The Rise of AI
Charting the Path of Artificial Intelligence
In recent years, artificial intelligence has become an increasingly common feature in tech products for a wide range of industries. While previously fodder for science fiction, AI is now a realistic and effective addition to the workflows of countless professionals. Let’s examine how AI reached its current state and why it has become so valuable.
The Evolution of AI
Over the past two decades, AI seemed to be making a lot of progress. In 1997, the IBM computer DeepBlue bested chess grandmaster Garry Kasparov. In 2011, IBM’s Watson won an episode of “Jeopardy!” against two human former champions. Both chess and trivia are closed-loop problems, so these systems could build fairly robust decision trees search for the solution and quickly get the answer, assuming they had enough indexes and subsequent computing power to draw from. While their successes made headlines, AI systems and business applications were inherently limited.
The problem was Polyani’s Paradox, which states that human knowledge and capability rely on understanding that is beyond our conscious awareness. We simply can’t tell a program everything we know about complex subjects, so many commentators thought that it would ultimately prove impossible for machines to rival human understanding.
That all changed in 2016, when Google’s AlphaGo program became the first AI to defeat a professional go player at the highest level without handicaps. Programmers broke through Polyani’s Paradox by removing humans and their biases from the process in favor of having AlphaGo run simulations against itself. It learned through learning, and in the process absorbed all those things about go that a human can’t describe. In effect, it was the first machine to understand common sense.
The Paradigm Shift
So, what made AlphaGo possible? It was the convergence of the three fundamental building blocks that ended the “AI winter” and allowed for us to forge a new path:
- Moore’s Law and Dennard’s Scaling
- Algorithm Evolution
- Data
Moore’s Law and Dennard’s Scaling state that processing power doubles and costs halve every two years, while increased computing power can be maintained at a near-equal voltage charge. This changed in 2005, when the process started slowing down because algorithms were requesting more computing power at an increasing rate.
What was pushing the algorithms? Data.
The rapid increase in data started in the mid-2000s, giving rise to the term “Big Data.” We are currently close to 3 zettabytes of information and will be north of 25 zettabytes in five to six years. More data requires more processing, which requires better algorithms, and better algorithms require more data, so we get a circle that is constantly feeding off of itself to create better and better AI.
Big Data Drives AI Innovation
The AI feedback loop means that advances in data both drives and requires innovation in the other fields. To ensure training times stay reasonable, processing solutions have evolved to support a more advanced use cases. Most use cases now push the limits of your traditional CPU. Because of this, graphical processing units (GPUs) were introduced to support better distributed processing needs. The latest, however, is tensor processing units (TPUs). TPUs provide the most powerful solution to date for the right use cases.
In turn, these boosts in computing power and data have allowed us to increase the power of our algorithms. With so much to draw from, we can now develop things like Generative Adversarial Networks (GANs). The power of GANs is that they create unique worlds that are so similar to our ours, it’s tough to tell real from imitation. GANs have the power to make music, paint pictures and even create videos and photos. Unique worlds can mean a personalized virtual environment for potential home shoppers that’s suited just for them.
As the cycle spins, we can expect AI to evolve, with process improvements and automation serving as core strategies. Digital developers at CoreLogic are keeping up with the latest innovations in machine learning and developing the next generation of digital solutions to transform the housing ecosystem. Expect to see big changes in what’s possible over the coming years.
