Understanding intent.

In marketing terminology, an intent is an implied goal.  It is different from interest.  It is different from search.  It is different from browsing.  An intent is an actionable vendor-defined product action.  It may include examples like: intent to buy shampoo, intent to buy organic whole grain muesli, or intent to subscribe to The Economist. 

Understanding intents requires understanding the main goal of each and every inventory item available.  It requires inferring the customer's goal from among many novel directions.  And it requires match making from one to the other.  For example there are over 50 white papers available on Executive Intelligence.  They sub divide into three broad topics - or intents:  neurological, computational, and economic.  How does one decide which white paper gets which tag or attribution?  And how does one decide that there are three broad topics, and not four?  All of these require not only vast processing and memory, but also sound executive judgment.  Understanding intent is being the customer liaison, is being the shopkeeper.  Tracking and matching these intents appropriately is by definition marketing.  It is one of the core factors for business profitability.  Understanding intents is understanding a bit about being human.

The early cognition research spans childhood.  Newborns prefer faces to non-faces (Mondloch, et al. 1999) and to human languages to non-language sounds (Eimas, et al. 1971).  Isolating a human target is the first step as it draws our attention resources to focus on intent sources. 

By 3 months, infants expect caregivers to interact in certain prescribed manners.  If caregivers' faces are still and expressionless while all else is unchanged, the infants smile less, avert their gaze, and display physiological stress.  (e.g. heart rates elevated; Kisilevsky, et al., 1998; Toda & Fogel, 1993; Tronick, et al., 1978).  Infants also follow adult gaze tracking (Butterworth, 2001) and pointing (Morisette, et al., 1995).  This shared attention establishes common associative ground, language, communication, and practice coordinating behaviors.  Attending to a human target is the second step as it draws our development to match normal human behaviors. 

By 6 months, infants expect people to behave differently on different targets (Legerstee, Barna, & DiAdamo, 2000).  For example, a human actor either speaks or reaches to a hidden target.  Upon revealing the target - which can be either a person or an object -  the infants show surprise (i.e. expectation violation) if the human actor were speaking to an object or reaching to a person.  The infants have come to develop a sense of continuity in target and goal tracking behavior.

Meltzoff (1995) showed 18-month old toddlers either a person or a toy robot trying and failing a task - say, pushing a button with a stick.  The toddlers will attempt to complete the task if a person tried, but not if a toy robot tried.  They learned that human sources provide a level of contextual explanation of intent that non humans do not.

Rapacholi & Gopnik (1997) explored if 18-month olds can correctly decide which snack to share with an adult.   The choices included crackers (toddlers' preferred snacks) and broccoli (toddlers non-preferred snack).  The initial inclination is for the toddlers to offer crackers since that is a preferred snack.  However, if the adults expressed preference for broccoli, the toddlers offered the broccoli instead.  By 18 months, toddlers are able to represent others' preferences separately from their own.

As the children develop into later childhood, they begin to develop the ability to hold stable abstract characteristic representations for human targets:  

Under age 5, the children tend to describe others via external observable characteristics.  "X is my best friend.  We play in the street.  I like him because I play with his toys." (Livesley & Bromley, 1973).

Over age 9, the children tend to describe others using stable dispositions leading to predicted behaviors.  "X is very modest…. He always seems good-tempered…He does not seem to voice his own opinions to anyone.  He easily gets nervous." (Livesley & Bromley, 1973.)

Social cognition development.

The research shows that humans develop the ability to understand others' intents incrementally with increasing sophistication.  But to scale this up without waiting 20 years or gambling with mass hirings, the question remains how exactly do humans develop this social cognition of intent understanding?  Proposals include:

The keyword is Mirroring.  Iacoboni, (2009) explored the concept of specific mirror neurons in the brain that simulate others' states and can predict their goals or intents.  Uddin, et al. (2009) located this Mirroring system in the frontal parietal system and proposed that it helps map others' goals to what the individual self would do - the proverbial putting oneself in another's shoes.  Further evidence comes from the opposite of mirroring, or autism, which strongly and disproportionately impairs intent understanding.   (Baron-Cohen,2001; Baron-Cohen, Leslie, & Frith, 1985).  mplies built in brain section.  This implies there is a specific built-in section of the brain dedicated to intent understanding.  It is a binary function - wither we have it or we do not.

However, Hickok (2015) explores this issue further and raises objections.  The issue is not so simple.  The brain is interleaved and distributed, without clear evidence of dedicated, modular functions.  The experimental evidence also defies simple binary functions.  Autistic individuals that fail on the basic intent understanding task can still pass on simplified forms less taxing on working memory processes.  This raises the possibility that intent understanding need not be a unique human (primate) function but could rely on information processing capability.  Understanding intent may require resources sufficient not only to monitor own goals, but also the sufficient to monitor and keep separate others'.

To understand intent, the individual needs to be able to track another's actions, track another's disposition, and put it all in context all while drawing upon own knowledge - yet keeping it sufficiently inhibited so as to only inform rather than bias the understanding.  This is a complicated chain of capabilities which can be taxing to even developed adults.  A Piagettian/Neo-Piagettian Theory of Childhood Development points towards expanding or increasingly efficient working memory processes as key towards cognitive development.  No specific brain module is required other than having sufficient processing capability. 

Children with more siblings tend to develop intent understanding faster (Jenkins & Astington, 1996), especially children with older siblings (Ruffman, et al., 1997).  This implies that intent understanding is neither due to a specific dedicated physical module nor due to general information processing development.  Rather, it is an arbitrarily-named function that arises simply because we find it useful.  As children encounter more social contacts in bigger and more dispersed "tribes," they would naturally require some form of intent understanding to successfully navigate and exploit other resources.  A "tribal" child would need to understand which snack - broccoli or crackers - to share with a tribe member not because they have a pre-wired module forcing them to do so but because doing so makes the tribe member happy and more likely to share their snacks in return.  A child with no tribe would not find sharing either snack helpful since there would be no quid pro quo.  

If we constrain intent understanding to natural language processing as a single modality to ease the analysis, then any insight gained here can lead to general purpose intent understanding later.  A natural language intent understanding manifests as a cogent conversation.  If a child speaks germane statements in a conversation with another person, that child understands the intent of the other person and of the conversation.  Each word and statement is a matter of choice, a decision the child makes.   The statements, "X is my best friend.  We play in the street.  I like him because I play with his toys…"  provide information to satisfy a perceived intent.  The questions stated or implied could have been, "Tell me about X.  Where do you see X?  Do you like him?"  Having multiple conversations play out in this manner scaffolds the child to expect and prepare these responses. 

Each response crafts from a pool of words and phrases the child encountered up to that point in time.  Each response addresses and directs conversation prompts the child has learned to expect up to that point in time.  With more experience, the statements, "X is very modest…. He always seems good-tempered…He does not seem to voice his own opinions to anyone.  He easily gets nervous…"  also provide information to satisfy a perceived intent.  But the older child in this case has encountered more possible intents and has grown - in a manner of speaking - from having a 5-choice multiple choice question to having a 500-choice multiple choice question.  The questions are more likely implied as well. 

The implication here is that intent understanding is an ongoing process.  It does not require any special brain machinery.  It simply requires ongoing exposure to language patterns and goals to expand a pool of word and phrase selection options that lead to an intent.