Original article was published by Ramil Jiwani on Artificial Intelligence on Medium
The Love Story of Quantum Computing and Artificial Intelligence | Quantum Artificial Intelligence
This is what happens when Quantum Computing and AI are merged and born out of their marriage is a child named Quantum Artificial Intelligence.
The human species started with fire, sticks and our brain. As we evolved, The fire turned into power plants and the sticks became nuclear weapons. But our brain has achieved the biggest evolution.
For the past 60 years, the power of computers has been growing exponentially. To put this into perspective, the most recent iPhones are 100,000 times more powerful than the Apollo 11 computer.
The Basics of Computers
Computers contain computer chips. These computer chips have many modules. And these modules have logic gates that contain transistors.
Transistors are the most basic data processing units in a computer. You can think of them as a switch that either blocks or opens their way for information. The information that these transistors receive is called bits.
Bits can be set to either 0 or 1 and several bits can be put into combinations to represent complex information.
Transistors are combined to create logic gates. Logic gates perform simple tasks such as sending an output of 1 if all inputs are 1 or sending an output of 0 if otherwise. Logic gates are then combined to create modules. Modules provide functions such as adding or multiplying. Combining modules makes a computer chip. Now that we know how classic computers work, let’s learn how quantum computers work.
What is Quantum Computing
Unlike classic computers, quantum computers process calculations based on the probability of the state of an object rather than the definite state. This means unlike before where we had bits that were either a 0 or a 1, we now have quantum bits or qubits which can be a mixture of possibilities such as the polarisation of a proton. This is called superposition. A simple example of this is a coin spinning in the air before it lands.
Qubits can be entangled with other qubits. This is what makes quantum computing so powerful. Entangled qubits interact instantly. But there is a fault with this. For qubits to be functional, quantum computers need to be cooled to near absolute zero. Even when this does happen, qubits don’t maintain their entangled state for long. This is what makes quantum computing very hard.
Why is Quantum Computing Important
One reason is that as transistors get smaller and smaller, we are getting close to the limit of how small we can go. Once we get to a certain point, phenomena such as quantum tunnelling appear. Quantum tunnelling is when electrons “go through walls” instead of being stopped. This means that transistors cannot work like usual transistors any longer. Quantum computing can solve this.
Another reason is that quantum computers are exponentially more powerful than classical computers. Let’s say we entangled 256 qubits. This would be equivalent to 10⁸⁰ classical bits. Putting this into perspective, there are 10⁷⁸ to 10⁸² atoms in the known, observable universe.
Quantum computing will transform complex tasks that take classic computers years to do into something that can be done in much less time. Some examples are:
Modern computers could take billions of years to find larger and larger prime numbers, however, a quantum computer containing 256 qubits would take only 100 seconds.
Volkswagen is working towards preventing traffic jams through a predictive system that could inform drivers up to 45 minutes in advance.
Quantum computing will also revolutionize the field of encryption. Quantum computers will be able to break encryptions in much less time than classical computers.
IBM’s Deep Blue computer defeated Garry Kasparov, chess champion, in 1997 because it could calculate 200 million potential moves every second. With a quantum computer, these calculations could be one trillion per second.
What is Artificial Intelligence
Before we move on to quantum artificial intelligence, let’s do a quick overview of Artificial Intelligence (A.I).
15 years ago, if you were to mention A.I to someone they would probably think of something like 2001: A Space Odyssey’s HAL or Star Trek’s Data. They would probably see it as something advanced humans from the 22nd-century use. But now almost everyone has heard of it and almost everyone uses it every day. Whether it is email filters, recommendations for your web search, finding you the fastest route to work, music and product recommendations or even mobile banking. AI has become part of our lives. So how does it work?
The easy definition for AI is: Artificial Intelligence is the capability of a machine or computer device to emulate human intelligence (cognitive process), acquire from experiences, adapt to the latest information and operate humans-like-activities.
That definition provides a good but very general understanding of AI because there are many branches of A.I. For today we’ll focus on machine learning.
Machine learning is the study of computer algorithms that improve as more data or experience is provided for the machine to process. Machine learning algorithms improve and become more precise as more data is provided. The process of providing the machine with more data is called training. After training is complete you can provide the model with an input and it will provide you with an output.
For example, if you have trained a predictive model, you can provide the model with data and it will make predictions based on previously inputted data and your inputted data. To improve predictive machine learning algorithms, 4 main techniques are used.
