Explainability and reliability: the two great challenges of quantum computing

The advent of artificial intelligence (AI) raises major ethical questions for societies. The quantum is likely to do the same.

When applying algorithms to sensitive topics – say HR data or justice – one of the requirements is that you have to be able to explain how you get the result. No question of having black boxes recommending this or that option without further argument.

If we still have to wait several years before really seeing operational quantum computing, the prospect of its “superiority” raises the same type of question.

It will indeed quickly become impossible to simulate, on a classical computer, the algorithms executed on a quantum material worthy of the name. And knowing the astronomical amount of parameters a quantum computer can ingest to search for an answer, how could a human plumb its depths to determine whether or not a calculation makes sense?

“If a quantum computer can effectively solve a problem, can it convince an observer that its solution is correct? », asks Marc Carrel-BilliardHead of Global Technology Innovation at Accenture.

Already relevant use cases envisaged for quantum computing

Quantum computing does not exist yet, but we know how it will behave and researchers understand better and better in which fields it can be used in a relevant way.

Head of science and technology at IBM, and head of IBM Research Quantum Europe, Heike Riel explains that “it’s not just a question of technological beauty. We seek to generate value. It’s a journey: develop the technology, find the most suitable phase-ahead applications, demonstrate and prove that value, and then develop the hardware and software. »

The journey has already begun, for example, at Eon Energy, which has joined the IBM Quantum Network. The transition to greener sources, such as solar and wind, has multiplied the types of energy to be managed by a power grid. Quantum computing could help optimize these networks if, in the future, companies and many households become producers of electricity via their own photovoltaic systems or electric vehicles, thanks to initiatives like the Vehicle to Grid (V2G) project of Aeon.

“It’s a journey: developing the technology, finding the most suitable early stage applications, presenting and proving value, then developing the hardware and software. »

Heike RielHead of Science and Technology at IBM

In this project, the batteries of electric vehicles are connected to the network as flexible storage media. Thus, it becomes possible to balance fluctuations in the production of renewable sources. Quantum computing would drive these processes more efficiently and effectively.

“All these sources have different characteristics; forecasts are becoming more complex,” recalls Heike Riel. “You have to optimize the system in real time. However, the complexity grows exponentially with the number of parameters, to end up becoming a problem that is difficult to solve using conventional computing. »

Another example of application, in science, the theoretical physicist and chief scientist at Cambridge Quantum Computing, Bob Coecke assures that the behavior of atoms and molecules – which are governed by the laws of quantum mechanics – must be able to be modeled on a quantum computer governed by these same laws.

“In view of the complex functioning of quantum mechanics, simulating a physical matter [au niveau moléculaire] is more and more expensive,” he explains. In fact, just in terms of storage, he explains, it would be impossible to adapt a traditional computer to these kinds of problems.

Simulating new materials and modeling the behavior of particles are two of the greatest use cases envisaged for quantum computers. In August 2021, Nicholas Rubin and Charles Neill, two research scientists at Google AI Quantum, wrote a blog post on an experiment that aims to create a complex chemical simulation using a Hartree-Fock model from computational physics.

“An Accurate Numerical Prediction [sur] chemical processes, starting from the laws of quantum mechanics which govern them, can open new ways in chemistry, and help to improve a broad spectrum of industries”, write the researchers.

Reliability of results and quantum noise

However, these promises come with their share of challenges. The two Google researchers find, for example, that their algorithms are still hampered by the high error rate of the first quantum computers.

“If a quantum computer can effectively solve a problem, can it convince an observer that its solution is correct? »

Marc Carrel-BilliardHead of Global Technology Innovation at Accenture.

Like the ability of classical neural networks to tolerate imperfections in the data, the pair explains that in their experiment, the VQE algorithm (for Variational Quantum Eigensolver) tries to optimize the parameters of a quantum circuit to reduce the “noise” that interferes with the algorithm.

IBM is working on the same problem. With few qubits, it remains possible to simply verify the result of a quantum algorithm on a quantum computer, by comparing it to the result of this same algorithm on a classical machine which simulates quantum computer behavior.

But the method is only possible as long as the number of qubits remains low enough, says IBM’s Heike Riel. The key point here is to understand how the “noise” of many qubits influences the system by producing erroneous results.

Today, IBM continues its roadmap with a 128-qubit system and wants to “provide proof that error correction can work”, says Heike Riel, “we are working to verify the results”.

Explainability

Mark Mattingley-Scott, Managing Director Europe at manufacturer Quantum Brilliance, raises another challenge: explainability.

“It’s one of the paradoxes of quantum computing. When we reach the stage where it is useful – when a quantum algorithm can perform calculations at a speed and with an accuracy impossible with a classical computer – it becomes impossible to directly verify the accuracy of the results obtained”, he summarizes. .

“We can check the correctness of the process on reduced versions of the same problem, which we do every day with classical algorithms, but there will be no means of ‘control’ as such. »

But quantum computing is, in essence, non-deterministic. Mark Mattingley-Scott therefore insists that the results produced are based on probabilities. “A quantum algorithm works by employing a quantum mechanism that constructively reinforces the ‘right’ response, and destructively suppresses the ‘wrong’ response,” he explains. But this construction remains the fruit of probability. “There is therefore always a certain amount of uncertainty. And using a classical computer to validate a quantum computer is only possible at the methodological level, not at the data level itself. »

“It’s a paradox of quantum computing. When it becomes usable, it also becomes impossible to directly verify the accuracy of its results.

Mark Mattingley-ScottManaging Director Europe at Quantum Brilliance

For his part, Bob Coecke, of the specialized company Cambridge Quantum Computing, considers that the principle of compositionality and the theory of categories can help to understand what is happening in a quantum computer.

The Belgian researcher explained this idea in a book co-written with Aleks Kissinger (“ Painting quantum processes “). From a general point of view, the book looks at how to break large quantum problems into smaller components. According to Bob Coecke, these “small blocks” are more understandable and verifiable.

In a similar way, the team of Mark Carrel-Billiard from Accenture is working on how to map certain problems into subgroups of mathematical problems. These “sub-problems” are then coded with SDKs and libraries from several quantum platforms. By testing the programs on different quantum hardware architectures, it then becomes theoretically possible to determine whether they produce consistent results.

In some cases, validation can also be done “in vivo”. In chemistry, for example, Michael Biercuk, CEO and founder of Q-CTRL, explains that “for a molecular structure or for chemical dynamics calculated on a quantum computer, it may not be possible to validate the calculation of the simulation itself. On the other hand, we can do a real chemical experiment [ou une analyse comparative avec des molécules connues] to validate the results. »

To understand or not to understand, that is the question

Quantum computing will also remain one approach among others. “If you have a complex optimization problem to solve, it doesn’t matter how or what type of computer you use as long as you get a result in the fastest and most efficient way possible”, assures Heike Riel from IBM.

From IBM’s perspective, a complex computational problem often has several distinct parts. Some will be treated with quantum computing, others with classical computing.

And even in the first case, an understanding of quantum mechanics will not necessarily be necessary. Once the basics are laid, “you will need a model developer who does not need to understand quantum computing in detail, but who must know how to describe the problem and use the best option to solve it”, predicts Heike IBM Riel. “The model developer does not have to bother with advanced quantum knowledge. “, she insists.

But explainability and reliability will remain two imperatives that are strongly intertwined with these technologies.

We want to say thanks to the author of this short article for this outstanding content

Explainability and reliability: the two great challenges of quantum computing


You can find our social media profiles as well as other related pageshttps://www.ai-magazine.com/related-pages/