Preface: A little bit of math in this. Written in collaboration with Gemini.

I’m getting pretty confused as I dive deeper into this topic. I just wanted to know about Epstein, but Minsky is too interesting of a character to pass on. Looking at his publications and books, he almost seemed too early for his time. I agree with most of the things he’s proposing, however, I am not a mathematician to fully understand his proofs. If this was the guy that was serious about cryonics, and this is who Epstein proposed his baby ranch to, we can know for sure they were both onto something substantial, and they both put their words into actions with their brilliant, potentially morbid mind, in Epstein’s case. While Minsky used his mind to explore the future of human consciousness and the nature of our intelligence and agency, Epstein used his mind to play crossbreeding with girls like they were plants and roses in Animal Crossing. Minsky died preserved hoping to be woken up later, Epstein perhaps isn’t dead and hoping to wake up later as well, same wish, yet how two people went about it were so different. It tells me once again that there ain’t anything wrong with wanting to live forever, yet there’s a difference as to how we achieve that goal. Same with any sort of biological, technical advancements. The end does not justify the means, at least that’s what I think. If the process must inflict pain on the weaker, then it’s not about “transforming” human lives, it’s about abusing the less powerful with new technology. If the process is unjust, no matter what result we yield, it will not okay the process. 

I’ve jumped around a lot when researching Minsky, I first wanted to find out what the infamous study of cryonics was about, only to find it is a very well researched, science-backed, up-and-coming industry that some people are going all in on. They are actively solving all the problems they can foresee, such as payment plans, putting your money in a trust so that even if you die the U.S. economy will pay for your “frozen body’s rent” in a cell, storage units being built… there’s a whole ecosystem of industry being put together. Yet that was hardly Minsky’s main focus or most famous work. He is most famous for his contribution to how we understand the human brain and artificial intelligence. 

To explain his importance in the history of AI, let me first give you a rundown of the history of AI.  The formal concept of AI emerged in the mid-20th century, focused on symbolic logic and the belief that machines could simulate human reasoning. Alan Turing published "Computing Machinery and Intelligence," proposing the Turing Test to determine if a machine could convincingly imitate human conversation. There’s a movie on this I’d recommend, called The Imitation Game, starring Benedict Cumberbatch. It’s a 2014 American biographical thriller film directed by Morten Tyldum and written by Graham Moore, based on the 1983 biography Alan Turing: The Enigma by Andrew Hodges. The film's title quotes the name of the game cryptanalyst Alan Turing proposed for answering the question "Can machines think?", in his 1950 seminal paper "Computing Machinery and Intelligence". It’s a pretty easy to digest dramatized intro to how Turing invented the first Automatic Computing Engine, Enigma, as a cipher device developed and used in the early- to mid-20th century to protect commercial, diplomatic, and military communication. It was employed extensively by Nazi Germany during World War II, in all branches of the German military.

In Alan Turning’s 1950 paper, "Computing Machinery and Intelligence", he proposed the famous Turning Test concept. It was designed to replace the abstract question "Can machines think?" with a more practical behavioral experiment. The test evolved from a Victorian parlor game called the "Imitation Game". In Turing's version, a human judge engages in a text-based conversation with two unseen participants: one human and one machine. The judge asks questions of both participants via computer terminal to avoid physical cues. If the judge cannot consistently tell the difference between the machine and the human, the machine is said to have "passed" the test and is considered to possess intelligence. There’s another great movie on what it looks like when a machine does pass the Turning Test, the famous Ex Machina, directed by Alex Garland. I consider this his greatest work so far, even though I pretty much loved all of his work. 

Turing argued that "thinking" was too poorly defined to serve as a scientific goal. He shifted the focus from the machine’s internal "conscious" state to its external behavior, believing that if a machine could perfectly simulate human intelligence, it was effectively intelligent. The Turing Test profoundly changed how scientists viewed and developed artificial intelligence. It placed "natural language" at the center of AI research, leading scientists to focus on how machines could process and generate human speech. For decades, it led scientists to believe that intelligence was a matter of logic and symbol manipulation—if you could code enough rules to fool a person, you had built a mind. Early programs like ELIZA (1966) were built specifically to exploit human psychology to pass the test, showing scientists that "intelligence" could sometimes be faked through clever pattern matching rather than true understanding. However, later critics, such as Minsky in his work on Perceptrons, argued that "passing" a conversation test didn't mean a machine could actually solve complex problems or understand the world geometrically.

In 1956, six years after Turing proposed the Turing Test, John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon organized a conference, called The Dartmouth Workshop, which coined the term "Artificial Intelligence" and established it as a formal academic discipline. Researchers created programs that could solve algebra problems, prove logical theorems, and play games like checkers. In 1966, Joseph Weizenbaum created ELIZA, the first chatbot, which simulated a psychotherapist. By the early 1970s, the field faced a slowdown as computing power proved insufficient for the ambitious goals set by early pioneers. In 1969, Marvin Minsky and Seymour Papert published Perceptrons, which mathematically demonstrated the limitations of simple, single-layer neural networks (specifically their inability to solve problems like XOR). This work shifts research focus toward Symbolic AI (rule-based logic) for over a decade. 

So the XOR problem proposed in Perceptrons goes something like this. To understand the problem, you first have to look at how simple logic gates work.

• AND Gate: Outputs "1" only if both inputs are "1."

• OR Gate: Outputs "1" if either or both inputs are "1."

• XOR Gate: Outputs "1" if exactly one input is "1," but outputs "0" if both are "1" or both are "0."

