Affective Artificial Intelligence

Synopsis

Affective Artificial Intelligence (Affective AI) aims to imbue machines with social and emotional intelligence (EQ). More specifically, Affective AI aims to create artificially intelligent systems and machines that can recognize, interpret, process, and simulate human social signals and behaviours, expressions, and emotions, to enhance human-AI interaction and communication. 

To achieve this goal, Affective AI draws upon various scientific disciplines, including machine learning, computer vision, speech / natural language / signal processing, psychology and cognitive science, and ethics and social sciences.

 

Background & Aims

Affective Artificial Intelligence (Affective AI) aims to imbue machines with social and emotional intelligence (EQ). More specifically, Affective AI aims to create artificially intelligent systems and machines that can recognize, interpret, process, and simulate human social signals and behaviours, expressions, and emotions, to enhance human-AI interaction and communication. 

To achieve this goal, Affective AI draws upon various scientific disciplines, including machine learning, computer vision, speech / natural language / signal processing, psychology and cognitive science, and ethics and social sciences.

Affective AI has direct applications in and implications for the design of innovative interactive technology (e.g., interaction with chat bots, virtual agents, robots), single and multi-user smart environments ( e.g., in-car/ virtual / augmented / mixed reality, serious games), public speaking and cognitive training, and clinical and biomedical studies (e.g., autism, depression, pain).

The aim of this module is to impart knowledge and ability needed to make informed choices of models, data, and machine learning techniques for sensing, recognition, and generation of affective and social behaviour (e.g., smile, frown, head nodding/shaking, agreement/disagreement) in order to create Affectively intelligent AI systems, with a consideration for various ethical issues (e.g., privacy, bias) arising from the real-world deployment of these systems.

 

Syllabus

The following list provides a representative list of topics:

• Introduction, definitions, and overview
• Theories from various disciplines
• Sensing from multiple modalities (e.g., vision, audio, bio signals, text)
• Data acquisition and annotation
• Signal processing / feature extraction
• Learning / prediction / recognition and evaluation
• Behaviour synthesis / generation (e.g., for embodied agents / robots)
• Advanced topics (e.g., continual learning)
• Ethical considerations (e.g., fairness, bias and fairness)
• Cross-disciplinary applications (via seminar presentations and discussions)
• Guest lectures (diverse topics – changes each year)

• Hands-on research and programming work (i.e., mini project & report) 

   

Objectives

On completion of this module, students will:

1. understand and demonstrate knowledge in key characteristics of affectively intelligent AI, which include:
   • Recognition: How to equip Affective AI systems with capabilities of analysing facial expressions, vocal intonations, gestures, and other physiological signals to infer human affective states?
   • Generation: How to enable affectively intelligent AI simulate expressions of emotions in machines, allowing them to respond in a more human-like manner, such as virtual agents and humanoid robots, showing empathy or sympathy?
   • Adaptation and Personalization: How to enable affectively intelligent AI adapt system responses based on users' affective states and/or needs, or tailor interactions and experiences to individual users based on their expressivity / emotional profiles, personalities, and past interactions?
   • Empathetic Communication: How to design Affective AI systems to communicate with users in a way that demonstrates empathy, understanding, and sensitivity to their affective states and needs?
   • Ethical and societal considerations: What are the various human differences, ethical guidelines, and societal impacts that need to be considered when designing and deploying Affective AI to ensure that these systems respect users' privacy, autonomy, and well-being?
2. comprehend and apply (appropriate) methods for collection, analysis, representation, and evaluation of human affective and communicative behavioural data.
3. enhance programming skills for creating and implementing (components of) Affective AI systems.
4. demonstrate critical thinking, analysis and synthesis while deciding on 'when' and 'how' to incorporate human affect and social signals in a specific AI system context and gain practical experience in proposing and justifying computational solution(s) of suitable nature and scope.

