No QuEST Meeting this week

QuEST is canceled for 1 February 2019.  Next QuEST meeting is 8 February 2019.

Weekly QuEST Discussion Topics, 25 Jan

QuEST 25 Jan 2019

 

QuEST this week will be focused on conversational bots – social bots – specifically an approach by colleagues at Facebook and Stanford:

 

Learning from Dialogue after Deployment: Feed Yourself, Chatbot!

 

arXiv:1901.05415v1 [cs.CL] 16 Jan 2019

 

•       The majority of conversations a dialogue agent sees over its lifetime occur after it has already been trained and deployed, leaving a vast store of potential training signal untapped. In this work, we propose the self-feeding chatbot, a dialogue agent with the ability to extract new training examples from the conversations it participates in.

•       As our agent engages in conversation, it also estimates user satisfaction in its responses. When the conversation appears to be going well, the user’s responses become new training examples to imitate.

•        When the agent believes it has made a mistake, it asks for feedback; learning to predict the feedback that will be given improves the chatbot’s dialogue abilities further.

•       On the PERSONACHAT chit-chat dataset with over 131k training examples, we find that learning from dialogue with a self-feeding chatbot significantly improves performance, regardless of the amount of traditional supervision.

Weekly QuEST Discussion Topics, 18 Jan

QuEST 18 Jan 2019

We want to continue this week with our traditional ‘Kabrisky Memorial Lecture’ –to start from scratch and baseline our use of terms and our goals for the group will be our focus.

n  QuEST is an innovative analytical and software development approach to improve human-machine team decision quality over a wide range of stimuli (handling unexpected queries) by providing computer-based decision aids that are engineered to provide both intuitive reasoning and “conscious” context sensitive thinking.

n  QuEST provides a mathematical framework to understand what can be known by a group of people and their computer-based decision aids about situations to facilitate prediction of when more people (different training) or computer aids are necessary to make a particular decision.

Last week we started with an update from our colleague George on concerns in assuming that modern reinforcement learning can solve some important real world problems versus in ‘game space’.

For the Kabrisky lecture we got through some definitions and concepts associated with the ‘subconscious’ ~ ‘Sys1’ system and were starting to consider consciousness.  We will continue from there this week.  The goal this week is to get to a common understanding of the guiding principles that will guide the generation of the ‘conscious’ representation.  Specifically we would like to have a convergence on our presumption that the goal of the conscious representation is to be stable, consistent and useful versus a representation of ‘reality’.  In the words of Hoffman fitness versus truth.

Perhaps the best way to capture why we think the way we do about consciousness is a discussion we had with respect to work by Hoffman:

• “ It should not surprise us that our engineered solutions to focused problems – like object ID from imagery might outperform humans – the idea that we can design solutions with a specific very tailored fitness function – probability of detection … – should if we do it right and provide all the data necessary win over a system that was designed with a different fitness function – survival and reproduction
•  It could be that the fitness function that humanity solves required consciousness?
•  I still like our distinction between agents and conscious agents – Hoffman attempts to make everything conscious –
•  but appreciate the idea of making agents versus the world out there – outside an agent are other agents – a rock or a table could be considered an agent – that is not a position I’ve taken before – its sensors are basically the physics that capture the way it can interact with the world – its sensors are its atoms – its internal representation are its physical makeup – and it effects the world as it interacts with that world – but there are agents that have what we might term more cognitive representations – some of those are in fact conscious –
• Do we need to change our definition of awareness as a result of this change – it used to be defined as the mutual information between an agent’s representation and reality – but reality is nothing more than a set of agents – I think we are ok – just the mutual information is the mutual information in the agent’s representation and the reality of the agents in the environment that it is representing – a rock has very low awareness – just those agents like the ground it is in contact with or being acted upon by
•  Now back to consciousness – some agents have as part of their representation a confabulated (simulation), situated and structurally coherent representation – those we will say are conscious / QuEST agents – now in the case of Hoffman the fact that such a representation can handle uncertainty / aspects can be combined and lastly have this Markovian property can be included in describing agents and doesn’t preclude our definition – but ours doesn’t require those aspects for consciousness –
•  Now back to the big question – what can we take that is useful out of this discussion – I will keep thinking about this as I’m on the road – but basically I think only the extension of our position to allow a framework that is all agents – thus a unified theory – I do not see anything here that says we have a new way / insight to the creation of QuEST agents

Matt Kabrisky

Dr. Matthew Kabrisky was an Air Force pioneer and innovator in technological advancements.  From Air Force aviator in the 1950s to professor, mentor, and researcher until the 1990s, Dr. Kabrisky made important discoveries and lasting impacts on Air Force research and development.

