User:Danhuynh cs/DRAFT

Bio: Currently a senior at Rutgers University. This account will be used for the W assignments within Rutgers's computer graphics course. Current assigned topics: 1) Computational Narrative 2) Human Aware Architectural & Urban Design

Computational Narrative
Computational narrative is the process of making more human-like artificial intelligence using computational models for narrative intelligence. Narrative intelligence is the process of comprehending and generating narratives. Artificial intelligence handles fact based reasoning very well, but it struggles when attempting more abstract concepts (Riedl, Mark O., 2016). Narratives are able to be used to educate and entertain. The Bible, for example, can impart lessons of morality and conduct. AI has difficulty with non-relational reasoning, processing emotion, and other inherent human cultural information that can be often found in narratives (Langley, 2012). A basic AI is unable to interpret and respond to emotion. A basic AI is unable to output information that is not known to it. Machines are not exposed to the narratives that are present in human culture, and thus, have not processed much of the human culture that propagates through narratives, making their behavior non-human-like.

History
Early Works in AI Development
 * Pyschology played a large role in creating early AI models. Herbert Simon and  Allen Newell, of the  Carnegie Institute of Technology, were two early pioneers that pursued a cognitive model in developing AI. They viewed AI has a system to model human cognition (Langley, 2012).

Current AI Limitations In Interpreting Narratives
 * AIs lack shared common knowledge and may not have the prior knowledge needed to interpret a text. In addition, narratives often employ many metaphors, which require more conceptual thinking. Lastly, human narratives are complex and are rarely bound by patterns that can be taught to AI (Riedl, 2016).

Computational Models
To enable AIs to process narrative information, there must be a defined process for AIs to be able to parse and interpret relevant information. The following models are just a few that have been proposed to handle this process.

Semantic Approaches
These approaches focus on processing the language found in narratives. A computer can be given a sentence template to parse; and upon encountering a known word, the computer can process it as the user desires.


 * Counterfactual Cognitive Model
 * Computers have a difficult time interpreting counterfactuals, which are often used in narrative context. Counterfactuals are prevalent in human speech, and they represent, and play a role in, higher order thinking. By using counterfactuals, variations on themes can be introduced, leading to conceptual thinking (Lakoff & Narayanan, 2010).

Non-Semantic Approaches
These approaches are an alternative to the language processing approach.


 * Narrative Event Chains
 * This model improves upon the standard verb-only approach of processing text. Narrative event chains are given an ordered set of events that revolve around a single protagonist. The AI interprets the actions of this protagonist, and the objective, ultimately, is for the AI to learn from the protagonist's behavior and actions (Chambers & Jurafsky, 2008).


 * Narrative chains are created using using probablistic measurements, ordered by some time. When the AI is presented with a list of events, it can infer which events belong to which narrative chain by their sentence structure.
 * The following are characteristics of a Narrative Event Chain:
 * a) Narrative events are defined as a verb and the dependents of that verb, but a single protagonist is used to characterize a narrative chain.
 * b) A protagonist is used to display only one perspective on the narrative event chain.


 * Benefits of using Narrative Event Chains Over Standard Semantic Approach
 * - Reduces total evaluated entries to only narrative chains with protagonist.
 * - 37% improvement in accuracy over the verb-only baseline (Chambers & Jurafsky, 2008).

Machine Enculturation
An approach that seeks to teach AIs human sociocultural values by having the AI infer from collected works of fiction from many cultures. This approach requires an AI that has human-level narrative comprehension (Riedl, 2016).
 * Current open questions include how an AI can infer which actions to take when it lacks common knowledge garnered through experience. One approach is to specify branching actions for the AI to take, while introducing a value-aligned reward signal. If the AI chooses an action that does not allow it to continue throughout the chosen path, then it recieves a punishment signal. However, actions that allow the AI to continue on the designated path results in a reward signal (Riedl & Harrison, 2016).

Generating Stories
“The Policeman’s Beard Is Half Constructed” is the first book written by a computer. Published in 1983. Racter, the software claimed to have written the book, was released in 1984. Racter used a syntax directive which specified how sentences should be written. The sentences generated were syntactically legal, but the words were generated pseudorandomly via a seed. The sentences lacked coherency (Chamberlain, 1984).

