Author Archive

Weekly QuEST Discussion Topics and News, 16 Mar

March 15, 2018 Leave a comment

QuEST 16 March 2018


Last week our ‘Bletchley Park team’ led a discussion on the ideas related to multi-agent systems – specifically how can one agent develop a model of another agent’s representation.   We want to continue this week with that discussion and provide them feedback on the types of ‘models’ and agents we are interested in. 


As a reminder of context – we’ve emphasized that ‘flexible AI’ will require these multi-agent systems that are not originally designed to work together be able to interoperate / compose / adapt.  We’ve defined the axes of flexibility to be in terms of Peer (relationships between agents – supervisor, peer, subordinate), task (being able to do multiple tasks) and cognition (flexible representational options). 


We’ve often discussed steps along the way to full flexibility to include Peer ~ interoperable, Task ~ composition, Cognition ~ adaptable.   Our team of mathematicians ‘Bletchley Park team’ have been working on these ideas and will continue a discussion on the first two concepts.


Title:  Agent representations for interoperability and composition


Speakers:  Cybenko, Erdmann, Oxley


Abstract:  A concrete multimodal representation for agents is proposed.

The representation is suitable for interoperability and composition, and is based on ontology and machine learning concepts.  Ideas from recent results on data topology and the manifold hypothesis will be presented and related to this representation.

Categories: Uncategorized

Weekly QuEST Discussion Topics, 9 Mar

In our recent discussions we’ve converged on the relationship between ‘Autonomy’ to Artificial Intelligence.   Specifically we’ve emphasized the need for ‘flexible AI’.  We’ve defined the axes of flexibility to be in terms of Peer (relationships between agents – supervisor, peer, subordinate), task (being able to do multiple tasks) and cognition (flexible representational options).  We’ve often discussed steps along the way to full flexibility to include Peer ~ interoperable, Task ~ composition, Cognition ~ adaptable.   Our team of mathematicians have been working on these ideas and will lead a discussion on the first two concepts.


Title:  Agent representations for interoperability and composition


Speakers:  Cybenko, Erdmann, Oxley


Abstract:  A concrete multimodal representation for agents is proposed.

The representation is suitable for interoperability and composition, and is based on ontology and machine learning concepts.  Ideas from recent results on data topology and the manifold hypothesis will be presented and related to this representation.

Categories: Uncategorized

Weekly QuEST Discussion Topics, 2 Mar

QuEST 2 March 2018

There are several topics we want to hit this week.

There are challenges in artificial intelligence but capturing what is solved and what remains to be solved and what may never be solved is always difficult to communicate.  We are always seeking examples that can help senior leaders understand.  Our recent interest in the DARPA DEFT work provided one view in the text processing area that is worth noting.

ACT3 is focused on ‘Improving every Decision’.  A recent series of discussions with a colleague Reggie B led us into a discussion on Cinefin, an IBM approach to business decision making.  The construct also provides a spring board into discussions relating to our recent ‘Beyond the OODA nonsense’ and cognitive / collaborative EW discussions.  We will briefly discuss their approach to decision making.

Last week we discussed modeling other agents in a pool of agents that may want to work/need together.  One comment from our colleague George C. was modern big-data suggest the answer lies in the data.  So that leads us down two related threads of discussions.  There was our previous discussions on the unreasonable effectiveness of Data, an article by by Alon Halevy, Peter Norvig, and Fernando Pereira, of Google. 

In addition there was the work by George C. entitled:  Matching Conflicts: Functional Validation of Agents. 

  • In most working and proposed multiagent systems, the problems of identifying and locating agents that can provide specific services are of major concern.
  • A broker or matchmaker service is often proposed as a solution.
  • These systems use keywords drawn from application domain ontologies to specify agent services, usually framed within some sort of knowledge representation language.
  • However, we believe that keywords and ontologies cannot be defined and interpreted precisely enough to make brokering or matchmaking among agents sufficiently robust in a truly distributed, heterogeneous, multiagent computing environment.
  • This creates matching conflicts between, a client agent’s requested functionality and a service agent’s actual functionality.
  • We propose a new form of interagent communication, called functional validation, specifically designed to solve such matching conflicts.
  • In this paper we introduce the functional validation concept, analyze the possible situations that can arise in validation problems and formalize the mathematical framework around which further work can be done.

One last topic is in the area of innovation.  In the discussions with Reggie B we also got some vectors to recent models of innovation cycles that are worth discussing.   One topic was the Schumpeterian Cycle of Innovation and Entrepreneurship.

The innovation theory of a trade cycle is propounded by J.A. Schumpeter. He regards innovations as the originating cause of trade cycles. The term “innovation” should not be confused with inventions. Inventions, in ordinary parlance, are discoveries of scientific novelties. Innovation is the application of such inventions to actual production (i.e., exploiting them).

It is innovations that are subject to cyclical fluctuations, not inventions. Innovation, thus, in economics means the commercial application of inventions like new techniques of production, new methods of organisation, novel products, etc.

