The Edge of Gamification : 無料・フリー素材/写真
The Edge of Gamification / jurvetson
| ライセンス | クリエイティブ・コモンズ 表示 2.1 |
|---|---|
| 説明 | Which products or business processes can be expressed as a game, and thus, might be ripe for the new wave of automation?About a year ago, I wrote about the “edge of automation” (MIT Tech Review), noting that many of the new jobs in the new economy (like Uber drivers and Mechanical Turkers) are cognition tasks just a bit beyond the realm of automation, and thus, are ever so ephemeral against the march of Moore’s Law. And since they are recently architected jobs, they are surrounded by modern digital interfaces, so much so that if your Uber driver were to be replaced by a mute body model from WestWorld, you might not even notice. But the Toronto Deep Learning conference helped refine that edge, and suggest where we may likely find it — the edge of gamification. Deep Learning applies particularly well to games, literally as the Deep Mind team demonstrated before the Google acquisition and through homologous problems as they have been demonstrating afterward. The training of Alpha Go was particularly prophetic: a learning algorithm can be more rapidly trained if the human trainer is replaced by another learning algorithm.We have also seen this in our startup portfolio in the field of autonomous vehicles. You can add billions of miles of driving experience, and hone the tricky corner cases, by building a Matrix for the vehicles’ sensors, feeding the computer the same digital streams that the camera and radar and LIDAR see when navigating real roads, but synthesized from the point cloud of a simulation not dissimilar to GTA. The big auto companies are doing the same to try to catch up with the leaders.Deep Reinforcement Learning is one of the promising subsets of machine intelligence that has delivered many of the recent gaming wins. Karpathy gives a good overview of RL and refines the type of gameplay that currently lends itself to automation: “there are many games where Policy Gradients would quite easily defeat a human, in particular, anything with frequent reward signals that requires precise play, fast reflexes, and not too much long-term planning, as these short-term correlations between rewards and actions can be easily noticed by the approach, and the execution meticulously perfected by the policy.” But that is for now; as memory structures roll into these models, the human knack for rapid abstract model building may also tip over the edge of gamification. So what’s next? > Shall we play a game? |
| 撮影日 | 2016-11-22 11:26:57 |
| 撮影者 | jurvetson , Los Altos, USA |
| タグ | |
| 撮影地 |