In view of the above, there will be at least 64 bits of incoming analog-to-digital register and 64 bits of outgoing digital-to-analog register at each surface of each cell. Given a cubic configuration, as proposed above, There will be at least 384 bits of data incoming and 384 bits of data outgoing per cell. The logic circuitry necessary to process 768 bits of incoming and outgoing data is still unspecified.
The mechanical movement of the cells relative to each other is determined by the state of at least two motor poles at each surface of each cell. In order to have forward and reverse motion in more than one axis at each cell, it may be necessary to have four poles and two sets of track latches at each surface. If each axis at each surface is four-state (two-bit) plus one bit for each latch, then there are an additional thirty-six bits of register information encoded in the motor elements in each cubic cell.
The tactile detection mechanism at each face of the cell can be used to detect contact between cells as well as mechanical contact from outside the sculpture. Given that there are two bits of mechanical pressure information at each face, then there are an additional twelve bits of register information in the tactile detectors of each cubic cell. This brings the total I/O register space of a single cell to 816 bits (102 8-bit bytes or 204 four-bit half-bytes).
The programming problem is unspecified at this point, except to the extent that we know the communication between cells. A software data structure can at least be constructed at this point which will simulate the state of data transmitted into and out of each cell. There will be at least a capability for passing messages from cell to cell, determining the spatial position and logical address of a cell and the ability for an addressed cell to decode and execute a command.
This model intends that each cell should be able to receive from the host a small, simple program to run semi-autonomously, processing messages it receives from the cells around it and from its external sensory environment. It is intended that the simplicity of the algorithm in each cell should give rise to complexity of functioning at the large scale by virtue of the sheer number of collective cells.
The host-resident program will ultimately take the form of an interpreter/compiler for a fine-grain parallel machine. It is unclear at this point whether the machine will in fact be a cellular automaton or a Single Instruction, Multiple Data (SIMD) machine, like the Connection Machine. It is known that a cellular automaton can be constructed in software on a machine with SIMD hardware architecture.
A cellular automaton model would possibly allow the cells to run asynchronously with minimal data communication, while SIMD design might necessitate clock and address leads which are transmitted throughout the structure. Compilers for SIMD machines are more well known than are compilers for cellular automata. An attractive solution might be to arrive at a nearly SIMD model in which the cells communicate asynchronously.
The only operation which will require precise synchronized action will be the moving of a bank (row, column or plane) of cells at a linear interface. The motor action must be timed for the movement to occur, but it is possible that a specialized message-passing sequence can be created which will allow rows of cells to operate in this synchronized manner.
The multiplexer/demultiplexer interface to the host computer will consist of a processor whose function is to interpret lines of instructions from the host computer and process them for parallel execution on the sculpture. The function of this processor will be to prepare message packets, distribute them to the cell faces in the base of the sculpture, collect messages from the cells and reassemble them into results to be sent to the host.
Given that the global structure of the sculpture, i.e., the position of all cells at any one time, will in general only be completely known at the highest level, the cells themselves will in general not have all this information. In order to limit the propensity for messages to "echo" about through the sculpture, it may be necessary to construct virtual data paths along which message passing will be concentrated.
It seems reasonable that a single "ping" message can be issued from the multiplexer interface into the base of the sculpture, the effect of which will be to establish the spatial and logical addresses of the cells and to establish the internal data pathways. Each cell may therefore have some knowledge of which cells are reachable through it in each direction.
Internal message data paths will have a tree-like form in which the "trunks" originate at the cell faces in the base interface and the "leaves" are the cells on the outer periphery of the sculpture. This structure has compelling potential to implement hierarchical processing algorithms, particularly relative to the peripheral cells as "sensory" inputs. It would be attractive to choose a natural neural model for solving these problems, employing behaviors such as lateral inhibition and so on.
The highest-level mode of functioning imaginable for this sculpture would be one in which each cell has an entirely self-contained ability to learn, the total effect of which when applied across all cells is a transcendental intelligence. The interactive learning process, between the artist, sculpture and viewing audience will in this case be an unprecedented and unpredictable synergistic phenomenon.
3-D Kinetic cellular Automaton -- Copyright 1995 Stewart Dickson