What kind of computing machinery do we need to advance Artificial Intelligence (AI) to human level? At the dawn of computing, one of the founding fathers, Alan Turing, believed that AI could be approached as software running on a universal computer. This was a revolutionary idea given that during his time, the term "computer" was generally referred to as a human hired to do calculations with pencil on paper. Turing referred to a machine as a "digital computer" to distinguish it from the human one.In the context of AI, Alan Turing is remembered for his Imitation Game, or later referred to as Turing Test, in which a machine strives to exhibit intelligence to make itself indistinguishable from a human in the eyes of an interrogator. In his landmark paper, "Computing Machinery Novumind Inc, Hardware Engineering, Santa Clara, California, USA Corresponding author: C.-P. Lu Email: cpl@novumind.com and Intelligence" [1], he tried to address the ultimate AI question, "Can machines think?" He reframed the question more precisely and unambiguously by asking how well a machine does in the imitation game. Turing hypothesized that human intelligence is "computable, " which has a precise mathematical meaning famously established by himself [2], as a bag of discrete state machines, and reframed the ultimate AI question as Are there discrete machines that would do well (in the imitation game)? [1] But what exactly are the discrete state machines to win the imitation game? Apparently, he did not know during his time; but witnessing the extreme difficulty of building a non-human, electronic computer himself [3], he envisioned only one machine, the Universal Digital Computer that could mimic any discrete state machine. Each discrete state machine can be encoded as numbers to be processed by a universal computer. The numbers that encode a discrete state machine become software, and the computing machinery became the "stored program computer" Thereafter, the history of computing has been mainly the race to build faster universal computers to answer the following challenge:Are there imaginable digital computers that would do well (in the imitation game)? [1] AI researchers and thinkers have been advancing AI without worrying about the underlying computing machinery. People might argue that this applies only to traditional rule-based AI. However, even connectionists have to translate their connectionist systems into algorithms in software to prove and demonstrate their ideas. We have been seeing advances and innovations in Deep Learning completely decoupled from the underlying computing machinery. Today, we use terms like "machines", "networks", "neurons", and "synapses", without a second thought about the fact that those entities do not have to exist physically. People ponder about a grand unified theory of Deep Learning using ideas like "emergent behaviors", "intuitions", "non-linear dynamics", believing that those concepts could be adequately represented or approximated by software. According to Turing, any fixed-function Deep Learn...