DARPA's future model of combat computing

---

DARPA's recent RFI on low-resource computing rethinks computation under battlefield conditions: scarce power, unreliable components, limited communications, cramped physical space and little opportunity for maintenance. Welcome to the burgeoning world of the sensor, drone, robotic scout and autonomous battlefield node.

This resembles the constraints that shape biological evolution. Living systems operate under severe energy budgets. Their components are noisy and failure-prone. They rely on local decision-making, tolerate damage, and continue functioning despite incomplete information; resilience matters more than perfection.

Future military systems will exhibit characteristics familiar in nature: distributed intelligence, local sensing and action, redundancy, graceful degradation and adaptation to specialised roles. An unattended sensor, a reconnaissance drone and a robotic logistics vehicle will look like different species occupying different ecological niches.

Natural selection optimises for survival and reproduction while military systems must remain understandable, controllable and trustworthy to their human operators. But it's going to be an SF jungle out there - it already is.

---

Machine Common Sense (MCS) - DARPA

The DARPA challenge

A past DARPA challenge kickstarted the self-driving car phenomenon. Will this new attempt to equip robots with common sense reasoning and interpersonal skills be as successful?

For some value of 'successful' of course.

DARPA's proposal starts with a short review of the disappointing record on 'common sense'.

"Since the early days of AI, researchers have pursued a variety of efforts to develop logic-based approaches to common sense knowledge and reasoning, as well as means of extracting and collecting commonsense knowledge from the Web.

While these efforts have produced useful results, their brittleness and lack of semantic understanding have prevented the creation of a widely applicable common sense capability."

DARPA breaks its new challenge into two substreams. The first bases itself on human infant cognitive development, as theorised by developmental psychology.

"The first approach will create computational models that learn from experience and mimic the core domains of cognition as defined by developmental psychology. This includes the domains of objects (intuitive physics), places (spatial navigation), and agents (intentional actors). Researchers will seek to develop systems that think and learn as humans do in the very early stages of development, leveraging advances in the field of cognitive development to provide empirical and theoretical guidance.

“During the first few years of life, humans acquire the fundamental building blocks of intelligence and common sense,” said Gunning. “Developmental psychologists have founds ways to map these cognitive capabilities across the developmental stages of a human’s early life, providing researchers with a set of targets and a strategy to mimic for developing a new foundation for machine common sense.”

To assess the progress and success of the first strategy’s computational models, researchers will explore developmental psychology research studies and literature to create evaluation criteria. DARPA will use the resulting set of cognitive development milestones to determine how well the models are able to learn against three levels of performance – prediction/expectation, experience learning, and problem solving."

The second stream is more bookish, mining the web.

"The second MCS approach will construct a common sense knowledge repository capable of answering natural language and image-based queries about common sense phenomena by reading from the Web.

DARPA expects that researchers will use a combination of manual construction, information extraction, machine learning, crowdsourcing techniques, and other computational approaches to develop the repository.

The resulting capability will be measured against the Allen Institute for Artificial Intelligence (AI2) Common Sense benchmark tests, which are constructed through an extensive crowdsourcing process to represent and measure the broad commonsense knowledge of an average adult."

I am impressed by neither approach.

It's too tempting to theorise a situated agent in terms of ungrounded abstractions, such as the belief–desire–intention software model. In this way we think of the frog, sat in its puddle, as busily creating in its brain a set of beliefs about the state of its environment, a set of desires such as 'not being hungry' and an intention such as 'catching that fly with a flick of its tongue'.

While we may describe the frog in such unecological folk-psychological terms - as is the wont of developmental psychologists - Maturana et al pointed out that is not what the frog does.

---

At this point it is tempting to bring in Daniel Dennett's ideas about first and second order intentionality but I distrust even this. Treating another animal as an agent (rather than an instrumental object in the environment) which is the hallmark of second-order intentionality, seems extraordinarily rare in the animal kingdom. The Wikipedia article, "Theory of mind in animals", suggests there is partial - but not compelling - evidence for it only in the case of some social animals. But the concept remains ill-defined.

We use a reified intentional language (note: language) with modal operators such as believes and desires to describe those objects we classify as agents. By virtue of  'possessing' their own beliefs, desires and intentions (= plans) they are taken to exhibit autonomy. We don't normally use such language to describe bricks and cauliflowers. We do use such language to describe spiders, mice and roombas - first-order intentional systems.

Some systems (such as people) seem able themselves to characterise objects in their environment (possibly including themselves) in intentional terms. They have the capability, for example, to look at us and see us as agents. We call these systems second order intentional systems. I would include cats and dogs and children here, noting how they manipulate us (spiders, mice and roombas don't seem to notice us as agents).

So how do they do that? That's the interesting architecture question DARPA is asking, and nobody knows.

I expect that a biological second order intentional system possesses neural circuitry which encodes a representation of intentional agents in its environment as persistent objects, together with links to situations (also neurally-encoded) representing beliefs, desires and intentions relativised to that agent. Think of the intuitions underlying Situation Semantics, implemented as computationally-friendly semantic nets: I wrote about it here.

