for the ImageNet competition.
ImageNet Classification with Deep Convolutional Neural Networks — Referenced for the AlexNet paper.
What I learned from competing against a ConvNet on ImageNet — Referenced for a blog post about competing with a ConvNet.
Mythic @ Hot Chips 2018 — Referenced for a blog post about Mythic's technology.
2.2 Basic operations of flash memory — Referenced for the basic operations of flash memory.
Mythic Multiplies in a Flash — Referenced for an article about Mythic's analog computing approach.
5 Myths About AnalogML — Referenced for a blog post about analog machine learning.
Deep physical neural networks trained with backpropagation — Referenced for a paper on physical neural networks.
The chips are down for Moore’s law — Referenced for the limitations of Moore's Law.
The Analog Thing — Referenced as the analog computer used in the video.
Mythic AI — Referenced as the company featured in the video.

Concurring Sources

The chips are down for Moore’s law — This Nature article, cited in the video, supports the claim that Moore's Law is slowing down.
Deep physical neural networks trained with backpropagation — This Nature paper, cited in the video, supports the idea that physical (analog) systems can be used for neural network computation.

External References

Contribution & Novelties

The video’s main contribution is its clear and accessible synthesis of the historical, technical, and economic arguments for a resurgence of analog computing, specifically in the context of AI. It effectively bridges the gap between the history of neural networks and modern hardware challenges, making a compelling case for why the ‘perfect storm’ of AI growth and digital limits makes analog a viable and necessary alternative. It also provides a concrete, real-world example of this technology in action at Mythic AI, which helps to ground the theoretical discussion.

Pour aller plus loin :

In-memory computing — This concept is central to the video’s discussion of Mythic’s approach, where computation is performed where data is stored.
Von Neumann architecture — The video discusses the ‘Von Neumann bottleneck’ as a key limitation of digital computers, and this article provides the foundational background.
Memristor — This is another type of analog component that is being researched for neuromorphic computing, offering a different approach to the same problem.

164 words

Radar Profile

The radar profile shows a video that is very strong across all dimensions, with particularly high scores in information quantity, quality, and reliability. The slightly lower score in technical level indicates that while the content is sophisticated, it is presented in an accessible way for a general audience.

Reliability 9/10

💬 Très positif. Sur les 30 commentaires analysés, le climat est extrêmement favorable, avec des éloges pour la clarté de l'explication, la pertinence du sujet et la qualité de la production. Plusieurs commentateurs partagent des anecdotes personnelles ou des connaissances techniques qui corroborent les points soulevés dans la vidéo.