Twin Pillars of Transhumanism: Deep Learning
*If you’re reading this blog, there’s a good chance you’ve heard of Longevity Escape Velocity. This article describes the overarching strategy the editors support for attaining that goal.*
Unless someone was keeping specific track, they could be forgiven for not knowing what deep learning, or even a neural network was. Even for those who work in an area like machine learning, it would not be far-fetched to have never heard of many of the more exciting projects which have been going on over the past year or so. AI or Artificial Intelligence is a term which has come to describe a broad family of applications, but when Transhumanists and those seriously involved in cutting edge AI research talk about it, they are usually referring to deep learning, a “Technique to perform machine learning inspired by our brain’s own network of neurons”. Deep learning AI fundamentally acts a lot more like applications people imagine in science fiction, closer to something that reminds people of Hal 9000 or the image enhancing technology from Blade Runner. Deep learning is what people think about when they hear about AI, and the technology has become a central aspect of Transhumanist thought.
Deep Learning’s development, perfection and implementation will be key to allowing those focused and devoted individuals with a burning and more permanent desire to live, to achieve Longevity Escape Velocity. We have a community made up of disparate methods, making use of distinct technologies to achieve similar ends. More needs to be shared, from funding to research, but most importantly, the time has come for the community to begin uniting around common goals regardless of avenue of research. Individually, powerful new architectures or medical discoveries can more easily be ostracized and delayed from entering the main stream, costing the lives of those who would have benefited from those advancements the most. Together, the twin pillars of transhumanism, biology and machinery, can join together to create a community which puts developing technology for the sake of helping our civilization at the forefront.
The current deep learning revolution is considered to have begun in 2012, at the time there was a competition called ImageNet, the goal of the competition was to create a machine capable of correctly classifying images in ImageNet’s dataset which includes millions of images. The most advanced AI’s years prior were not deep learners, they learned the same way most modern computers do, essentially just brute force training and recognition after being trained on countless examples. Alexnet, a CNN (Convolutional Neural Network) patterned after human neurons was successful in recognizing images with a top-5 error rate of just over 15%, to casual onlookers not a massive achievement, but those who were paying attention understood that this was the future.
Almost 10 years later, deep learning neural networks are a multi-billion-dollar industry, the growth has been exponential instead of incremental in nature, which means that the sophistication of AI systems has become ever more complex with each passing year. In 2018 the deep learning architecture which captured the most interest were GANs (Generative Adversarial Networks), they were capable of impressive feats, such as creating photos of people so realistic, almost no one could tell the difference, by that year deep learning was already fueling AI applications worth billions. By this time, many compared the intellectual capabilities of GANs to small insects, and each one was generally good at just one specific task or type of task, “narrow AI’s” built for specific purposes.
It was only about a year ago that AIs considered more general in their capabilities began reaching the public, most notably transformers such as the language model GPT-3 have become part of a trend which places an emphasis on the ability to perform a larger variety of tasks and includes qualities such as “meta-learning” which is the ability to learn how to learn. GPT-3 and other transformers are capable of learning at a level surpassing previous architectures, it can hold conversations with people, and learned how to code even though it was not specifically trained or programmed to do so. The coding applications in particular are being spun out into a new program called Codex and are set to revolutionize software engineering within a few years at most. This barely scratches the surface of what lies on the cutting edge or future of AI, the longer-term implications are almost endless, but in the short-term AI encompasses a gold mine for those interested in subjects such as age reversal and super longevity.
Deep Learning and Medical Research
The possibilities for artificial intelligence to improve human longevity are manyfold, there are several potential avenues by which sufficiently advanced AI could result in indefinite lifespans. As will have been covered in another article, geneticists and other medical researchers are close to finding a way to reverse the aging process. With the aid of artificially intelligent systems, already promising research could be rapidly accelerated, resulting in refinements in the field of genetics that few had even dared to imagine before. Although this technology has not been applied directly to anti-aging research, the idea that deep learning will revolutionize the life sciences is not hypothetical. Alphafold 2 is a transformer built by Deepmind for the purpose of protein folding, and was successful mapping the entire human proteome at least in part, an achievement such prominent publications as Nature have said “has the potential to revolutionize the life sciences”.
Although research in anti-aging has made significant progress, at this stage there is a good argument for the idea that AI is the fastest path to LEV (Longevity Escape Velocity). If every major Biotech incorporated high level AI solutions into their R&D, the progress in medicine and specifically areas like aging research would skyrocket. While implementation can be challenging, the potential for deep learning to change the face of medicine and biology is only the tip of the iceburg. Alphafold 2’s accomplishments may have been impressive, our understanding of biology is about to be upended, but compared to the future this was just a demo. In the future, deep learning tools will be able to help scientists analyze nearly every aspect of the human body with pinpoint accuracy. Compared to those emerging capabilities, even what is being practiced in the most advanced areas of medicine is primitive when one considers what has become possible with the relatively crude and unrefined deep learning architectures of the modern day.
The difference between modern and AI-enabled medicine is so stark, the output it comparable to something like the going from the renaissance to the industrial revolution in the space of a few years. The techniques made possible with deep learning assisted research are so vast that it is inevitable that it will one day dominate modern medicine. So far, at least in terms of research there have been two main groups of transhumanists, those who research areas like genetics in search of cures to longevity and the capacity to modify human biology, and those in AI who seek to create more powerful tools. More powerful tools have now arrived. In order to achieve the best results, these twin pillars of transhumanity must unite and begin feeding into each other to be able to achieve both dreams. This technology is only going to become more impressive, the systems which become public knowledge within a year or two will dwarf the capabilities of the most advanced systems today. We will remain with you as this story unfolds, and give updates on this rapidly accelerating arena of technology.