Archive for December, 2015

Weekly QuEST Discussion Topics and News, 18 Dec

December 17, 2015 Leave a comment

QuEST 18 Dec 2015

I want to start with a discussion on unexpected query and the AFRL conscious content curation effort (AC3).  Specifically our colleague Scott C. provided us with some ‘unacceptable responses’ from our current system and we had a discussion of the implication to our Quadrant view of the unexpected query space – either we don’t have the right awareness of the environment or we don’t have an appropriate model.  Our discussion then turned to the idea of how can a conscious wrapper (QuEST agent) facilitate transfer learning for these examples.  We will have a discussion on the use of consciousness to reprogram the subconscious reflexive system.

We also want to review some recent information about Viv Labs (they are working a very interesting problem – how to computers solving complex task through learning enabled automatic program synthesis  – this hits at one of our key interests – solutions that scale – it helps reduce issues associated with the limitation of human coding new solutions for every envisioned interaction between systems – Beyond Siri – this is consistent with our discussions on the unexpected query – example:

On the screen at the end of the room, a green V appears. Green bars radiate, and then it connects. This is Viv, their bid for world domination. It’s a completely new concept for talking to machines and making them do our biddingnot just asking them for simple information but also making them think and react. Right now, a founder named Adam Cheyer is controlling Viv from his computer. “I’m gonna start with a few simple queries,” Cheyer says, “then ramp it up a little bit.” He speaks a question out loud: “What’s the status of JetBlue 133?” A second later, Viv returns with an answer: “Late again, what else is new?”

To achieve this simple result, Viv went to an airline database called and got the estimated arrival time and records that show JetBlue 133 is on time just 62 percent of the time.

Onscreen, for the demo, Viv’s reasoning is displayed in a series of boxes—and this is where things get really extraordinary, because you can see Viv begin to reason and solve problems on its own. For each problem it’s presented, Viv writes the program to find the solution. Presented with a question about flight status, Viv decided to dig out the historical record on its own. The snark comes courtesy of Chris Brigham, Viv Labs founder number three.

Now let’s make it more interesting. “What’s the best available seat on Virgin 351 next Wednesday?”

One other topic I want to hit for the entrepreneurs in the crowd is the business model:

Priceline pays Google about $2 billion a year to get displayed at the top of cheap-flight searches. The entire Internet sales model is based on finding something, if you can find it, then going to the Web site or the app and looking some more and entering your dates and credit card. But Viv knows what Cheyer’s looking for. It knows if he likes hotels with swimming pools and the best deals on his favorite entertainment options, even the airport he usually flies from. And although some of this interactivity is already available on Google’s Siri clone, Google Now, Viv also knows how to enter all Cheyer’s personal data and credit-card numbers and execute the transaction—one-stop shopping without the stop.

Along those lines I want to expose people to a spectacular example of choosing a business model – success is NOT about the technology – an example provided by a valued colleague Curt C at Practice of Innovation – turbo tap:


The last topic for this last meeting of the QuEST group for the year is the What is QuEST one pager – for those interested in contributing to how we want to capture our goals succinctly we can get you the current draft – and then we also want to hit the current outline for the Kabrisky Memorial Lecture – the state of QuEST lecture I will give to the group 8 Jan – that is our annual effort to re-sync our focus and allow everyone to come up to speed on our positions and provide everyone with the collateral necessary for them to talk to anyone about what we are attempting to do.

news summary (36)

Categories: Uncategorized

Weekly QuEST Discussion Topics and News, 11 Dec

December 10, 2015 Leave a comment

QuEST 11 Dec 2015

See below for a link to Capt Amerika’s GEOINT talk from June 2015 (starts around 15 min mark)

In our continuing effort to tie together topics from the year to include in the Jan ‘Kabrisky memorial Lecture’ – ‘What is QuEST?’ – we will return to the topic of the unexpected query and specifically attempt to resolve the relationship between the Pan article on Transfer Learning – our use of the phrase ‘unexpected query’ – our quadrant diagram on automation versus autonomy and the unexpected query and the word Sandy V used last week ‘context’.