References (Computational Narrative)

 * Chamberlain, Bill. (March 1984). "Racter, 1984, Introduction". http://www.ubu.com/historical/racter/index.html.
 * Chamberlain, Bill. (March 1984). "Racter, 1984, Introduction". http://www.ubu.com/historical/racter/index.html.
 * Chamberlain, Bill. (March 1984). "Racter, 1984, Introduction". http://www.ubu.com/historical/racter/index.html.
 * Chamberlain, Bill. (March 1984). "Racter, 1984, Introduction". http://www.ubu.com/historical/racter/index.html.
 * Chamberlain, Bill. (March 1984). "Racter, 1984, Introduction". http://www.ubu.com/historical/racter/index.html.
 * Chamberlain, Bill. (March 1984). "Racter, 1984, Introduction". http://www.ubu.com/historical/racter/index.html.
 * Chamberlain, Bill. (March 1984). "Racter, 1984, Introduction". http://www.ubu.com/historical/racter/index.html.
 * Chamberlain, Bill. (March 1984). "Racter, 1984, Introduction". http://www.ubu.com/historical/racter/index.html.

Human Aware Architectural & Urban Design
Humans have many methods of communication with one another. This ranges from direct speech to indirect body signals. As artificial intelligence progresses, it's imperative that they're developed with human aware architecture in order to appropriately interact with humans (Bellotti & Edwards).

Context
Context in human-computer dialog is the ability of the computer to interact with the user under certain pre-defined situations with some given input data. A user can click the backspace on a keyboard, and, depending on the context, the backspace may perform different functions (Winograd, 2011).

Context Awareness and Adaptive Systems
As artificial intelligence plays an increasingly larger role in the daily lives of humans, an AI must be equipped with the necessary tools to properly interact with humans and the environment (Bellotti & Keith, 2016).

Human Awareness
Humans are difficult to model due to their unpredictability. As such, an AI needs to be able to respond appropriately to causing discomfort or even injury to humans.
 * Action execution using a Hierarchical Task Network (HTN):
 * HTN use action costs to punish or reward certain defined AI actions. By incorporating these values to common social rules, an AI can learn to become more aware of its actions amongst humans (Alami et al., 2013).

Environmental Awareness
Computers, once occupying large rooms, are now both more powerful and more portable. A challenge of this portability is to create an environmentally aware computer system. By equipping mobile devices with GPS, gyroscopes, light sensors, and accelerometers, the computer system is able to deal with changes in environmental context (Ohlin, 2012).

Human Physiology
Computer systems can benefit from integrating human physiological traits as data within their context. For example, modern computer systems have begun incorporation human speech to input commands. Another proposed use is to have eye movements and pupil dilation to adjust a screen's brightness (Shye et al.).
 * Proposed implementation by using a User Management Unit (UMU): This system monitors user feedback by processing user physiological data to the computer system. If a user steps away from the screen or is no longer looking at the screen, the UMU can process this information and decide to dim or turn off the screen in order to be more energy efficient (Shye et al.).

In Practice
Human aware architecture extends into many fields, including robotics and multimedia.

Robotics
Presently, robotics has played a role in phasing out manual labor tasks. Amazon is reported by Bloomberg Technology to have 30,000 robots across their warehouses. These robots transport large shelves of goods throughout the warehouse, minimizing the need for human-workers to move about (Bhasin & Clark, June 2016).

Video Games
Game studio, Valve, uses the player's estimated arousal, to modify gameplay patterns of  Left 4 Dead in an attempt maximize the player's enjoyment.
 * - Other tested features: Portal 2, another popular game from Valve, used an eyetracker to aim, and hand movements to move.
 * - Potential uses: Heart rate as a measure of excitement and adrenaline. Other potential uses include game matchmaking, where similar players may be paired together based on their passiveness (Ambinder, 2011).

Electronics
Mobile devices, today, are equipped with many features to handle changes in context. These features include the GPS, gyroscope, light sensor, and accelorometer (Ohlin, 2012).

References (Human Aware Architectural & Urban Design)

















 * Ambinder, Mark. (2011). "Biofeedback in Gameplay: How Valve Measures Physiology to Enhance Gaming Experience". Valve Software Publications, 2011.


 * Bhasin, Kim., & Clark, Patrick. (June 2016). "How Amazon Triggered a Robot Arms Race". https://www.bloomberg.com/news/articles/2016-06-29/how-amazon-triggered-a-robot-arms-race.