Analog 2.0 - When it comes to simulating a human brain in the not-too-distant future it will likely run on
Memristors.
Built into neuromorphic circuits, computer architectures will emerge that have very little in common with current static
computer hardware.
These processors are no longer programmed but optimize themselves analogously to evolutionary processes. The hardware becomes software.
The signs of such a future of computers is current to read the promises of Big Data, Intelligent Assistants (Siri, Cortana) as well as autonomous driving.
Behind these different ventures is a common concern to identify patterns in big data, as close to real time as possible. The focus here is less on the precision of
data analysis is less important than recognizing fuzzy or noisy trends and shapes. Accordingly, digital precision is not required. These projects suggest
to resort to analog computing methods, which are far ahead of the digital in terms of speed.
As the fundamental fourth electrical component, along with the resistor, capacitor and coil, the memristor differs from these in that, due to its inherent
hysteresis, it memorizes its past state. As a variable non-volatile resistor, it is used in hardware neural networks as a memory for the continuous
connection weights. On the other hand, when operated digitally, the memristor suggests a different logic.
The conjunction used in CMOS architectures is replaced by the implication, which can be seen in the sense of a conditional as well as a temporal link,
and thus embodies Günther's polycontextural logic, a temporal and context-dependent logic.
According to Michael Conrad such architectures are urgently necessary, in order not to continue to pay the 'price of programmability', at the expense of programmability
polynomial growth in the interaction of the computing units. Rather, all computational units should be linkable with each other and mutually
evolutionary optimization. This approach would say goodbye to the Von Neumann architecture, with its separation of memory and computational units and the associated
bottleneck problem, the latency that occurs when data is loaded from memory into the processor. There is then talk of memory computers, computing elements
used as both memory and computational modules, operating in parallel.
By extension, this refers to a machine unconscious, because these sub-symbolic
analog computational processes run below the discrete digital symbols, and it is impossible in retrospect to reconstruct the state changes.