The first one is supervised learning. Supervised learning begins with a set of data and a basic understanding of how that data is classified. This data has a set of features that are labelled and these features will be used by the machine to define the data. An example of this is inputting many pictures of fruits and labelling each of them. Then when you input your own image, the machine can try to figure out which fruit you have inputted.
The next technique is unsupervised learning. Unsupervised learning is used when the data set is not labelled. Unsupervised learning makes inferences from the data set and can be useful to find the hidden patterns.
The third technique is reinforcement learning. Reinforcement learning is different from the other techniques as it doesn’t start with a data set that is provided from before, instead, reinforcement learning models self improve by trial and error. When there is a successful decision, the model is ‘reinforced’ with that decision.
The fourth technique is deep learning. Deep learning is a technique that uses neural networks. Neural networks are layers of cells that work together to produce a result. Deep learning is used when a pattern needs to be found in unstructured data. Neural networks are inspired by the process of identification of the human brain. Some examples of deep learning are face, voice and image recognition.
With basic understandings of both these fields, let’s move to Quantum AI.
What is Quantum Artificial Intelligence and Why Does it Make Artificial Intelligence Better
Did you know that every minute of every day 3.2 billion people post 9,722 pins on Pinterest, 347,222 tweets, 4.2 million Facebook likes. Accounting for all other data, every day, the human population produces 2.5 exabytes of data. 2.5 exabytes is equal to 250,000 Libraries of Congress or the data storage capacity of 5 million laptops. We are getting to the point where our AI algorithms cannot compute any more data on traditional computers. This is where Quantum AI comes in.
Running AI and machine learning algorithms on quantum computers means that we can process data at an exponentially higher rate. Processing more data means that our models are more accurate. If a classic computer would take thousands of days to train an accurate model, a quantum computer would be able to do it in a matter of days. This means that with Quantum A.I, we will be able to get a result from algorithms quicker and with more accuracy, spot patterns in a larger data set and optimize solutions.
Optimization of solutions has the biggest promise. Quantum AI is good for very specific problems that account for a lot of variables such as facial recognition. This may seem like a downfall but this will help to curate the best solution out of all other possible ones. Optimization problems fall under this. This means theoretically we will be able to have the best and most efficient solutions for problems such as drug discovery.
Right now most of our machine learning algorithms are ready and in format for a quantum speedup. With more breakthroughs in quantum computing, we will be able to run existing algorithms and previously complex ones and attempt to solve problems we once thought were too complicated.
What Are Its Possibilities
Now we know what Quantum AI is, what are its real-world applications?
We can start with drug discovery. The estimated cost for discovering a new drug is reported to be US$2.6 billion. And most of this money on the nine out of ten candidate therapies that fail somewhere between phase 1 trials and regulatory approval. Pharmaceuticals are starting to pair up with companies such as IBM to use AI to hunt for therapies. Although this will make the drug discovery process quicker, with Quantum AI, drugs could be discovered in a matter of days.
Let’s move on to reinforcement learning. Reinforcement learning models are great for handling tasks with many variables. The only issue with this is that it takes a lot of time to train these models. For things like making a game bot this doesn’t matter as much but when you want to train models for things like self-driving cars or for trading and finance, the training time increases significantly. With Quantum AI, the capabilities of reinforcement learnings will increase significantly. We will be able to train these models in much less time and with higher accuracy. Raising the capabilities of models used in Natural Language Processing, healthcare, engineering, self-driving cars, marketing and advertising and robotics.
Next up we have handling large amounts of data. It’s a simple equation. The more data, the higher the accuracy. In situations such as healthcare, where accuracy is vital, large amounts of data is necessary. This amount of data takes classical computers months or even years to process and train on. But with Quantum AI we will have accurate models ready within days that are ready to use rather than spending years developing a solution that may not be necessary by the time it’s done.
Quantum AI can also improve supervised learning. With Quantum AI, we can have image recognition models and algorithms that are much more accurate and efficient. This will help fields such as national and personal security with facial recognition models that identify criminals or can detect threats such as gun violence. We can also use improved supervised learning models to detect cancer or other diseases.
Wrapping all of this up, Quantum AI has the potential of improving our AI models by decreasing their training time while also improving their accuracy. And because of this, we will be able to solve problems that might have taken a classic computer thousands of years to do, in exponentially less time. We will be able to produce pharmaceuticals at a higher rate while keeping costs low, improving world safety, advancing modern medicine, improving self-driving cars and much more. While our advancements from fire to powerplants and sticks to bombs might have slowed down, the advancement of our brains to computers has only just begun.