geeky NEWS: Navigating the New Age of Cutting-Edge Technology in AI, Robotics, Space, and the latest tech Gadgets
As a passionate tech blogger and vlogger, I specialize in four exciting areas: AI, robotics, space, and the latest gadgets. Drawing on my extensive experience working at tech giants like Google and Qualcomm, I bring a unique perspective to my coverage. My portfolio combines critical analysis and infectious enthusiasm to keep tech enthusiasts informed and excited about the future of technology innovation.
A passionate team of Silicon Valley engineers at OpenAI is in a race to develop an artificial intelligence bot capable of defeating the world champions of one of the most lucrative esports – Dota 2. And they’ve got one year to do it. The outcome could alter the way we think about the frontiers of machine learning and of humanity’s relationship to AI.
For more than 80 years humans and machines have competed in chess, Go, poker, Jeopardy! The latest battlefields are hugely popular multiplayer online games – like Dota 2 – that are worldwide sensations. In tihs in-person screening of Artificial Gamer, it talked about OpenAI’s efforts to build the most powerful esports bot. It begins at the 2017 annual eSports world championship, also known as “The International.” OpenAI announces that an AI agent will play against some of the world’s best DOTA 2 players. Dota 2 is a multilayered experience involving bluffing, coordinated five-player strategy, and epic psych-outs — a level of gaming considered too sophisticated for the traditional reach of computers. As OpenAI’s software not only successfully matches the world’s top DOTA 2 players, but wins by a landslide in one-on-one competition, the creators behind the AI seek to step up their game and create a software that can successfully compete and win against five players in a five-on-five match.
The documentary went on interviewing key players in the field, displaying fun and exciting sequences of gameplay that effectively change the pace of the film, and delivering a solid story arc balancing the viewpoints of the OpenAI creators as well as the DOTA 2 community. I enjoyed learning how the AI developers push the boundaries of machine learning, extracting and repurposing algorithmic learning to take advantage of processing power that can cram 180 years’ worth of Dota play every day, learning via self-play. It trains using a scaled-up version of Proximal Policy Optimization running on 256 GPUs and 128,000 CPU cores — a larger-scale version of the system built to play the much-simpler solo variant of the game.
Using a separate LSTM for each hero and no human data, it learns recognizable strategies. This indicates that reinforcement learning can yield long-term planning with large but achievable scale — without fundamental advances, contrary to our own expectations upon starting the project. OpenAI Five is given access to the same information as humans, but instantly sees data like positions, healths, and item inventories that humans have to check manually. Our method isn’t fundamentally tied to observing state, but just rendering pixels from the game would require thousands of GPUs. OpenAI Five averages around 150-170 actions per minute (and has a theoretical maximum of 450 due to observing every 4th frame). Frame-perfect timing, while possible for skilled players, is trivial for OpenAI Five. OpenAI Five has an average reaction time of 80ms, which is faster than humans. More information can be found here.
Following the screening, there is a panel discussion and Q&A featuring Open AI team’s David Farhi and Susan Zhang. Some backgroudn about David Farhi who is a research lead at OpenAI, where he led the DotA 2 project during its seminal achievement of superhuman performance in a top competitive e-sport. He studies reinforcement learning in complex environments and is especially interested in environments with unbounded emergent behavioral complexity. He completed his PhD at Harvard in theoretical physics and undergraduate degree at MIT.
In the front lobby area of the Computer History Museum, there is also a Waymo (part of Google) self drive car, called Firefly. Google initially designed the Firefly back in 2013. From the start, the car was intended to be an experiment that would allow engineers to explore different ideas about how autonomous vehicles should work or be configured, they said. Although it was seen all over the place in cities like Mountain View while Google was testing it, the Firefly was never intended to be a production vehicle. Along the way, the Firefly racked up some impressive achievements for the development of autonomous vehicles, including million of miles driven and the first completely autonomous trip. Visitors are allowed to get in the Waymo and try out the ride the auto seldrive car. Pretty cool!
