Quantum computing and artificial intelligence (AI) have arrived at the tech scene at almost the same time and both are promising to be the next big frontier. However, it is becoming clear to technology companies, tech enthusiasts, and developers that these two technologies should co-exist and are capable of complementing and supplementing each other.
The biggest caveat with AI is its need for massive compute performance. To run thousands of iterations of an algorithm to reach an inference and that too instantaneously requires enormous power. Andrew Ng, one of the profound scholars in the field of AI and machine learning, has been a vocal advocate of high-performance computers from the start.
While high-performance computers are made accessible thanks to performance chips being produced by the likes of NVIDIA, Intel, AMD, and others and data centres from companies like Amazon Web Services, Microsoft Azure, Google Cloud, and others, there is still a search ongoing for alternatives. That alternative could prove to be quantum computing.
What is Quantum Computing?
Quantum computing can be described as the exact opposite of classical computing. Classical computing stores data in binaries and at any given time, there can be only one instance. Quantum computing changes that paradigm altogether with its ability to hold many different possible outcomes in the “quantum state”.
This ability to store different possible outcomes is precisely what AI is used for right now. Imagine a healthcare company using AI to predict different outcomes for a research medicine or a trial of generic medicine. With current computing powers, this healthcare company will need to run each instance in a separate compute node.
This process takes a lot of computing power and associated energy requirements need to be factored in as well. However, with a quantum computer, a single solution could deliver multiple possible outcomes without any hassle. This is mainly possible because of how a quantum computer stores its data.
Instead of Boolean linear algebra functions of 1s and Os, quantum computing allows the use of linear algebra in the form of quantum bits or qubits. These qubits are composed of numbers, vectors, and matrices interacting in quantum states, including superposition, entanglement, and interference.
Quantum Computing and AI
As mentioned earlier, an AI algorithm is currently run on compute nodes that might include CPUs and GPUs, which are dependent on classical binary computing. As a result, they come with the aforementioned limitations. Quantum computers thus offer the possibility of a quantum leap in performance and bring capabilities enabling a range of use cases.
AI could prove to be one of the biggest use cases and that use case could come to be defined as quantum AI. The central element to this ability is the capability of a quantum computer to store multiple states at once, which is precisely what those using AI programs want when looking for an inference or result.
One of the most commonly cited abilities of a quantum computer is to use brute-force methods to guess the passcode used to encrypt a piece of data using a 256-bit algorithm. The rise and accessibility to quantum computers could mean data encryption being done with AES-256, which is considered to be much more secure and cannot be cracked using a brute-force attack.
Another example that has been mentioned is the use of a quantum computer to figure out the most efficient path to reach a particular geographical location, one of the most compute-heavy applications in the world. This means the likes of UPS and FedEx will switch to quantum computers to compute the route for their delivery person, which could lead to savings in terms of time as well as fuel.
Let’s switch our attention to AI and machine learning and how quantum computers might play a huge role. The area where quantum computing could be applied almost immediately is in the field of deep learning. Deep learning can be dubbed as the latest iteration of machine learning requiring a large amount of computing performance.
OpenAI’s GPT-3, a new transformer model with 175 billion parameters, takes months to train on a classical computer featuring CPU and GPUs. However, it is abundantly clear that GPT-3 is just the start and has paved the way for large transformer models featuring a number of billions or even trillions of parameters.
In order to train a transformer model with trillions of parameters, it will take several months, which won’t be ideal. Quantum computers could reduce this amount of time required for training massively and thus allow for new progress with artificial intelligence and machine learning.
Google launched TensorFlow Quantum in March 2020 as an effort to bring the TensorFlow machine learning development library to the world of quantum computers. Developers will be able to develop quantum neural network models running on quantum computers with Google’s TensorFlow Quantum. However, we need to acknowledge that running AI applications on quantum computers is still in its earliest stages.
Right now, Google has shown how AI development can be brought to quantum computing. NASA has seen the potential and has begun working with Google. IBM is another big player in the field of quantum computing with IBM Research demonstrating “mathematical proof” of a quantum advantage for quantum machine learning.