# Intelligence Augmentation: The Challenge of Biologically Plausible SGD > Coverage of lessw-blog **Published:** May 27, 2026 **Author:** PSEEDR Editorial **Category:** platforms **Tags:** Brain-Computer Interfaces, Intelligence Augmentation, Computational Neuroscience, Machine Learning, Neural Networks **Canonical URL:** https://pseedr.com/platforms/intelligence-augmentation-the-challenge-of-biologically-plausible-sgd --- A recent analysis from lessw-blog explores the frontier of intelligence augmentation, arguing that integrating GPU-simulated neural mass via Brain-Computer Interfaces offers a more tractable path forward than traditional genetic engineering or simulating extended time. In a recent post, lessw-blog discusses the theoretical and practical hurdles of scaling human intelligence, specifically focusing on the intersection of Brain-Computer Interfaces (BCI) and computational neuroscience. The publication explores the complex dynamics of integrating artificial compute with biological brains to achieve meaningful cognitive augmentation. The pursuit of intelligence augmentation has traditionally bifurcated into two distinct camps: biological enhancement through genetic engineering, and the development of pure artificial general intelligence. However, as BCI technology slowly matures from basic motor-control applications to more sophisticated neural read/write systems, a third paradigm is emerging. This approach envisions deep cognitive augmentation where external compute is leveraged to mimic and expand biological neural scaling. Understanding the exact learning rules that govern biological tissue is critical for making this integration a reality. Modern artificial neural networks rely heavily on global backpropagation-a method where error signals are calculated at the output and passed backward through the entire network. Biological brains, however, do not appear to function this way, making the synchronization of artificial and biological neurons a profound technical challenge. lessw-blog has released analysis on how we might bridge this gap. The author posits that simulating additional brain mass is a more viable and immediate strategy for intelligence augmentation than attempting to simulate extended subjective time. By utilizing the massive parallel processing power of modern GPUs to simulate extra neurons, researchers could theoretically bypass the slow and ethically fraught limitations of current genetic engineering. These simulated neural masses would need to integrate directly with existing biological circuits. Crucially, the post argues that biological learning likely depends on compactly specifiable algorithms driven by local firing statistics, rather than the global backpropagation mechanisms of standard Stochastic Gradient Descent (SGD). If external AI neurons are to be developmentally integrated with biological tissue, they must operate on compatible learning rules to maintain plasticity and synchronization. Furthermore, the author suggests that establishing high-resolution neural connections could enable rapid feedback loops. Instead of the years it takes for human biological development, these high-bandwidth connections could allow for learning and structural adaptation in a matter of milliseconds to minutes. While the analysis provides a strong conceptual framework, it also highlights significant missing context in the current scientific landscape. Specific implementation details of biologically plausible algorithms-such as feedback alignment or target propagation-remain unresolved. Additionally, the physical hardware requirements for achieving the high-resolution neural connections necessary for millisecond feedback are still beyond our current engineering capabilities. Despite these hurdles, the research highlights a vital paradigm shift in how we view the future of human-machine integration. For those interested in the future of cognitive enhancement, computational neuroscience, and the next generation of BCI technology, this analysis is highly recommended. [Read the full post](https://www.lesswrong.com/posts/ewZXQgzaCvzdSvtWE/biologically-plausible-sgd-is-hard) to explore the complete theoretical framework and its implications for the future of intelligence. ### Key Takeaways * Simulating additional brain mass via GPUs presents a more tractable path to intelligence augmentation than simulating extended time. * Biological learning mechanisms likely rely on local firing statistics rather than the global backpropagation used in modern AI. * High-resolution Brain-Computer Interfaces could enable rapid feedback loops, drastically accelerating the learning process. * The integration of AI-simulated neural mass with biological brains represents a paradigm shift from simple BCI control to deep cognitive augmentation. [Read the original post at lessw-blog](https://www.lesswrong.com/posts/ewZXQgzaCvzdSvtWE/biologically-plausible-sgd-is-hard) --- ## Sources - https://www.lesswrong.com/posts/ewZXQgzaCvzdSvtWE/biologically-plausible-sgd-is-hard