Quantum computers are now the platform for the groundbreaking artificial intelligence technology that powers chatbots.
Transformer is an artificial intelligence concept that is accomplished much more quickly than other computer science innovations. In 2017, Google researchers initially presented the idea of a transformer, a type of deep learning—a computer concept based on neural networks in the brain. Seven years later, the transformer drives the present-day AI revolution. It allows ChatGPT and other chatbots to develop advanced outputs in response to user inputs rapidly. Even if this artificial intelligence architecture is already impressive, imagine how powerful it may be on a quantum computer.
That could sound like some hurried mash-up by an excited technological entrepreneur. However, due to pure interest, quantum computing developers are exploring an identical topic to make computers perform new things. A recent study in Quantum suggested that more advanced quantum-AI combinations can theoretically solve essential issues in fields like chemistry and encryption by demonstrating that essential quantum transformers could function using cheap hardware.
Knowing which components of the input it receives are more critical than alternatives and how firmly those components relate is a transformer’s superpower. Consider the phrase, “She’s eating a green apple.” “Eating,” “green,” and “apple” are the three essential terms in the sentence that a transformer might identify. Then, it would determine that the activity of “eating” had considerably more associated with the object “apple” and less to deal with the colour “green” based on patterns found in its training data. This characteristic, which computer scientists call an “attention mechanism,” focuses on the most significant words in a phrase, pixels in a picture, or molecules in an entire sequence. The attention mechanism performs an operation that is fundamental for most youngsters but that computers had trouble with—until the ChatGPT era—by replicating how individuals perceive language.
Even though algorithms for attention today operate on supercomputers with highly efficient processors, they still rely on simple binary bits that can only store values of 0 or 1. Scientists refer to these devices, including PCs and smartphones, as “classical” machines. Contrarily, quantum hardware uses the peculiarities of quantum technology to resolve issues that are too complex for traditional computers. It is because quantum bits, or qubits, can exist in various nations, including 0 and 1. So, could scientists use qubits to create a better attention mechanism? No one expects quantum computers to be a computational miracle. Yet, we will only know if we try, according to Christopher Ferrie, a researcher on quantum computing at the University of Technology Sydney, which was not part of the current work.
 Jonas Landman, one of the study’s authors, has created quantum versions of other brain-like AI architectures compatible with quantum technology. “We were interested in examining transformers as they appeared to represent the peak of profound learning,” says Landman, an expert in quantum computing from the University of Edinburgh and member of QC Ware, a computer company. He and his fellow researchers modified a transformer for medical analysis for the new study. The quantum model classified every image from a collection of 1,600 retinal scans—some from people with diabetes-related blindness, some from people with healthy eyes—into one of five categories, ranging from minimal damage to the worst.
The procedure of creating their quantum transformer involved three steps. They had first to make the quantum circuit for a transformer—or, to put it another way, a quantum program’s “code”—before they could handle any quantum hardware. They created three variants, and mathematical arguments showed that each one could, in theory, pay attention more effectively than a conventional transformer.
The research authors evaluated their proposals on a quantum training program, a qubit emulation that operates on conventional hardware with math support. Today’s real quantum computers still have issues with temperatures, electromagnetic waves, and other disturbances that can cause qubits to get confused or become completely worthless. Emulators overcome this issue.
All the quantum transformers on the simulation device correctly classified a collection of retinal photos with a precision ranging from 50 to 55 per cent; this is an improvement above the 20 per cent efficiency that could have resulted from randomly classifying retinas in one of five different groups. Two traditional transformers with far more sophisticated networks obtained an accuracy level of 53 to 56 per cent, roughly identical to the 50 to 55 per cent range.
The scientists could only proceed to the third stage if they had successfully operated their transformers on actual IBM quantum computers, utilizing up to six qubits simultaneously. Even yet, the accuracy of all three quantum converters remained at 45–55 per cent.
It doesn’t have that many qubits—six. Specific experts believe that computer scientists must write code utilizing hundreds of qubits to construct a feasible quantum transformer that could compete with gigantic chatbot companies like Google’s Gemini or OpenAI’s ChatGPT. Such large quantum computers are currently available, but the disruptions and possible mistakes required to build a comparatively massive quantum transformer make it impracticable. Though they did not achieve the same level of accomplishment, the researchers explored more significant qubit numbers.
The team is one of many working on transformers. Scientists at IBM’s Thomas J. Watson Research Center presented a quantum variant of the graph transformer last year. Additionally, Ferrie’s group in Australia has created their own transformer quantum network idea. The team is currently working on the initial stage that QC Ware completed, which involved mathematically evaluating the design before using it.
Imagine, still, that there was a dependable quantum computer—one with over a thousand qubits and minimal interference. So, will the advantage of a quantum transformer always hold? Perhaps not. It is not appropriate to compare Quantum and conventional transformers side by side because they most likely have different advantages.
One advantage of traditional computers is their level of understanding and investment. According to Nathan Killoran, director of software development at quantum computing company Xanadu, who was not part of the new research, it will require a long time for quantum machines to reach a level up to that scale even as the technology advances. In the meantime, traditional computers will continue to expand. In our lifetimes, it might not be worthwhile to completely replace classical machine learning with a newly developed technology such as quantum computer technology because it is so solid and well-funded.
Furthermore, the kinds of issues that conventional machine learning and quantum computers excel in vary. Using their initial training data, advanced deep learning methods find patterns. Although it’s unclear if qubits are the best option for the job, it’s plausible that they could write the same patterns. It is thus because qubits are most useful in “unstructured” problems, i.e., problems whose data initially lack identifiable patterns. A quantum computer can locate a word in the phone book at a square root of the time, which would take a conventional computer, but imagine trying to discover a name without any categorization or order.
Yet there is no need to choose between the two. Many quantum researchers think a hybrid system combining classical and quantum elements will be the best application for a quantum transformer. While a traditional system processes vast amounts of data, quantum computers could tackle the most challenging chemistry and materials research issues. Additionally, data generated by quantum systems (such as decrypted cryptographic keys or the characteristics of as-yet-undiscovered materials, which are difficult for classical computers to process) may prove helpful in training classical transformers to carry out tasks that are currently mainly inaccessible.
Quantum transformers also have further benefits. At the scales at which they are currently in use, classical transformers use sufficient power, so the United States companies are forced to continue operating coal-spewing, carbon-spilling plants to fulfil the power requirements of new data centres. The desire for a leaner, more energy-efficient machine that reduces the energy burden is also the aspiration of a quantum transformer.
