Artemis has a powerful wireless technology called pCell. It appears that pCell might have a high computational cost resulting in scaling issues. They might be able to solve this problem by replacing portions of their algorithm with neural networks. If they have an accurate simulator, they could even replace the entirety of their algorithm with deep reinforcement learning. This could be a good application of analog computer based neural networks.

It is known that quantum mechanics is Turing complete. If the Einstein field equations are Turing complete, then a “relativity computer” could be built. (It is known that some differential equations are Turing complete.) Unification of quantum mechanics and general relativity is then a matter of converting quantum programs to relativity programs.

The author of Gorgonia, a neural network library, had an interesting Strange Loop talk about reimplementing AlphaGo. The problem they ran into is that it would have cost them $70,000 to train a competitive model, so they focused on training their algorithm for games with a smaller board. What if a small neural network could be scaled to work on a larger board? The approach that I propose is based on program synthesis.

CRDTs, conflict free replicated data types, are data structures that can be synchronized across an unreliable network with consistency guarantees. Here is a talk on CRTDs from Martin Kleppmann. His group has produced a CRTD library called automerge which gives an application a JSON-like data structure that is replicated across multiple machines. Why is this significant? Open Source currently can’t compete with SAAS because users want their data backed up and available on all devices, and unfortunately servers cost money.

There are noninvasive approaches for reading neural activations from the brain, but a noninvasive method for writing into the brain currently doesn’t exist. The value of write capability can be seen from this article. The article describes how neural activity was transferred from better trained rats to rats with less training. Then the rats with less training performed like the better trained rats. The rats had electrodes placed inside their brains, but I think it is possible to do such an experiment on humans without the electrodes.

In the paper Mining Energy In An Expanding Universe a source of free energy is described. In an expanding universe free energy can be produced by connecting widely separated bodies with strings. Another way of producing free energy is the Casimir effect. This works by placing two conducting plates parallel to each other. A net force between the plates is then produced. So, there are two known ways of producing free energy.

There is an interesting article on the computer scientist Constantinos Daskalakis. The article talks about his work on Nash equilibrium, machine learning, and neurology. The obvious idea here is that perhaps neuron behavior could be explained by game theory. Each neuron seeks an optimal strategy for playing some game. After all organisms operate through competition. This could mean that consciousness is a Nash equilibrium. In order to create AGI all we need to do is figure out what game neurons are playing.

Wikipedia edit wars are a well known tragedy of the commons problem. What is needed is an ownership model similar to Github, but then which version of an article is the correct version? Machine learning could be used to produce one correct article from all of the different article versions. The different article versions would be auto merged together. The Transformer machine learning algorithm could solve the auto merge problem: Generating Wikipedia by Summarizing Long Sequences.

Mind pod is an iPod that attaches to both your brain and ears. The purpose is to generate music from your brain activity. There are two enabling technologies required for this idea to work. One enabling technology is a wearable brain sensor comparable to an MRI machine: Openwater, and the second enabling technology is a neural network capable of generating music: SampleRNN. Music generated from SampleRNN can be found here. To create music from brain activity all one has to do is record the brain activity from a user listening to some music and then use the recorded brain activity and the corresponding music to train a neural network.