I used to think that the only way forward was to design the best higher-order intentionality architecture possible, embody it in a robot child and expose it to the same experience-history as that involved in human infant socialisation. But I notice that DeepMind and similar have made huge leaps forward in simulated environments which decouple cognitive challenges from the messy (and slow) domains of robotic engineering.

So I imagine that's where the smart money will be.

Naive generate-and-test won't hack it

When I was young I toyed with the following idea.

Pretty much any concept can be adequately expressed in a mini-essay of a thousand words.

Simply generate all possible articles of a thousand words and somewhere you will find the answer to all problems.

Want the design of a stardrive engine? Immortality? The theory of perfect governance?

It's all in there somewhere.

---

How many essays though? Apparently the average educated speaker of English knows about 40,000 words. So for our first estimate, we could simply raise 40,000 to the power of 1,000 .. but most of those 10^4,602 essays would be wildly ungrammatical. We can do better.

I reviewed a sample text: the introductory quote in Peter Seibel's "Practical Common Lisp".

The first five sentences comprised 100 words in total which broke down into:

• nouns: 20%
• verbs: 15%
• adjectives: 10%
• others: 55%

A certain amount of hand-wavy rounding of course. Assume we adopt the very restrictive constraint of exactly one syntactic structure for the entire set of essays, then the total number reduces to a product of:

(number-of-English-words-in-category) ^{(number-of-words-of-this-category-in-essay)}

or,

8,000²⁰⁰ * 6,000¹⁵⁰ * 4,000¹⁰⁰ * 22,000⁵⁵⁰ = 10^4,092

That's still a big number*. Suppose only one 'essay' in a billion was semantically sensible and we could read one essay per second. That's 10^4,083 seconds .. or 3 * 10^4,066 billion years.

The merits of a compact notation.

---

Exhaustive search through the space of all possible candidates isn't a very good way of proceeding. And this has important implications for DARPA's third wave - contextual AI - which I wrote about previously.

In his excellent exposition (YouTube), John Launchbury highlighted the very large number of training instances needed to force convergence for today's artificial neural networks. By comparison, children learn new concepts from very few examples.

John Launchbury's proposed solution was - correctly - to identify additional constraints which might dramatically collapse the search space. His chosen example showed the benefits of adding the dynamics of handwriting characters to the resultant bitmaps normally used for training. It turns out that if you consider how the image might have been created, it makes recognition a lot easier.

It's not hard to identify the extra constraints about the world which children use. They interact with new objects, touch them, throw them, bite them and try to break them. Thus are acquired notions of 3D structure, composition and texture to augment what their visual systems are telling them.

I really do think that a high priority should be given to embodied robotics in the next wave of AI research.

---

Another example John Launchbury discussed was the Microsoft Internet-chatbot "Tay".

Apparently this was the least-offensive tweet Launchbury could find. But what would an AI have to know about contemporary mores to self-reject statements like that?

For extra credit, discuss the 'situated cognition' thesis that only through active and corporeal participation in the social world can one truly understand social concepts.

Particularly emotionally-charged ones.

---

* Since

(i)  I don't consider all the syntactically-permissible permutations of the ways in which nouns, adjectives, verbs and others could be mixed up in the thousand words, while

(ii)  the size of the 'others' vocabulary is likely to be way smaller than 22,000 (so if, for example, the 'others' vocabulary size was 2,200, this would reduce the overall essay-set size by a factor of 10⁵⁵⁰ - a distinction, however, without a practical difference),

this calculation counts as pretty bogus. I only wanted to demonstrate, however, that no matter how you cut it, the numbers involved are simply ginormous.

DARPA: three waves of AI

High production values for DARPA's US Military roadmap and vision for AI (February 2017).

This will be the basis of funding going forward. The images below are taken from this slide-pack, more sophisticated than anything I've seen from the likes of Accenture.

Click on any of the pictures to make larger - or better, review the entire slide-set.

Although this 'three wave' model is not too surprising, it's still an accurate view as to where research is heading.

---

If human beings are taken as exemplars of neural nets which can explain their own, contextual operation, it's worth noting that such explanations have a curious character.

No human can explain their own sub-conscious neural processes. If asked to explain how you know that a picture of a cat is indeed that of a cat, you are not going to elucidate details of early visual processing in your visual cortex.

Instead, you are going to traffic in high-level, symbolic descriptions of putative intermediate stages in scene interpretation. The talk will be of features such as fur, shape, the environment of said animal.

These intermediate-level symbolic descriptions are remote indeed from the actual neural processes which it is claimed implemented them .. and indeed will have only a contingent (although highly correlated if accurate) relationship with them.

Self-deception is never far away in the third wave!

---

If you have sixteen minutes, John Launchbury's presentation of DARPA's strategy is excellent.

Interestingly, John Launchbury is British.