AI research must focus on theory, practice, and how to make machines more intelligent. AI Perspectives & Advances (AIPP) publishes high-quality work that aims to address key pieces of the “AI puzzle” –fundamental topics of the field that beg for a breakthrough to move AI forward– and practical applications thereof that are poised to inspire and define the next generation of AI-based automation. On the applied front we want to see representatives from all major industrial, societal, and governmental operations, but especially forward-looking operational prototypes tested in industry settings and lab mockups of practical deployment. In basic research we seek submissions on any and all topics related to cognition and relevant computational methods across the entire research focus spectrum, especially work that aims to bridge the gap between existing work in, and leap beyond, various subfields of AI. Defining clearly the questions that we seek to answer, and supporting the selection with arguments from more than one perspective, should be fundamental in every research field. The same must hold for the methods chosen to answer them. Any methodology that is fruitful should be considered good. We welcome any papers proposing or using unified theories of cognition and AI research methodology, separately or in relation to other topics.
Topics of interest include, but are not limited to:
Learning & Knowledge Representation
Experimental, constructivist & cumulative learning
- how actionable knowledge can be built incrementally from ongoing information gathering
- how an intelligent agent can use systematic experimentation to create new knowledge
Analogy-based learning
- how analogies at various level of detail can allow an intelligent agent to learn faster by creating useful hypotheses about novel phenomena
Representation of cause-effect relations
- how an intelligent agent can model the world based on experience
- the relation between cause-effect relations and grounded knowledge
Concept-based knowledge representation
- what role do concepts play in cognition and how they are represented
- language use and language development
Understanding
- what kinds of models enable understanding and how does it relate to reasoning
- how a machine can automatically improve its understanding of its environment
- what role do concepts play in cognition and how they are represented
- what is the relation between understanding and ‘common sense’
Reasoning & Generality
Reasoning for generality
- methods for generalizing (inducing) from limited examples, using existing knowledge to support cumulative learning
Effective general control
- giving systems a way to handle vast amounts of information and operate under insufficient knowledge and resources
Automatic explanation generation
- how a system can generate useful explanations of its actions, knowledge, and environment
Ampliative reasoning
- how an intelligent system can combine many types of reasoning to learn tasks, solve problems, and make plans
Ampliative control
- how an intelligent system can combine many types of control - reactive, predictive, reflective - in the process of learning, solving problems, and making plans
Generality & Cognitive Autonomy
Worlds with infinite complexity & Empirical (“real-world”) control
- methods for achieving generality and autonomy in worlds with infinite combinatorics and novel phenomena, like the physical world, as opposed to axiomatic tasks like board games
- approaches and perspectives on unification of multiple control strategies for achieving goals and making plans
Autonomous knowledge management
- how an intelligent system can incrementally and autonomously build actionable and expandable models from its ongoing experience in a self-supervised manner
Seed-programmed autonomy & learning
- what is sufficient for an intelligent agent to know at “birth” to allow it to bootstrap its learning (from a “seed”), without outside help
General-purpose planning
- enabling systems to learn plan-making in a domain-independent way
Autonomous goal generation & management
- methods for achieving autonomous generation and management of subgoals
- approaches to autonomously managing multiple goals, goal evaluation, conflicting goals, goal prioritization, goal modification and goal abandonment
Reflection
- how an intelligent control system can inspect its own operation, steer its development, and improve it over time through cumulative learning
Many of these topics are increasingly mentioned by leading AI researchers as being of central importance to the advancement of AI. Attempts to address them in a holistic way have largely been based on methodologies dating back to the middle of last century, before the advent of modern programming techniques. Among them are formal logic and artificial neural networks, both of which have well-known limitations for achieving general autonomy, reasoning, and learning. Examples include methods for. To create truly intelligent machines requires integration of these (and other) cognitive processes in a single unified system that is capable of cumulative learning, autonomous learning of cause-effect relations, automatic explanation generation, and empirical (“real-world”) reasoning. Such unification calls for new approaches consisting of a coherently matching theory, methodology, and cognitive architecture – the “wholly trinity of AI research”. The Journal of AI Perspectives & Advances is not limited to these topics, but welcomes any and all research addressing these, as well as other, topics of AI research.
