Analog and Probabilistic Computers: From Quantum Atom To Living Body

Описание: This highly accessible and short TED-talk-like video explains why analog and probabilistic computers in all of physics (atoms), chemistry (molecules), or biology (cells and brains) can inspire incredibly fast and incredibly energy efficient architectures in A.I. and in all of computer science. It suggests that they may take us back to the future, back to principles that the founder of quantum computing, Richard Feynman, and the founder of computer science, Alan Turing, already understood. Such mixed-signal collective analog computers are programmable and scalable with traditional digital languages, microprocessors, and bits to arbitrary precision and complexity. In fact all digital computers are special cases of such analog computers, not the other way around. Quantum computers are also just a special case of such analog computers, which can actually mimic many of their advantageous properties at room temperature and at orders-of-magnitude lower cost and energy in VLSI chips.

Many ground-breaking VLSI integrated-circuit chips, starting from when Dr. Sarpeshkar was in the Physics of Computation Mead lab at CalTech (Mead is the founder of VLSI and a National Medal of Technology and Kyoto prize winner) have reduced these ideas to practice as outlined in the author's book in the video (written when he was a tenured professor at M.I.T) and in multiple publications spanning three decades.

Nature uses both continuous and discrete mathematics to compute in physics, chemistry, and biology. It is time for man to also do so! Multiple applications in A.I, embedded A.I., portable A.I, highly-energy-efficient A.I., cybersecurity, drug design, simulation, chemistry, optimization, materials science, and other applications can leverage such mixed-signal supercomputing chips. They can thus accelerate performance without blowing power budgets, a HUGE problem right now.

Furthermore, such A.I. can preserve privacy and security so your information is yours by default with advanced hardware level security that no software can hack; and not just be based on traditional neural-network A.I. and large-language models.