A new approach to integrating sensory input and responsive action– an intriguing paper: “Learning Sensorimotor Control with Neuromorphic Sensors: Toward Hyperdimensional Active Perception” by Anton Mitrokhin and Peter Sutor (abstract link)
Category: AI
Cogitai’s learning agent lives in the cloud, but it’s not called Skynet.
Continual Learning AI startup Cogitai teams up with Sony
Cogitai, Inc. is dedicated to building artificial intelligences (AIs) that learn continually from interactive experience with the real world.
Sony joins forces with Cogitai
A complete discussion of continual learning can be found in Mark Ring’s dissertation, Continual Learning in Reinforcement Environments.
I guess it is time to sign up for a machine learning course. This one by Littman (Cogitai) Isbell (Cogitai) and Kolhe looks good.
UPDATE (6/6/2016):
In case Cogitai and other AI agent builders are becoming _too_ successful, the human race are investigating the possibility of an AI agent “kill switch”.
Interestingly, while the Conclusion states:
… We have shown that some algorithms like Q-learning are
already safely interruptible, and some others like Sarsa are
not, off-the-shelf, but can easily be modified to have this
property. We have also shown that even an ideal agents
that tends to the optimal behaviour in any (deterministic)
computable environment can be made safely interruptible.
However, we may not be safe from Skynet Apocalypse, as the Conclusion also suggests:
… However, it is unclear if all algorithms can be easily made
safely interruptible …