This topic then leads to a discussion on when we don’t have the right model for the agent to acceptably respond – conceptual combination –

  • Conceptual combination is the process of creating and understanding new meanings from old referents. Our ability to understand novel word compounds, such as octopus apartment or fame advantage, is predicated upon the inherently constructive nature of cognition that allows us to represent new concepts by mentally manipulating old ones.
  • Central to research into how people process such combinations is an understanding of what constitutes the representations of these concepts.

how do we form new models out of previously developed representations – we’ve covered this topic a couple of times this year – in fact this week I was asked to comment on a new article:

Human-level concept learning

through probabilistic

program induction

Brenden M. Lake,1* Ruslan Salakhutdinov,2 Joshua B. Tenenbaum3

11 DECEMBER 2015 • VOL 350 ISSUE 6266  Science

People learning new concepts can often generalize successfully from just a single example, yet machine learning algorithms typically require tens or hundreds of examples to perform with similar accuracy. People can also use learned concepts in richer ways than conventional algorithms—for action, imagination, and explanation. We present a computational model that captures these human learning abilities for a large class of simple visual concepts: handwritten characters from the world’s alphabets. The model represents concepts as simple programs that best explain observed examples under a Bayesian criterion. On a challenging one-shot classification task, the model achieves human-level performance while outperforming recent deep learning approaches.We also present several “visual Turing tests” probing the model’s creative generalization abilities, which in many cases are indistinguishable from human behavior.


We have previously discussed related works:


Interpretation and Representation: Testing the
Embodied Conceptual Combination (ECCo) Theory V2
Louise Connell (
School of Psychological Sciences, University of Manchester
Oxford Road, Manchester M13 9PL, UK
Dermot Lynott (
Decision and Cognitive Sciences Research Centre, Manchester Business School, University of Manchester
Booth Street West, Manchester M15 6PB, UK

  • The Embodied Conceptual Combination (ECCo) theory differs from previous theories of conceptual combination in two key respects.

–     First, ECCo proposes two basic interpretation types: destructive and nondestructive.

–     Second, ECCo assumes complementary roles for linguistic distributional information and perceptual simulation information. Here, we empirically test these assumptions using a noun-noun compound interpretation task.

  • We show that ECCo’s destructive/nondestructive interpretation distinction is a significant predictor of people’s successful interpretation times, while the traditional property/relation based distinction is not.

We also demonstrate that both linguistic and simulation systems make complementary contributions to the time course of successful and unsuccessful interpretation. Results support the ECCo theory’s account of conceptual combination

There was also the work:

Embodied conceptual combination
Dermot Lynott1* and Louise Connell2
1 Manchester Business School, University of Manchester, Manchester, UK
2 School of Psychological Sciences, University of Manchester, Manchester, U

Frontiers in psychology – November 2010 | Volume 1 | Article 212

  • Conceptual combination research investigates the processes involved in creating new meaning from old referents.

–     It is therefore essential that embodied theories of cognition are able to explain this constructive ability and predict the resultant behavior.

  • However, by failing to take an embodied or grounded view of the conceptual system,existing theories of conceptual combination cannot account for the role of perceptual, motor, and affective information in conceptual combination.
  • In the present paper, we propose the embodied conceptual combination (ECCo) model to address this oversight.

–     In ECCo, conceptual combination is the result of the interaction of the linguistic and simulation systems,

–     such that linguistic distributional information guides or facilitates the combination process,

–     but the new concept is fundamentally a situated, simulated entity.

  • So, for example, a cactus beetle is represented as a

–     multimodal simulation that includes

  • visual (e.g., the shiny appearance of a beetle)
  • and haptic (e.g., the prickliness of the cactus) information,

–     all situated in the broader location of a desert environment under a hot sun,

–     and with (at least for some people) an element of creepy-crawly revulsion** Affective aspects **

  • The ECCo theory differentiates interpretations according to whether theconstituent concepts are destructively, or non-destructively, combined in the situated simulation.
  • We compare ECCo to other theories of conceptual combination, and discuss how it accounts for classic effects in the literature.

news summary (35)

Categories: Uncategorized

Weekly QuEST Discussion Topics and News, 4 Dec

December 3, 2015 Leave a comment

This week we look forward to our colleague Sandy V presenting some of her latest findings on an ‘inference generating algorithm, based on an abstract representation of qualia space’.  

news summary (34)

Categories: Uncategorized