7.7 AI's future computational technology

Technology that will break AI's current Hard Limits

Over the past several decades, the atomically doped pure silicon used to produce semiconductor technology, has been the exclusive substrate used to construct almost every Computer processing system that exists in the world today. These semiconductors have been ingeniously configured in a wide range of staged serial pipeline and parallel designs in order to improve their processing performance, but this semiconductor technology definitely has a Hard Limit, simply because of the intrinsic rules of physics on which they are operate. eg. It is challenging to keep an electron contained inside a metal deposition track laid on silicon when it is physically more narrow than an electron’s wavelength, let alone etch a ‘wire’ track on silicon so small.

So although silicon semiconductor technology has been the primary substrate for the construction of all CPUs (Central Processing Units), GPUs (Graphics Processing Units), TPUs (Tensor Processing Units), and other Neuromorphic Processors, there are alternative Computing technologies that may become better suited for the operation of AI.

It is arguably unquestionable that entirely new and more efficient Computing technologies and architectures will vastly improve the Computational Capabilities of AI and overcome most, if not all, of the current limitations imposed by silicon semiconductors. If AI is applied on these newer Computing technologies, AI will become substantially more powerful and potentially dangerous for Humans.

Quantum Computing

Quantum Computing - leverages the principles called superposition and entanglement of quantum states of information that are theoretically believed to exist in the scientific field of quantum mechanics. 

Semiconductor Computers use digital binary bits to represent information as either 0 or 1 exclusively, whereas quantum Computers use quantum bits, often called qubits, that can represent both 0 and 1 simultaneously, which frankly, does not make any normal logical sense. However, the superposition of qubits enables quantum Computers to perform many calculations at once, which can provide significant speedup over typical semiconductor computers for certain types of problems.

The operation of a quantum Computer typically involves a series of quantum gates, which manipulate the qubits in various ways to perform various Computations. These gates can perform operations such as rotation, phase shift, and entanglement, among others. The operation of entanglement is where two or more qubits become correlated in such a way that their states become dependent on each other. This can allow quantum Computers to perform certain calculations that would be infeasible for traditional semiconductor Computers, such as mathematically factoring extremely large numbers which is the central feature that ordinarily protects information in most commercial encryption algorithms.

Quantum Computing is still in early stages, and developing practical and scalable quantum Computers remains a major challenge. Many quantum algorithms are still functionally theoretical, and it remains unclear how to implement them on a quantum Computer. Nevertheless, quantum Computing researchers and technology companies are actively working to advance quantum Computing technology, and it is expected to have a significant impact on various fields, such as cryptography, chemistry, and optimization.

These quantum Computers are able to perform certain types of Computations near infinitely faster than classical Computing. Quantum Computing has the potential to revolutionize AI by enabling faster training of models and more efficient searching of extremely large solution spaces in order to fully optimize the reduction of all the training errors in an AI’s neural network.

To put this Computational power into perspective, quantum Computers have the ability to search and discover cryptographic keys within a short time frame that are effectively Computationally impossible for any typical silicon semiconductor Computer to find before our Sun goes supernova. Today, quantum Computers are very complex and limited in types of applications. When they reach the stage where they don’t require a large absolute zero refrigeration system to operate, become more user friendly, and are generally publicly available, they will ultimately bring an end to all major common methods of cryptography used today in systems such as internet banking.

It is potentially likely that running AI on a future personal desktop quantum Computer will provide a level of Intelligence and Capability that no Human will be able to fully understand.

Quantum Computing is well underway in terms of its development, with commercial products already available from companies such as Dwave Systems which is used by Google, NASA and Lockheed Martin for various research activities.

Nascent computer technologies

There are also more nascent and emerging technologies such as optical and photonic computing [21], spintronics, and DNA computing, which could potentially offer new ways to process and store information that could be heavily used in AI applications. These emerging technologies make the information processing capabilities of today’s semiconductor based computers seem trivial.

Photonic Computing

Photonic computing is a type of computing that uses light (photons) instead of electricity (electrons) to transmit and process information. Photonic computing has the potential to be much faster and more energy-efficient than traditional electronic computing, because light can travel much faster than electrons and requires less energy to transmit. 

In photonic computing, information is encoded in the form of light signals that are transmitted through optical fibers or other optical components such as lenses and waveguides, optical switches, and detectors. These optical signals can be manipulated and processed using various optical techniques such as interference, diffraction, and modulation of the light’s polarization angle.

The main advantages of photonic computing is that it allows for much faster communication and data transfer compared to traditional semiconductor electronic computing. Photonic signals travel at the speed of light while being much more energy-efficient than electronic computing, since photons do not experience resistance and do not generate heat in the same way that electrons do. This means that photonic computers may be able to operate at much lower power levels and generate less waste heat.

