July 28, 2018

Artificial Intelligence has received unprecedented worldwide attention over the past several years. However retrospect, which AI breakthroughs were truly revolutionary? Here's a list of our top 10 picks (in no particular order):

• 1) Deep Blue

  'Alas, the legendary chess game, IBM's Deep Blue, vs. world champion (human) Garry Kasparov. This was the match up that proved machines were here to stay. In previous attempts, Artificial Intelligence machines simply did not stand a chance against chess masters. Kasparov fortunately won the 1989 match, but did not stand a chance the following years. Deep Blue, believed to have an ability to play chess 5 moves ahead of it's opponent, seemed to calculate 20 moves ahead, (compared to 10-15 for Kasparov), and easily won the remaining matches. Kasparov was less than thrilled by the results.

• 2) ELIZA

  The original chatbot, ELIZA, became the first form of AI that could truly converse with humans on any topic. It was developed in 1965 by Joseph Weizenbaum, a professor at MIT, with the intentions of emulating a psychotherapist.

• 3) Siri, Cortana, Jarvis, Duplex

  Virtual assistants were the next generation evolution of chatbots like ELIZA. These "super chatbots" were made to manage your schedule, book appointments, and even tell jokes. Although conversational software did exist for many years prior to Siri's release, the quality of Siri and it's competitors far exceeded any of it's predecessors. Instead of having a million hard-coded responses, these newer assistants had the ability to decide what the best answer was based on your questions, using machine learning techniques.

• 4) Computer vision

  In 2012, a Neural Network developed by Google managed to correctly differentiate between cats and dogs by watching YouTube videos. Over the last 5 years, AI Computer Vision has evolved tremendously, and has made it possible to classify thousands of objects using the computing power of an average smartphone.

• 5) Jeopardy!

  In 2011, IBM’s Watson defeated Jeopardy! champions Rutter and Jennings. This impressive feat demonstrated how machines have indeed reached a human level understanding of Natural Language.

• 6) Back-propagation & Neural Networks

  The driving force behind modern day Neural Networks; Back-propagation was originally developed in the mid-1980's, but until the early 2000's, they were shunned by the AI community, as sub-par and inefficient. Advances in computational power, including the widespread use of computer graphics cards, or GPUs, turned the tables and gave Neural Nets the leading edge.

• 7) Prolog

  A prolog is just a computer programming language that enables logic. Suited for solving problems like: Condition 1: “If it is raining, take an umbrella”, Condition 2: “It is raining” Result: “Take an umbrella” Simple enough.

• 8) Reinforcement learning

  Reinforcement learning is a modern day technique that enables a computer to learn by doing. Being a top notch Mario World player (MarI/O), or a stickman that can learn to walk and run like a human (DeepMind), takes practice and reinforcement. The concept sounds intuitive, but the execution can prove to be rather difficult.

• 9) Self-driving cars

  A car that can drive and park by itself, straight out of a 70's fiction movie, is now a commonplace occurrence. Using radars, laser light, GPS, cameras for computer vision, these robotic cars can detect and recognize anything in their path, be it a road sign, pedestrian, or your not-so-friendly neighbor ready to be pulverized! Jokes aside, autonomous cars promise increased safety and security, mobility and reduced crime rates. Get ready to work longer hours. Commute times will not exist.

• 10) Alpha Go / Zero

  AlphaGo was a computer program designed by Google’s DeepMind team for playing the  popular game "Go". Long story short, AlphaGo demolished Lee Sedol, (then the Go world champion). It's powered by a deep neural network that uses an algorithm called Monte Carlo tree search. "Go" is much more complex than chess, and brute force search algorithms weren't enough to crack the game. AlphaGo’s newer version, AlphaGo Zero, achieved even better results, and more importantly the results were achieved without any human data. It learned to play the game by constantly playing against itself. If it were to give chess a go (no pun intended), it would probably learn to play in one day!