Since the development of Artificial Intelligence (AI) and the early substantial work of Alan Turing and others, the primary goal has been to create a digital or physical machine that acts intelligently.
The term intelligence itself is a suitcase word. It can summarize behavioural traits that we assign to biological organisms operating in and manipulating their environment advantageously, for example. Since the underlying algorithms of intelligence have not yet been understood, the last decades of R&D around AI have focused on behavioural patterns and skills that we consider intelligent, including computer vision, agent-oriented behaviour or logic.
In our current era of AI, we do not create intelligent behaviour in machines, we create imitations of intelligent behaviour. Today’s most sophisticated AI systems, like self-driving vehicles, leverage the latest innovations in AI. Despite the performance of these systems continually improving, something fundamental seems missing as we strive to develop Artificial General Intelligence (AGI).
We call today’s AI algorithms that operate with limited performance in narrow domains “narrow AI”. Many tasks that narrow AI and future AGI systems depend upon will continue to be efficiently computed by classical high-performance computers (HPC), graphics processing units (GPU), neuromorphic processing units (NPU) and specialized hardware. These tasks include all data pre-processing, as well as significant parts of the algorithms that equip AI systems with the ability to learn from data in a manner similar to humans.
However, we’re still missing the explicit description of the algorithms which make our brains work. For the last 120 years, we have understood that the universe is based on quantum physics. Whether we are talking about fundamental forces, the nuclear fusion in stars, the brains of biological organisms or even a simple chair, they all function according to the laws of quantum physics.
In recent decades, we have tried to demystify and artificially reproduce the function of the human brain. In recognizing that it works according to the laws of quantum physics opens up an exciting new way of thinking about technological intelligence. Unless we use a computer that inheres the architecture of quantum information, we won’t be able to implement a true AI that isn’t limited to mimicking human intelligence in narrow domains – let alone an AGI. Quantum circuits are explicit procedural representations of neural networks, allowing us to understand how an implicit algorithm is built and why the output is probabilistic
Nevertheless, we will not need to do everything quantumly. Tasks that classical algorithms can already do well, such as pre-processing data, and significant parts of the algorithms we use today, will continue to be done classically. Yet, they are and will be influenced quantumly. Such as in the brain, the firing of neurons won’t only be governed by macroscopic electrochemical interactions. Instead, a quantum-classical transition dynamic will be responsible for the microscopic quantum effects becoming relevant macroscopically on the level of neurons, for instance.
Current AI research is headed towards developing algorithms consisting of both classical and quantum parts to be processed in a hybrid quantum computing (HQC) model – a technological platform equipped with classical HPC, GPU, quantum processing units (QPU), and more specialized hardware.
HQC is not only pertinent for the running of AI algorithms, but it can also help us address intriguing problems we cannot solve purely through classical methods. Among these problems is the search for the interaction of molecules with one another – relevant to the automotive industry’s development of improved electrochemistry in vehicle batteries. In terms of optimization, the applications running in an HQC are also manifold, ranging from optimizing traffic to production processes to the orbital mechanics and trajectories of satellites.