 

Assessment - Part II Students

• Seminar presentation (individual): 20%
• Discussant activity (individual): 10%
• Project proposal (as a team of 2): 5% 
• Final mini project report & code (as a team of 2): 65%

The assessments listed for this module may be subject to change following the outcome of the consultation on the addition of vivas to Module assessment modes.

Self-certification for this module may not be permitted.  This matter will be considered at the meeting of the Faculty Board for Computer Science and Technology on Tuesday 13 October and the outcome will be communicated to students after the meeting.

 

Further Information

This module is shared with ACS. Assessment will be adjusted for the two groups of students to be at an appropriate level for whichever course the student is enrolled on. More information will follow at the first lecture.

Assessment - MPhil / Part III Students Breakdown

• Seminar presentation: 20%
• Discussion Activity: 10%
• Project Proposal: 5% 
• Final mini project report & code: 65%

Seminar & Discussion Session: Information & Assessment

The goal of the seminar session is to learn to understand, evaluate and discuss research papers in the field of affective artificial intelligence, and present a scientific paper to peers with an appropriate level of description and explanation.

Each research paper will be presented in the following manner:

• One (pre-assigned) student presents.
• Another (pre-assigned) student acts as an assigned discussant.
• The discussant prepares in advance and provides, on the day after the corresponding presentation, a formal critical response.
• Then the presenter replies.

Each seminar session will therefore consist of:

1. Student Presentations

   MPhil / Part III students:

   Each student will be assigned one paper to present during the term. Each student will prepare and deliver a 7-minute presentation in class (the specific date will be announced during the lectures), and engage in a scholarly discussion with the pre-assigned discussant after their presentation (5 minutes).

   Part II students:

   Each student will be assigned one paper to present during the term. Each student will prepare and deliver a 4-minute presentation in class (the specific date will be announced during the lectures), and engage in a scholarly discussion with the pre-assigned discussant after their presentation (3 minutes).

2. Assigned Scholarly Discussant

   Each student will be assigned a research paper for which they will act as a scholarly discussant. The discussant role is intended to support inclusive and equitable scholarly engagement by emphasizing prepared analytical response rather than spontaneous classroom participation.

   The role of the discussant is to provide constructive scholarly critique by asking the presenter questions about the research paper and engaging in scholarly discussion.

   The discussant may focus on various aspects of the presented work, including but not limited to:

   • Research question(s)
   • Theoretical framing
   • Methodology
   • Assumptions
   • Evidentiary strength
   • Limitations
   • Alternative interpretations
   • Ethical considerations
   • Future directions

Assessment

Please ensure that you submit the PDF file of your presentation on Moodle so that your seminar marks can be entered into the system. Your submission must include the Honesty Code Coversheet as the first slide. See the end of this document for details on how to include this.

The assessment has two parts:

• The seminar presentation contributes 20% of the marks for the L344 Affective Artificial Intelligence course.
• The assigned scholarly discussant role contributes 10% of the marks for the L344 Affective Artificial Intelligence course.

Each seminar presentation will be assessed based on the following criteria:

• Understanding of the aims of the topic.
• Placement of ideas in the context of the literature.
• Critical analysis of findings, strengths and weaknesses.
• Preparation, depth and breadth of presentation.
• Organisation of ideas and topics.
• Clarity in communicating ideas and concepts.
• Use of visual aids, such as slides, graphics, animations and illustrations.
• Appropriate level of explanation when addressing lecturers and peers.
• Ability to drive the discussion and answer questions.
• Ability to keep to time.

Each discussant engagement and discussion will be assessed based on the following criteria:

• Depth of engagement.
• Quality of critique.
• Fairness and professionalism.
• Evidence-based reasoning.
• Understanding of research methods and literature.

Further Information

Current Cambridge undergraduate students who are continuing onto Part III or the MPhil in Advanced Computer Science may only take this module if they did NOT take it as a Unit of Assessment in Part II.

This module is shared with ACS. Assessment will be adjusted for the two groups of students to be at an appropriate level for whichever course the student is enrolled on. More information will follow at the first lecture.