Artificial Intelligence (AI) surrounds us, and it has become an inherent part of our everyday lives whether we are aware or not. Digital AI machines, also known as bots, are the mechanism for almost everything we experience daily, ranging from smartphone to TV apps. From Netflix’s “Recommendation Engine” tool to the Instagram “explore” portion of the app, all the recommendations and pictures are brought to life by AI bots. AI enables social media giants such as Facebook and Twitter to target ads for their clients with extreme precision. AI also exists in our homes and can be in the form of smart devices that react to a change in temperature or lighting. The shift towards autonomous vehicles powered by AI has become a reality and is no longer something you see solely in sci-fi movies. We live in exciting times, at the forefront of a technological ‘AI revolution’. Now you are probably wondering, what is AI? Is it a robot that walks and talks and has feelings? No — at least, not yet.
Defining Artificial Intelligence
Artificial Intelligence exists at the convergence of computers, math, and data, which aims to mimic the human decision-making process.
Think of AI as an umbrella, under this umbrella, there are two key components: Machine Learning and Deep Learning. Machine learning is the subset of AI which uses statistical learning algorithms to build smart systems. Deep learning, on the other hand, is the subset of AI that uses artificial neural networks to mimic how a human brain filters information.
The design of these two learning systems is to discover patterns using historical data and apply these learned patterns to new, unseen data. AI systems learn by minimizing errors, similar to humans. As previously said, AI is an imperfect replication of our decision-making process.
Think back to when you were young, and you touched a hot stove, you burned your hand and know not to touch hot things again because you now know it will hurt. AI learns similarly, initially making lots of errors and then eventually learning from them. The bots need to make these errors as a part of the learning process, as do humans. AI systems are constructed to minimize the mathematical distance between the error and the correct output. This is called training.
The most interesting way to assess the intelligence of modern AI is not mathematically but philosophically. Before a child can walk independently, those infants rely on adults for years. In a matter of minutes or hours, we can train an AI system from scratch to do a task with optimal accuracy. Similarly to training humans, we can show a child three photographs, one of a dog, one of a cat, and one of a horse, and these three unique animals are now indelibly imprinted on the child’s mind. It takes tens of thousands of images to train an AI model to do the same task properly. Beyond that, the child can simultaneously do multiple things. This includes classifying pictures of animals, as well as talking, walking, etc. The zenith of modern AI systems is still limited to singular tasks and cannot be trained to perform cross-functional tasks.
The AI Revolution Today
Human neural networks have about 86 billion neurons and 150 trillion synapses, and an average number of 1,744 synapses per neuron. Contrastly, one of the world’s largest artificial neural networks has a capacity of 175 billion machine learning parameters or artificial “synapses.” 150 trillion/150 billion = 1,000. In other words, the most powerful AIs would have to scale by a factor of one thousand for their connections to be on par with the brain.
However, it is not the size capacity of AI models that limits their abilities. It is our understanding of the brain which limits the scope of modern machine learning systems. Humans don’t possess a deep enough understanding of the brain to replicate it dependably and completely.
One of the most renowned computer scientists of our generation and often revered as the ‘father’ of deep learning, Geoffrey Hinton, has many quotes with this same sentiment. “The brain has about ten thousand parameters for every second of experience. We do not have much experience about how systems like that work or how to make them so good at finding structure in data.”
For Artificial Intelligence systems to scale to human intelligence, the entire paradigm may have to be changed, which is hopefully something we will witness in our lifetimes. Hinton also said, “[talking about AI] …we’re very used to this in banking, for example. ATM machines are better than tellers if you want a simple transaction. They’re faster, they’re less trouble, they’re more reliable, so they put tellers out of work”, and this is a great analogy for where the AI revolution is today.