Schumpeter regards trade cycles as the offspring of economic progress in a capitalist society. Cyclical fluctuations are inherent in the economic process of industrial production. When there are internal changes taking place on account of innovation, the development process begins…

Categories: Uncategorized

Weekly QuEST Discussion Topics, 23 Feb

February 22, 2018 Leave a comment

QuEST 23 Feb 2018

Some of us have been having a side-bar discussion on meaning – specifically related to the idea of ‘conversations as a platform’.  One of the ACT3 major concerns is having agents be able to work together – this may require modeling agents – one particularly successful area of agent models is in collaborative filtering – let’s look at that area for some insights

To expound on that issue we want to revisit some prior discussions on ‘big-data’ and a discussion on collaboration tools like ‘Slack’.

If you work at one of the 50,000 companies that pay to use Slack for workplace collaboration, you probably spend hours on it, swapping information, bantering, and sharing files with your colleagues. It’s a casual, flexible way to interact—you tap out brief messages in group chat rooms (called channels) instead of sending e-mail, and it feels more like a smartphone app than typical office software.

  • Collaborative filtering (CF) is a technique used by some recommender systems.
  • Collaborative filtering has two senses, a narrow one and a more general one.
  • In general, collaborative filtering is the process of filtering for information or patterns using techniques involving collaboration among multiple agents, viewpoints, data sources, etc

We’ve previously discussed the idea of ‘meaning’ being agent centric – and the key issue is how to capture a representation that can exhibit flexibility in ‘meaning making’.

For example:  Big Data – QuEST perspectives v11 short deck – AC3 inserts (AFRL conscious content curation).

  • From the IQT (In-Q-tel) quarterly (vol 7 no 2) fall 2015 issue – discusses “Artificial Intelligence gets Real”.
  • Predictions with Big Data By Devavrat Shah:

–     We know how to collect massive amounts of data (e.g., web scraping, social media, mobile phones),

–     how to store it efficiently to enable queries at scale (e.g., Hadoop File System, Cassandra) and

–     how to perform computation (analytics) at scale with it (e.g., Hadoop, MapReduce).

–     And we can sometimes visualize it (e.g., New York Times visualizations).

But from a QuEST perspective:

  • Current approaches to big-data bring extremely valuable insights – even in very large data sets with low information density
  • These approaches do so finding correlations
  • Most often they don’t attempt to answer questions on causation
  • QuEST seeks to deliver a simulation based deliberation approach (not correlation not causation)

–     degrees of freedom for simulation possibly chosen via ‘big-data’ infrastructure

  • Using the situated simulation consciousness provides an alternative to the issues above – you don’t have to have the experiences and been able to articulate a model to be able to understand causation – BUT – you also don’t have to have experience all of the data to be able to relate to prior data – the simulation approach provides something between or maybe outside of those – better than both?

Modeling Relationships in Referential Expressions
with Compositional Modular Networks
Hu – UC Berkeley

  • People often refer to entities in an image in terms of their relationships with other entities.
  • For example, the black cat sitting under the table refers to both a black cat entity and its relationship with another table entity.
  • Understanding these relationships is essential for interpreting and grounding such natural language expressions.

Most prior work focuses on either grounding entire referential expressions holistically to one region, or localizing relationships based on a fixed set of categories

From our prior discussions on meaning:

  • Meaning, value and such like, are not intrinsic properties of things in the way that their mass or shape is.
  • They are relational properties.
  • Meaning is use, as Wittgenstein put it.
  • Meaning is not intrinsic, as Dennett has put it.
  • And here’s the point: if you know everything there is to know about that web, then you know everything there is to know about the data.

And the precursor to that work:

Neural Module Networks
Jacob Andreas Marcus Rohrbach Trevor Darrell Dan Klein
University of California, Berkeley

  • Visual question answering is fundamentally compositional in nature—a question like where is the dog? Shares substructure with questions like what color is the dog? And where is the cat?
  • This paper seeks to simultaneously exploit the representational capacity of deep networks and the compositional linguistic structure of questions.
  • We describe a procedure for constructing and learning neural module networks, which compose collections of jointly-trained neural “modules” into deep networks for question answering.
  • Our approach decomposes questions into their linguistic substructures, and uses these structures to dynamically instantiate modular networks (with reusable components for recognizing dogs, classifying colors, etc.).
  • The resulting compound networks are jointly trained.
  • We evaluate our approach on two challenging datasets for visual question answering, achieving state-of-the-art results on both the VQA natural image dataset and a new dataset of complex questions about abstract shapes.

Categories: Uncategorized

Weekly QuEST Discussion Topics and news, 16 Feb

February 15, 2018 Leave a comment

QuEST 16 Feb 2018

The discussion last week led to the concept that we are fighting an adaptable foe – and that reminded us of some prior QuEST discussions specifically – A Change in strategy in the war on cancer, the GWOT, Cyber Warfare and flexible AI = Autonomy, those discussions also provide some relevant points to consider relevant to our current focus on creating a ‘knowledge platform’.  This week we will have a discussion on some of those points.