In order to build artificial intelligence that can function like human brains, we have to start with the simplest logic, and we have to be able to translate the logic into circuit boards of diodes that can read in 1s and 0s, have the same input yield the same output as our brian logic. In the graph above, we can see that for NOT, the output should be opposite of input, for OR, as long as one of the inputs is not 0, output yields 1; and for AND, both inputs have to be 1 to yield 1, otherwise, it’s 0. This is consistent with our logic so far. However, the “XOR” gate, which is not shown in the graph, is not in the graph because it cannot be represented simply by a single diode. The “XOR” gate, checks if the two inputs are the same, which means that if the two inputs are the same, say “0 and 0” or “1 and 1”, output yields 0; but if they are different, “1” and “0”, doesn’t matter which comes first, it should yield 1, which means the two aren’t the same. This task of identifying if two inputs are the same is impossible to achieve by a single diode, as proven mathematically somehow, in the ways I don’t understand, in Minsky’s first book  Perceptrons. 

According to this book, XOR is not linearly separable. If you plot the four possible input points on a graph, the "True" results sit at opposite corners: (0,1)and (1,0). The "False" results sit at the other corners: (0,0) and (1,1). You cannot draw one single straight line that keeps the "True" points on one side and the "False" points on the other. Minsky and Papert proved that because of this, a simple perceptron is mathematically incapable of "learning" XOR—it literally cannot represent that logic in its architecture. 

The graph at the beginning of this article visualizes the fundamental mathematical hurdle that Minsky and Papert identified in early AI. A simple perceptron is a linear classifier. This means its decision-making logic is restricted to drawing a single straight line across this graph. As you can see visually, the "True" points sit at opposite corners from each other. No matter where you try to draw a single straight line, you will always end up with at least one "True" point on the same side as a "False" point. This just means that on a 2D plane, there will always be a point where the “false” and “true” interact, which makes that point both true and false at the same time, which makes this unsolvable on a 2D plane. However, it is possible to solve this on a 3D or high dimensional matrix or matrices. To solve the XOR problem, you have to move beyond a single line. The "solution" that eventually ended the AI Winter was the Multi-Layer Perceptron (MLP). Instead of one neuron trying to do everything, you add a hidden layer between the inputs and the output.

Think of the hidden layer as a way to create two different "lines" or "filters" that look at the data simultaneously. I don’t fully understand the layers, but I’m putting it here for context. According to this theory, we can have one hidden neuron acting like an OR gate, separating (0,0) from the other three points. Another hidden neuron acts like a NAND gate, separating (1,1) from the rest. A final neuron combines these two results. It essentially says: "If the first line is TRUE AND the second line is TRUE, then it's an XOR match". By using multiple layers, the machine is no longer restricted to a single flat plane. It can effectively fold or warp the mathematical space so that the "True" points end up on one side and the "False" points on the other. In the book, the authors actually acknowledged that multi-layer machines could solve these problems. However, they raised two major objections that remained true for nearly 20 years. While they knew the architecture worked, there was no known mathematical way to "teach" the hidden layers. They proved that as problems become more complex, the number of neurons and the size of the "weights" (coefficients) needed can grow exponentially, becoming physically impossible for computers of that era to handle. However, these problems were solved later with Backpropagation Algorithm and the rise of Graphics Processing Units (GPUs). More on this in the next post. ☀️

Reference / Recommended Readings

Historical & Technical Foundations

• Minsky, M., & Papert, S. (1969). Perceptrons: An Introduction to Computational Geometry. MIT Press. (The foundational text that proved the XOR limitation and the scaling problems of simple neural networks).

• Turing, A. M. (1950). "Computing Machinery and Intelligence." Mind, 59(236), 433–460. (The paper that proposed the "Imitation Game" or Turing Test).

• McCarthy, J., Minsky, M. L., Rochester, N., & Shannon, C. E. (1955). A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence. (The document that established AI as a formal discipline).

• Weizenbaum, J. (1966). "ELIZA—A computer program for the study of natural language communication between man and machine." Communications of the ACM. (The first chatbot that simulated a psychotherapist).

• Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1986). "Learning representations by back-propagating errors." Nature. (The breakthrough paper that provided the mathematical "Backpropagation" method to train multi-layer networks, solving the XOR problem).

Media & Popular Culture

• Hodges, A. (1983). Alan Turing: The Enigma. Burnett Books. (The biography upon which the film The Imitation Game was based).

• Tyldum, M. (Director). (2014). The Imitation Game [Film]. Black Bear Pictures.

• Garland, A. (Director). (2014). Ex Machina [Film]. DNA Films.

To Understand Marvin Minsky’s Philosophy

• "The Society of Mind" (1985) by Marvin Minsky: This is far more accessible than Perceptrons. Instead of math, he uses the metaphor of "agents"—tiny, mindless processes that work together to create an intelligent mind. It explains his view of how consciousness emerges.

• "The Emotion Machine" (2006) by Marvin Minsky: His later work where he argues that emotions are not separate from thinking, but are actually just different "ways to think" that our brain uses to solve different types of problems.

To Understand the Ethical & "Morbid" Side of Futurism

• "The Singularity is Near" by Ray Kurzweil: If you are interested in cryonics and living forever, Kurzweil (a contemporary of Minsky) explains the "technical advancement" perspective in detail.

• "To Be a Machine" by Mark O'Connell: This is an excellent journalistic look at the transhumanist movement, cryonics, and the quest to solve death. It captures the "brilliant but potentially morbid" energy you mentioned.

To Demystify the "Hidden Layers" and Math

• "But what is a neural network?" by 3Blue1Brown (YouTube): This is widely considered the best visual explanation of how the "hidden layers" and "Backpropagation" actually work without requiring a math degree. It visualizes the "warping of space" you described.

• "Architects of Intelligence" by Martin Ford: A collection of interviews with modern AI pioneers (including those who "solved" Minsky's XOR problem) that discusses both the technical history and the ethical "means vs. ends" concerns of AI development.