By Jennifer Yang
Delving Into The Past![]()
The origins of AI began, where most things seem to, in Ancient Greece. One myth mentions an intelligent robot Talos, a creation of Hephaestus—the god of metalworking and forges—designed to protect Europa of Crete. The Greeks themselves dabbled in the making of robots. They can claim to have created the first working robot in history. Made in the shape of a life-size woman, this robot was capable of pouring wine and water. There are also records from various peoples—such as the Chinese, Arab, and English—of creating automatons and contemplating the imitation of human thinking.
Both literature and the film industry have also spent a lot of time toying with the idea of AI. There was Mary Shelley’s Frankenstein, a story delving into the creation of a being capable of feeling and thinking subjectively. Star Wars is also no stranger to AI, with its beloved robot characters R2-D2, C-3PO, and now BB-8, who all display intelligence, emotion, and the capability to take action based on the previous two traits. Familiar with I, Robot? The whole book, and its movie adaptation, centers around the idea of robots who have become intelligent enough to take actions based on what they believe is the best course of action. To humans, this sort of thinking occurs without much difficulty, but where is the line drawn? I’ll pull an example from the work. There are two people drowning in a large river: a middle-aged man and a young girl. Only one can be rescued. The typical reaction of a human would be to save the child first, but the robot in the movie saves the middle-aged man, based on the calculation that the man had a better chance of survival. The worlds that authors and movie writers have been dreaming about for so long may come to life sooner than you might think, if the progression of AI is anything to judge by. The idea of artificial intelligence (AI) first gathered steam in the 1950s, bringing with it the notion that machines might be capable of thinking (Turing’s Test) and therefore capable of manipulating symbols the same way they do numbers. 60 years later, the accomplishments of game AIs, such as the groundbreaking chess program and ELIZA, the first chatterbot, now look childishly simple in comparison to recent events. Back To The Future
In this era, with the availability of big data and faster processors, AI has been able to take huge leaps—quite literally.
Take Google’s DeepMind. One of DeepMind’s recent accomplishments has involved the use of reinforcement learning (RL or machine learning done so that an agent feels motivated to choose a path that will lead to a reward) to train agents with varying body shapes to navigate through obstacles such as uneven ground, hurdles, and walls. To clarify, the programmers’ only involvement was providing the agents with sensors for interacting with the environment and rewarding them if they did not “fail”, or fall over. To us, the agent’s method of movement (displayed at 1:27 of the video embedded above) seems amusing at most. As silly as it looks though, Deepmind’s agents are learning how to move efficiently through trial and error. Their strange, whimsical movements have gotten them farther than agents of other simulations, who often cannot adapt when introduced to strange environments. A notable phenomenon is that an increase of number of tasks an agent went through caused an improvement in the agent’s movement and the agent’s recollection of the way it moved past obstacles. RL usually doesn’t influence an agent deeply if the environment changes by too much, but these AI agents have shown that complexity is possible. This is a small taste of what DeepMind has accomplished: the company has used the fruits of their research to beat the world’s best player at Go. Go is an ancient Chinese strategy game played between 2 players with the goal of capturing more area than your opponent. Despite having deceptively simple rules and bearing similarities to chess, Go is much more complicated. Louis Victor Allis calculated that there are roughly 10^50 possible positions in chess. In Go, there are 2 x 10^172. In 2016, DeepMind’s AlphaGo faced the world’s best player with these odds and won. Only a year later, DeepMind released AlphaGo Zero, which quickly left its predecessor in the dust within 3 days and became proficient enough to be considered the best Go player in 40 days. Games aside, DeepMind has also been working on refining Google’s energy efficiency and improving medical treatment and care in hospitals. Providing a world with all sorts of services can be a daunting task, but DeepMind’s AI has been applied to reduce Google’s energy bill by 40%. By analyzing at least one hundred variables, AI was able to maximize the efficiency of the power conversion and reduce water usage. Google isn’t the only benefactor from DeepMind. Patients in the UK can receive high quality care under the National Health Service, NHS, but many still suffer and possibly die due to lack of quick diagnosis. DeepMind AI has been able to prevent such tragedy through an application called Streams. Through applying deep learning, the AI can track a patient’s status and notify hospital staff immediately, through the app, if there are abnormalities in the patient’s condition. Ever seeking to improve itself, DeepMind is also attempting to recognize common symptoms of certain conditions, such as kidney disease, and bring them to a doctor’s attention. Citations
“AITopics.” A Brief History of AI, AAAI, aitopics.org/misc/brief-history.
“AlphaGo Zero: Learning from Scratch.” DeepMind, deepmind.com/blog/alphago-zero-learning-scratch/. “Go and Mathematics.” Wikipedia, Wikimedia Foundation, 2 Nov. 2017, en.wikipedia.org/wiki/Go_and_mathematics. NatureVideoChannel. “The Computer That Mastered Go.” YouTube, YouTube, 27 Jan. 2016, www.youtube.com/watch?v=g-dKXOlsf98. Oppy, Graham, and David Dowe. “The Turing Test.” Stanford Encyclopedia of Philosophy, Stanford University, 8 Feb. 2016, plato.stanford.edu/entries/turing-test/. Vincent, James. “DeepMind's AI Is Teaching Itself Parkour, and the Results Are Adorable.” The Verge, The Verge, 10 July 2017, www.theverge.com/tldr/2017/7/10/15946542/deepmind-parkour-agent-reinforcement-learning. ALL IMAGES AND VIDEOS BELONG TO THEIR RESPECTIVE OWNERS. SUGGESTED READINGS
To delve deeper into understanding AI, interesting ponderings, or if in search for a book about robots, look no further!
Asimov, Isaac. I, Robot. HarperCollins, 1996. Gordon, Charlotte. “What A.I. Researchers Can Learn From Frankenstein.” Slate Magazine, 23 Jan. 2017, www.slate.com/articles/technology/future_tense/2017/01/what_artificial_intelligence_researchers_can_learn_from_frankenstein.html. Kurzweil, Raymond. The Age of Spiritual Machines: When Computers Exceed Human Intelligence. Penguin Books, 2000. Piercy, Marge. He, She and It. Del Rey, 2016. For more about DeepMind, read the following articles about their latest project, DeepMind Ethics and Society, which is aimed towards considering the ethical implications of their AI technology on the world. Nelson, Daniel. “Google's DeepMind Launches AI Ethics Research Unit.” Science Trends, 20 Oct. 2017, sciencetrends.com/googles-deepmind-launches-ai-ethics-research-unit/. Temperton, James. “DeepMind's New AI Ethics Unit Is the Company's next Big Move.” WIRED, WIRED UK, 4 Oct. 2017, www.wired.co.uk/article/deepmind-ethics-and-society-artificial-intelligence.
0 Comments
Leave a Reply. |
The PrimerStriving to deepen the understanding of STEM-related topics. Archives
June 2018
Categories
All
|