‘Magic Bullet’ fallacy

  • Nobel Laureate P. Ehrlich introduced the concept of ‘magic bullet’ over 100 yrs ago (compounds to selectively target/kill tumor cells or disease causing organisms without negative impact on normal cells).
  • Problems we face in asymmetric war or cyber warfare are related – Is it reasonable to think that there exists some technological solution that will allow us to have a warfare ‘magic bullet’ in either of these wars?
  • Even worse does the thought of a ‘magic bullet’ dominating our work cause us to not pursue technology avenues that would bring great value!

Current focus in all of the original three domains and now Autonomy

  • Cancer war – driven by implicit assumption that a magic bullet will be found – even in detection the idea is to find, pursue and kill all cancer cells (sometimes to the detriment of patient) – in fact the idea that you can ‘kill’ the cancer before it metastasizes is or may be flawed
  • Cyber warfare (defense or offense) – driven by assumption that we will be able to identify the malicious processes (or all the bots in a DNS attack) or the enemy cyber targets via a magic bullet recognition process and prevent any possibility of damage (sometimes to the detriment of the mission) – flawed idea of perimeter cyber defense – Science 10
  • Asymmetric warfare – similarly we are pursuing a path that assumes via some technological ‘magic bullet’ (both in information processing and/or small precise munitions) we will be able to target the insurgents hiding in the civilian populace without collateral damage to the ‘normal cells’ = civilian populace.  That we will prevent all bombings!(behavioral signatures work to detect to the left of the boom make this assumption)  The current discussion on Counter Terrorism versus Counter Insurgency falls into this same trap – why search for a PETN detection solution
  • Autonomy – in our pursuit of a knowledge platform that can in an agile fashion be applied to a range of Autonomy needs we have suggested that there are some common aspects of knowledge creation that can be put into a platform – and by doing this not only to you reduce the development time you can also improve performance by facilitating transfer learning

Maybe insert some discussion of counter insurgency and counter terrorism – that drives home the idea of some actions have to be prevented …

news summary (72)

Categories: Uncategorized

Weekly QuEST Discussion Topics, 9 Feb

February 8, 2018 Leave a comment

Beyond the OODA Nonsense

Zealots suggest that autonomy (smart software/hardware) will achieve the shortening of the kill chain (mission effect chain) for hybrid/irregular or cyber warfare!

One major challenge in hybrid/irregular and cyber warfare, as in all other forms of warfare, is obtaining timely and accurate information in a useable form.  Information is critical to prosecuting a target.  These targets may be easy to ultimately prosecute but very hard to find, as they often hide in plain sight.  This is true for terrorists and insurgents and especially true for network intruders and malicious software.  Additionally, achieving desired end-state goals may not be possible due to the potential for unintended damage caused during target prosecution (e.g., to winning the population’s hearts and minds or preventing Blue Force use of our own networks).  The operational cyber domain and the advancement of artificial intelligence leading to Autonomy offer new challenges and complexities that may drive a reshaping of traditional war fighting doctrine. Automated techniques and human-in/on-the-loop insights must be synergistically integrated to achieve ultimate success.  Without integration of automated capabilities into how humans process data into information will lead to failure.

This will be a discussion on how the common use of the Observe, Orient, Decide, Act (OODA) construct and the misleading assumption that autonomy will allow the acceleration of the kill-chain are potentially damaging.  We present five issues with OODA.  These issues illustrate some of the daunting challenges the Air Force face.   The last issue we discuss leads to a conclusion that autonomy will not be the magic bullet.  We must instead pursue the more efficient option of integrating humans and computers to generate a more robust solution.

Categories: Uncategorized

Weekly QuEST Discussion Topics, 2 Feb

February 1, 2018 Leave a comment

QuEST 2 Feb 2018

After last week’s discussion we have spent the week focusing on the ‘knowledge platform’.  How can we take the problems we are addressing and use the resulting advances to mature a ‘knowledge platform.’  The platform provides the business model and the technological framework to in an agile fashion be able to deliver capability for a range of applications.  The key question is “what is the ‘representation’ approach for this knowledge creation platform?”.  The discussions have focused on what is a simulation / situated nature of the representation.

This week we want to return to some of the material that over the years have supported our ideas on simulation / situated nature of the conscious representation.  For example:

620 Barsalou

Annu. Rev. Psychol. 2008.59:617-645. Downloaded from

by EMORY UNIVERSITY on 02/13/08.

Grounded cognition rejects traditional views that cognition is computation on amodal symbols in a modular system, independent of the brain’s modal systems for perception, action, and introspection.

Instead, grounded cognition proposes that modal simulations, bodily states, and situated action underlie cognition. Accumulating behavioral and neural evidence supporting this view is reviewed from research on perception, memory, knowledge, language, thought, social cognition, and development.

Theories of grounded cognition are also reviewed, as are origins of the area and common misperceptions of it. Theoretical, empirical, and methodological issues are raised whose future treatment is likely to affect the growth and impact of grounded cognition.

Categories: Uncategorized