Dr. Kabrisky served as a Professor of Electrical Engineering at the Air Force Institute of Technology (AFIT) from February 1961 to January 1992 and an active Professor Emeritus until his death on February 2011.  His fifty years of service to AFIT, the Air Force, Department of Defense, and nation made exceptional contributions to technological advancements through graduate education, research and development.  Dr. Kabrisky distinguished himself as an outstanding teacher, technical consultant, lecturer to students from grade school to graduate school and community groups, and leader of innovative research that was internationally recognized.

As an internationally recognized expert in signal processing, Dr. Kabrisky made significant scientific and engineering contributions. His research made major advances in the areas of understanding of how information is processed in mammalian brains, how motion sickness is characterized by physiological symptoms, speech recognition algorithms applicable for voice process control, and imaging for object / target and face recognition.

Dr. Kabrisky pioneered the use of electronic silicon devices for taking measurements directly off the surface of the brain cortex using multiplexed two dimensional arrays. This research led the way for universal applications in bio-engineering protheses. Specifically, Dr. Kabrisky developed theories of how the human brain processes information to recognize visual objects. Applying these theories, Dr. Kabrisky developed a computer chip that could be implanted and not only measure brain electrical activity but also electrically stimulate the nervous system.  This work directly led to the innovation of implanted electrodes for those afflicted with diseases such as epilepsy and injuries that resulted in paralysis.

Dr. Kabrisky was also the leading international expert on the physiological symptoms of space adaptation sickness, motion sickness. The result of this research was a unique theory which models the onset of sickness as a form of seizure. This theorywas confirmed by 1988 experiments, through testing a large repertoire of treatments which had been tested successfully for similar seizure phenomena. This research led Nasa to a better understanding and an approach to mitigate the effects of space environments on astronauts.  His work on motion sickness (space adaptation sickness) solved an extremely important problem associated with astronauts getting sick during flight and not being able to perform their mission.

Dr. Kabrisky also pioneered the early development of speech recognition / processing systems.  He developed a series ofAFIT algorithms for processing speech independent of the speaker and applicable to tasks which require the recognizer to function with large vocabularies. Dr. Kabrisky’s research in this area of robust speech recognition laid critical foundations forfostering the development of Department of Defense and private industry products ranging from voice activated controls in advanced tactical aircrafts, to aides for the disable and handicapped and industrial process control. He also led a team that developed compression algorithms that maintained speech quality using his model of information processing in the human brain.  His work directly led to modern compression techniques used in speech transmission and storage.

In the area of image processing, Dr. Kabrisky developed algorithms which recognize objects independent of distortions found in imagery and applicable to multiple sensors. Dr. Kabrisky’s foundational research in this area led to the development of systems that were used to prepare for a potential war against the former Soviet Union where the targets were large numbers of tanks and in also fielded in the first Gulf War to detect Scud missiles.  In the 1990s, Dr. Kabrisky helped lead a team of engineers that developed the world’s most accurate breast cancer detection system.  This highly successful product is still sold internationally and since its inception, has helped in the detection of thousands of breast cancers before they would have otherwise been detected.

Dr. Kabrisky was a personal mentor to over 250 graduate research students.  He was a highly effective and innovative teacher, motivating thousands of military officers in the areas of human effectiveness, pattern recognition, and signal processing.  Dr. Kabrisky was a renown and internationally respected speaker, researcher and author.  He author over 450 publications and four books including seminal work in the area of modeling the human visual system.  Dr. Kabrisky was a Senior Member of the Institute for Electronic and Electrical Engineers (IEEE).

Weekly QuEST Discussion Topics, 11 Jan

QuEST 11 Jan 2019

Our traditional ‘Kabrisky Memorial Lecture’ – the first QuEST meeting of a calendar year – to start from scratch and baseline our use of terms and our goals for the group will be our focus.

n  QuEST is an innovative analytical and software development approach to improve human-machine team decision quality over a wide range of stimuli (handling unexpected queries) by providing computer-based decision aids that are engineered to provide both intuitive reasoning and “conscious” context sensitive thinking.

n  QuEST provides a mathematical framework to understand what can be known by a group of people and their computer-based decision aids about situations to facilitate prediction of when more people (different training) or computer aids are necessary to make a particular decision.