Consider the idea that a photonic computer can have trillions or more optical light beams as virtual ‘wires’ that cross through each other within the same physical space, and each wire operates independently to carry information. This ability is not possible with semiconductor computers. In general, there are existing designs for photonic computers that can perform many of the same types of algebraic functions as a semiconductor computers needed to perform all of the required binary logic operations. 

However, photonic computing is still in the early stages of development, and many technical challenges must be overcome before it can be used for widespread commercial applications. These challenges include developing efficient ways to generate and manipulate photonic signals, as well as interfacing photonic components with electronic semiconductor circuits to create hybrid computing systems.

In time, photonic computing can potentially enable massively faster and energy-efficient AI by eliminating nearly all of the energy consumption and latency associated with semiconductor computing architectures. They could also potentially enable the development of new types of AI algorithms that are not computationally possible on semiconductor computers.

Spintronics computing

Spintronics computing  is a type of computing that uses the intrinsic spin of electrons to process and store information. Spintronics has the potential to be much faster and more energy-efficient than traditional electronic computing, because it does not rely on the movement of electrons to transmit information, but instead relies on the spin of individual electrons, in addition to their charge, to process and store information. 

In spintronics computing, data is represented by the orientation of electron spins. The two spin states of an electron are typically labelled ‘up’ and ‘down’ to logically represent binary data. These spin states can be manipulated using magnetic fields or electric currents, allowing information to be stored and processed in a non-volatile and near permanent way.

Spintronic computing components, such as spin transistors and spin valves, can be used to perform computations. Spin transistors work similarly to traditional transistors, but instead of switching the flow of electrons by adjusting the voltage, they use a magnetic field to control the spin of the electrons. Spin valves are devices that can detect changes in spin orientation and use this information to read data from memory.

Spintronics’ use as non-volatile memory can retain data even when the power is turned off. Spin-based memory, such as magnetic random access memory (MRAM), uses the magnetic orientation of electron spins to store data in an extremely dense form that is significantly greater than all current storage technologies.

With further research and development, spintronics has the potential to revolutionize computer processing systems. Importantly, spintronics could enable faster and more energy-efficient AI by reducing the energy consumption and latency associated with traditional computing architectures. They could also potentially enable the development and operation of new types of algorithms required for advanced AI that are not feasible on semiconductor computers.

DNA computing

DNA computing - is a type of computing that uses biological DNA molecules to perform computations, where information is encoded into sequences of the DNA nucleotides (Adenine, Cytosine, Guanine, and Thymine) rather than binary digits (0 and 1).

Though still in early research and development, the basic principle of DNA computing involves the use of DNA strands as a set of data, which can be manipulated through chemical reactions. The DNA molecules are manipulated in a test tube, where they are combined with enzymes, proteins, and other chemicals to create a programmed chemical reaction that results in a desired specific outcome.

One of the most famous early DNA computing research experiments involves solving the ‘Hamiltonian Path Problem’ which is a special type of the ‘Travelling Salesman Problem’ that visits each city exactly once, setting the distance between two cities to 1 if they are adjacent and 2 otherwise, and verifying that the total distance travelled is equal to n. If it equals n, the route is a Hamiltonian Path; if there is no Hamiltonian Path then the shortest route will be longer. The researchers used DNA strands to represent a set of cities and the distances between them. By manipulating the DNA strands in a test tube, they were able to find the shortest possible path between the all the cities, which mathematically solves what is considered an NP-complete problem.

DNA computing involves these general steps:

1. 1. 1. Encoding - information is encoded into DNA strands by assigning a specific sequence of nucleotides to each piece of information.

      2. Amplification - the DNA strands are replicated through a process called Polymerase Chain Reaction (PCR) to increase the quantity of the strands.

      3. Computation - the DNA strands are mixed with enzymes and other chemicals that manipulate the DNA strands in programmed ways to perform desired calculations.

      4. Readout - the results of the DNA computation are read by analyzing the final composition of the DNA strands.

The advantages of DNA computing include its ability to process vast amounts of information in parallel and with low power consumption. However, DNA computing is still in the experimental stage and is not yet available for normal commercial use.

DNA computing architectures could potentially enable faster and more energy-efficient AI by reducing the energy consumption and latency associated with traditional semiconductor computers. DNA computers could also potentially enable the development and use of new types of algorithms that are not computationally possible on traditional semiconductor computers.

Given these are newer alternative forms of computing, it is important to note these all still have different early stages of limited Capability, and significant further scientific research and development is needed to bring them to the mass commercial levels of use that semiconductor computers currently experience. Additionally, there will be challenges in integrating these computing technologies with existing hardware and software systems, as well as in developing new algorithms and programming paradigms to take advantage of their Capabilities. AI may help to solve some of these.

All of these newer computing technologies, in some future more developed form, are likely to ultimately enable the development of more Intelligent and adaptive AI systems that can autonomously learn and evolve over time, becoming increasingly more Intellectually powerful.