Is Quantum Computing the Ultimate of Artificial Intelligence?
Artificial Intelligence now-a-days plays a significant role in modern computer science and engineering, as the ultimate objective of the computer science research community is to develop a humanoid that can act perfectly as a human with human like intelligence. The problem undoubtedly, is many fold complex of the present state-of-the art status and needs serious thought. The task is mammoth and needs collective work.
It is very difficult to define Human intelligence. However, the power of decision making and problem solving is the back bone of human intelligence, and artificial intelligence is practically the same power possessed by a machine. For artificial intelligence, a machine should be fully capable of generating solutions of a problem or should be capable generating decisions under a certain situation. This generation must be unambiguous, unique and fast. Such generations should be based on online learning. As the designed humanoid is preferred to act as a human, we must kep in mind that it must be interactive in nature with its inputs from the surrounding external 3D world. These inputs are basically multimodal in nature, e.g., stationary and non-stationary images which could be meaningful or nonsense, texts of the same kind, audio, music, speech etc. The humanoid should have the learning capability in the interactive online mode. Attention based learning mechanism, generation of embedding spaces for each subject and underlying encoding-decoding mechanism is a huge task to be carried out by a computer. It should be noted that even the learning mechanism alone for a single classification problem takes a week-long time. Thus, we can feel how important the computing power of a machine is. It is almost impossible to think of such a humanoid behaving like a human with human-like intelligence.The computing power must be huge, storage must be huge and the overall machine must be superfast. One should therefore search for such a super computer.
Conventional computers use bits. They are either a stream of electrical or optical pulses representing 1s or 0s. They are viewed as binary digits. Images, speech, videos, songs, any kind of transactions, i.e., everything we come across in our daily life are all data. These data are nothing but strings of 1s and 0s. Similarly, computer programs are also data. Computer programs operate on data and produce our desired results which are again strings of 1s and 0s. All kinds of computations use binary digits. It is, therefore, not very difficult to understand the impact of binary digits in understanding and functioning of our computers.
A quantum computer uses quantum bits or qubits in the place of bits in conventional computers. Qubits are subatomic particles such as electrons or photons, could be also ions whose charge or polarization can act as a representation of 0 and/or 1. An engineering challenge to the research community is therefore, to generation and manipulate qubits in an easy and straightforward way. Some big companies, like Google, IBM, Rigetti Computing etc. use super conducting circuits at very low temperatures. Trapping individual atoms in electromagnetic fields on a silicon chip in ultra-high-vacuum chambers, are also a feasible and done by some other companies. In both the cases, the goal is to have the qubits in a controlled quantum state. A connected set of qubits is found to possess more processing power than an equal number of binary bits. This phenomenon produces two effects known as superposition and entanglement. Before getting a broader picture about the superposition and entanglement, we explain qubits in a greater detail.
Bloch Sphere Representation of Qubits: A Fundamental Concept of Quantum computers
In computer science, we all know that in 1930s, Alan Turing developed a theoretical machine, known as Turing Machine, that consists of
(a) A head which can read or write a symbol 0 or 1 at a time and move either to the left or right or remain in the same position, depending on the symbol read from the tape.
(b) An infinite tape on either side of the head, marked-off into square cells in which symbols can be written. The tape is considered to be unbounded 1-D memory, filled with blank characters, 0s unless otherwise specified.
A read-write device reads these symbols and blanks, which gives TM required instructions to perform a desired program. A Quantum Turing machine, on the other hand, can be thought of as consisting of a tape that exists in a quantum state, as does the read-write head. This states that the symbols on the tape can be either 0 or 1 or a superposition of 0 and 1. Thus, the symbols are both 0 and 1, and all points between them at the same time. Note that, when a Turing machine (TM) can only perform one calculation at a time, a quantum Turing machine (QTM) can perform many calculations simultaneously.
So, unlike the conventional computers that manipulate bits in one of the two states, Quantum computers manipulate qubits that are in many states, i.e., qubits are not restricted to two states only. These states' information are encoded as quantum bits, or qubits. We have already noticed that qubits may represent photons, electrons, atoms or ions. Their control devices, working together, may provide computer memory or a processor.
Now, it is clear that due to the simultaneous occurrence of qubits in multiple states, quantum computers are many times more powerful than today's awfully powerful supercomputers. The simultaneous occurrence of qubits or superposition lends quantum computers inherent parallelism; as a result of which a quantum computer can do millions of computations at time when the most powerful supercomputers can do only a few computations.
To have a rough idea about the superposition of qubits, let us consider the spin state of an electron in a magnetic field. The spin of the electron is a quantum state and it can be either 1 or 0. 1 means spin-up in which the electron is aligned with the magnetic field, while 0 means spin down which is the reverse direction of the magnetic field. One can change the direction of this electron's spin by using laser energy in the form of pulses. If q units of laser energy is needed to change the spin-state from spin-up to spin-down state, then the electron will enter a superposition state and its spin can be represented using superposition law of Quantum Mechanics. Thus, each qubit can be a superposition of 1 and 0; and so the number of computation that a quantum computer can do is 2N, where N is the number of qubits. For N=500, Margaret Rouse in her blog-post has told thata quantum computer comprising of 500 qubits would have a potential of doing 2500 calculations in a single step. This number obviously is awfully huge. The machine, therefore, does truly parallel processing with capacity beyond any parallel processors.
Geometric representation of qubits is very important to visualize the states of qubits. This is normally done by Bloch sphere. Ian Glendinning, in February 16, 2005, has described in an article ``The Bloch Sphere" in QIA, TechGate, how states of qubits can be demonstrated. The Bloch sphere, shown below, provides a concept of generation of all qubits.
We can generate all the points on the Bloch sphere. But only one half of the Bloch sphere needs to be considered because the points in the lower half differ only by a phase factor of -1.
The qubit value in superposition can be determined using a quantum mechanical phenomenon, called entanglement. Under an external force to two atoms, the second atom act according to the behaviour of the first atom. An atom can spin in all directions but the moment one of its values is chosen, the second one immediately at the same time sets its spin in the opposite direction. This behaviour is very peculiar and is known as entanglement. Thus, it helps to get the value of the qubit. But at the same time we musk that to need some control devices also to control or manipulate the qubits. This is normally done by some techniques widely known as superconducting circuits, semiconductor impurities, ion-traps, quantum dots etc.
One more problem in quantum computers is heat and it increases when every qubit pair is added to the system - one of the great findings by Dzurak.
It is therefore, clear that a quantum computer has supremacy over any conventional computer by a factor of a million and is quite capable of handling today's artificial intelligence problem very easily but the situation is far behind as such a quantum computer is not yet available.
We believe scientist will breakthrough and achieve remarkable success.
Dr. Sambhunath Biswas
Professor, Machine Intelligence Unit,
Indian Statistical Institute, Kolkata (ISI)
Head, Computer Science & Engineering
Techno India University, West Bengal.