We will however this year take a slight digression at the beginning and have our QuEST colleague George C present an update on his mathematics of defeating reinforcement learning based AI as some of us have recently been focusing on counter AI/ counter autonomy.

Matt Kabrisky

Dr. Matthew Kabrisky was an Air Force pioneer and innovator in technological advancements.  From Air Force aviator in the 1950s to professor, mentor, and researcher until the 1990s, Dr. Kabrisky made important discoveries and lasting impacts on Air Force research and development.

Dr. Kabrisky served as a Professor of Electrical Engineering at the Air Force Institute of Technology (AFIT) from February 1961 to January 1992 and an active Professor Emeritus until his death on February 2011.  His fifty years of service to AFIT, the Air Force, Department of Defense, and nation made exceptional contributions to technological advancements through graduate education, research and development.  Dr. Kabrisky distinguished himself as an outstanding teacher, technical consultant, lecturer to students from grade school to graduate school and community groups, and leader of innovative research that was internationally recognized.

As an internationally recognized expert in signal processing, Dr. Kabrisky made significant scientific and engineering contributions. His research made major advances in the areas of understanding of how information is processed in mammalian brains, how motion sickness is characterized by physiological symptoms, speech recognition algorithms applicable for voice process control, and imaging for object / target and face recognition.

Dr. Kabrisky pioneered the use of electronic silicon devices for taking measurements directly off the surface of the brain cortex using multiplexed two dimensional arrays. This research led the way for universal applications in bio-engineering protheses. Specifically, Dr. Kabrisky developed theories of how the human brain processes information to recognize visual objects. Applying these theories, Dr. Kabrisky developed a computer chip that could be implanted and not only measure brain electrical activity but also electrically stimulate the nervous system.  This work directly led to the innovation of implanted electrodes for those afflicted with diseases such as epilepsy and injuries that resulted in paralysis.

Dr. Kabrisky was also the leading international expert on the physiological symptoms of space adaptation sickness, motion sickness. The result of this research was a unique theory which models the onset of sickness as a form of seizure. This theorywas confirmed by 1988 experiments, through testing a large repertoire of treatments which had been tested successfully for similar seizure phenomena. This research led Nasa to a better understanding and an approach to mitigate the effects of space environments on astronauts.  His work on motion sickness (space adaptation sickness) solved an extremely important problem associated with astronauts getting sick during flight and not being able to perform their mission.

Dr. Kabrisky also pioneered the early development of speech recognition / processing systems.  He developed a series ofAFIT algorithms for processing speech independent of the speaker and applicable to tasks which require the recognizer to function with large vocabularies. Dr. Kabrisky’s research in this area of robust speech recognition laid critical foundations forfostering the development of Department of Defense and private industry products ranging from voice activated controls in advanced tactical aircrafts, to aides for the disable and handicapped and industrial process control. He also led a team that developed compression algorithms that maintained speech quality using his model of information processing in the human brain.  His work directly led to modern compression techniques used in speech transmission and storage.

In the area of image processing, Dr. Kabrisky developed algorithms which recognize objects independent of distortions found in imagery and applicable to multiple sensors. Dr. Kabrisky’s foundational research in this area led to the development of systems that were used to prepare for a potential war against the former Soviet Union where the targets were large numbers of tanks and in also fielded in the first Gulf War to detect Scud missiles.  In the 1990s, Dr. Kabrisky helped lead a team of engineers that developed the world’s most accurate breast cancer detection system.  This highly successful product is still sold internationally and since its inception, has helped in the detection of thousands of breast cancers before they would have otherwise been detected.

Dr. Kabrisky was a personal mentor to over 250 graduate research students.  He was a highly effective and innovative teacher, motivating thousands of military officers in the areas of human effectiveness, pattern recognition, and signal processing.  Dr. Kabrisky was a renown and internationally respected speaker, researcher and author.  He author over 450 publications and four books including seminal work in the area of modeling the human visual system.  Dr. Kabrisky was a Senior Member of the Institute for Electronic and Electrical Engineers (IEEE).