Archive

Archive for November, 2016

Weekly QuEST Discussion Topics and News, 4 Nov

November 3, 2016 Leave a comment

QuEST 4 Nov 2016

Next we want to continue discussing our main topic:  “QuEST for Consciousness”:  the purpose of the discussions over the last several weeks have been to generate a single slide and an elevator speech that captures the ‘what’ part of a pitch for initiating an effort to build a conscious computer – this week we will continue to discuss the ‘what’ – specifically this week we want to establish types of qualia and computing with qualia

All ‘conscious’ mental states involve qualia – we will argue that generating and computing with qualia will be the key break through in making emotionally intelligent machines (machines that can ‘feel’ and distinguish between a range of ‘feelings’ / ‘conscious experiences’ in their representation of the environment and themselves {part of the representation of the environment can include a representation of the ‘feelings/experiences’ of other agents}) –

This will be done in a dual process framework that integrates big-data approaches to capturing data level experiences and efficient representations to facilitate quick reflexive responses to stimuli that are close enough to prior experiences at the data level – and secondly by generating a qualia based representation that models our Theory of Consciousness (tenets) –

We argued last week that this vocabulary of qualia could be extracted from the ‘big-data’ analysis done in the subconscious processing – this week we need to address the types of concepts we need to represent in the qualia vocabulary –

To have the types of qualia discussion we need to remind everyone how we use that term – definitions are pretty worthless – but through providing examples we can come to an understanding of how we use that term – so discussions that focus on ‘what it is like to have a given experience’ can provide a framework to move forward with –

Then we can move to our Theory of consciousness – the 3 big tenets –  situation based / simulation / structurally coherent and the sub tenets – a discussion of what we mean by these terms will allow us to move forward to a discussion of how do current major thrusts in machine learning can be used and/or modified to engineer these ‘conscious computers’ –

To have the discussion on modern machine learning we will use three recent great examples – the first is a Darpa perspective on AI provided to us by the Director of DARPA I2O – John Launchbury – I had the pleasure of interacting with him in association with us both providing talks at a recent Jason’s meeting – John speaks to the waves of machine learning

The second perspective on modern machine learning is from our DARPA colleague Tran T. and is also a historical view of AI

The last perspective will follow the book the master algorithm by Pedro Domingos – in that work he parses the space into the five tribes of machine learning –

The goal of these discussions on machine learning will be to discuss where those representations are either consistent with our types of qualia tenets or they can be manipulated to a representation that is consistent AND then lead to the discussion of how the resulting representation can be used for inference – (of course there is the middle step which is how do they each currently generate a model for a given problem space)

  • In reality, three orthogonal problems arise:

–     • choosing a representation language,

–     • encoding a model in that language,

–     • performing inference on the model.

 

news-summary-34

Advertisements
Categories: Uncategorized