Artificial Intelligence (AI) has become an essential part of our lives, more so than ever before thanks to the vast amounts of data readily available today. Along with AI we often read about machine learning and specifically deep learning. Before delving into deep learning, let us briefly go through what AI is and how machine learning works.
Artificial Intelligence is whenever a machine can execute tasks that would normally need human intelligence and input. One way that AI develops is by machine learning, where software learns from experience after many trials and lots of data benchmarks fed into the self-teaching algorithms.
Deep learning is a subset of machine learning that mimics human neurons and the networks they form in our brains. Deep learning works by feeding massive amounts of relevant data, which the machine then proceeds to use to repeat the desired task as many times as needed with tweaks to optimize the process, just like a human would learn from trial and error and experience.
Deep learning and neural networks are nothing new, and have been around for decades. However, before the Internet became ubiquitous in almost every part of daily life, the necessary amounts of data were nowhere near enough to create the desired outcomes. With social media and an increasingly connected population and environment, big data is plentiful and deep learning and neural networks can finally realize their potential and develop artificial intelligence capable of performing tasks like humans would do, using a similar decision-making logic, with little to no supervision from human handlers.
Perhaps the best way to understand the impact of neural networks in today’s world and tech is to give real-life examples that illustrate their current abilities and how these will develop in the very near future.
1- Autonomous Cars
Perhaps the most talked about and covered technology using AI deep learning, is self-driving cars. Tesla, Google, and even more traditional automakers are all already integrating at least partial autonomy in their cars. As the Internet of Things (connected cities, traffic lights, etc.) becomes more and more available in cities across the world, self-driving cars will be better equipped and “experienced” to more fully take over for human drivers, especially freight trucks that put a massive toll on drivers who work long hours with very little sleep and rest.
A big part of a physician’s work is tapping into his or her bank of knowledge on similar cases to decide the best course of treatment for their patients. Deep learning is a big improvement in the analysis of medical images and data to better diagnose patients in a fraction of the time it might take an overwhelmed medical team and with human error minimized in this process. Another place where AI can help in healthcare is doses of medication and anesthetic, which are among the top reasons patients die, or their situation deteriorates in healthcare facilities.
3- Voice Commands
Our devices seem to be becoming smarter by the week, with services like Apple’s Siri and Google’s Voice Assistant not only being able to understand speech but also beginning to get the context in which words are spoken, even when a sentence includes words from two different languages. This is thanks to neural networks that make speech comprehension similar to how humans understand it, and not just word-for-word translation and comprehension.
4- Image Recognition
If you’re prompted after signing in to identify “traffic lights” or “bicycles” in a photo grid, you’ve experienced first-hand what deep learning has allowed machines to do after endless hours of learning to identify objects, people, and even the weather. Image recognition might seem like a piece of cake, but if we think of a dog, it’s incredibly difficult for a piece of software to recognize that both a Chihuahua and a St Bernard are “dogs”, and that another four-legged mammal in between is not.
5- Image Captioning
This is where image recognition takes a turn to something even more impressive. After identifying objects and persons in a photo, another neural network can analyze what’s actually happening and create an appropriate caption. As early as 2014, algorithms of Recurrent Neural Networks (RNNs) have been showing impressive results and writing stunningly accurate captions.
6- Colorization of Old Footage, Adding Sound to Silent Movies
Peter Jackson’s upcoming documentary “They Shall Not Grow Old” about World War I, with color and sound added to grainy black and white footage from the early part of the 20th Century, will become a lot more common and less time-consuming thanks to deep learning. AI can analyze different sets of data, such as the color of each army’s uniform, and color them automatically accordingly.
The cherry on top is when the AI can recognize what’s in the footage, and link it to sound clips or create sounds that correspond to things like leaves ruffling, guns firing or children giggling that were once silent and without color.
This is perhaps the deep learning innovation that most users detest: personalized advertising. AI neural networks can tap into a wide array of data about individual users, such as location, income level, education completed, likes and dislikes, etc. and show them the most relevant ads, learning from the impact of each tweaked campaign and perfecting them over time and with more data. This maximizes return on investment, by targeting potential customers in real-time and avoiding users who the brand will probably not convert with a particular ad campaign.
8- Predicting Earthquakes
Apart from the work in healthcare diagnostics and treatments, there are other ways that deep learning can save lives and avert tragedies. Harvard researchers have been able to cut down computer calculations on potential earthquakes by a whopping 50,000%. This speed in prediction will undoubtedly help innumerable lives who will now have enough of a head start to get to a safe place and avoid being trapped or crushed under rubble.
Finally, even though neural networks are not something new, their effectiveness has only now become workable across different industries and technologies. Big data is how machine learning thrives, and deep learning makes it more and more similar to how our own brains think, with less need for human oversight on the decision-making process.
Neural networks are in your smartphones, your cars, your hospitals, your bank, and even institutes trying to predict earthquakes. In the coming months and years, deepa learning will redefine what we think of as AI. It’ll make our Siris and Teslas even smarter, more human-like in their thought process, only much faster and better informed.