A Cognitive Mind-map Framework to Foster Trust (Extended Paper)

The explorative mind-map is a dynamic framework, that emerges automatically from the input, it gets. It is unlike a verificative modeling system where existing (human) thoughts are placed and connected together. In this regard, explorative mind-maps change their size continuously, being adaptive with connectionist cells inside; mind-maps process data input incrementally and offer lots of possibilities to interact with the user through an appropriate communication interface. With respect to a cognitive-motivated situation like a conversation between partners, mind-maps become interesting as they are able to process stimulating signals whenever they occur. If these signals are close to an own understanding of the world, then the conversational partner becomes automatically more trustful than if the signals do not or less match the own knowledge scheme. In this (position) paper, we therefore motivate explorative mindmaps as a cognitive engine and propose these as a decision support engine to foster trust. 1. Explorative Mind-maps The principles of explorative mind-maps Mi have already been described in [11], where we accent that mind-maps rely on the natural principle on sensations and the corresponding propagation of stimuli to a final destination. Indeed, explorative mind-maps share this principle through an associative architecture that incrementally processes accepted stimuli to a consistent informational structure. This is similar to the natural paradigm, but on contrast to a verificative processing of a user’s thoughts, the explorative mind-maps are built from the bottom up, meaning that their existence exclusively interdepend on incoming signals. Explorative mind-maps share a sub-symbolic architecture that is composed of interacting entity cells ei. As mentioned above for the natural principle, these cells foster on a processing of data streams and a stimulation/inhibitionprinciple of adjacent connections. The activation of such a connectionist architecture bases on a dynamic construction of cell structures during the processing of the input stream. In the stimulation phase, a stream data is stimulated and absorbed by receptor (input) cells ri, which decompose the Figure 1: Merge between the existing mind-map a) and a (newly) mini-network b) to an updated mind-map c). Entity cells e2 and e5 are higher activated; the connection in between has been learned, the activation increased as well. stream to its entities. For example, the text streams are decomposed into the word entities, transactional streams to item entities, and so on. Using filter cells fi, those receptor cells ri are inhibited that do not address a semantic interest. In the Mini-Network phase, the collection of entities, which occur at such a specific time-point, form a mini-network [6] with fully connected mini-network cells mi. The Mindmap Merger starts once the mini-network is established: in this phase, the mini-network is sent to the mind-map and is merged with the existing entity cells in the mindmap (initially, the mind-map is empty). In this regard, the specified merge references an action of mini-network and entity cells (mi and ei) that share a same representation to a unique entity cell. The activation status act(ei) of an entity cell is then increased, in case it has been merged. If two adjacent entity cells ek and el of the mind-map are activated by the mini-network at the same time, their connection weight ω(ek, el) is increased by a learn rate φ (Hebbian Learning). If two adjacent entity cells remain inactivated, their connection weight is decreased by a fragment ar X iv :0 90 8. 33 94 v1 [ cs .A I] 2 4 A ug 2 00 9 of φ. The association becomes forgotten if the connection weight is below the forget parameter σ. The mind-map may degenerate due to less intensive stimuli. The usage of memories depends on skeletons kj(t), which are higher-activated clumps of entity cells that occur temporally and that consist of entity cells sharing a strong activity. As for the natural example, these skeletons may be sent to memories: depending on its state and temporal eligibility, the mind-map may keep existing skeletons inside the shortterm memory (STM) or send it to the long-term memory (LTM). Whereas the short-term memory comprehends itself as a location on short-notice, the long-term memory is a place of long-duration that is only accessible to skeleton patterns being permanently present or recurring. In a final phase, during communication with the outside world, comparison between the actual state of the mind and the outside contents is realised. This is to be done either on demand the user explicitly sends a request or unsolicited. In so far, the consequence of the brief descriptions above is not only that explorative mind-maps, which are non-deterministic regarding its size, appearance, and communication. Also that explorative mind-maps stand for a life-cycle process that depends on the intensity of incoming stimulations and the activity inside the mind-map. Figure 2 presents a figurative demonstration of the mindmap. When the corresponding mini-networks arrive on the left side, they are already assigned to the receptor cells ri and filtered to fi cells. These mini-networks are entered into the outer area STM. Inside the STM, mini-networks are compared with the individual mind-map structures one by one. Note that these mind-maps are temporarily taken from the LTM (inner core) part. If the mini-network skeleton (kj(t)) supports the structure of any existing mind-map, the skeleton is properly mapped and merged into the mind-map itself. Here, depending on the strength and relevance, clumps of the entity cells (skeletons kj(t)) may be sent to the LTM from the STM and/or kicked out to the outer area. Once being in the outer area, they (the skeleton or its subset) are released from the mind-map. Currently, some implementations, which are described in [1], have been realised and tested. First, an intrusion detection system has been designed and implemented. The explorative mind-map is used as an internal communication center for circulating immune cells in order to exchange novel intruder-related patterns and news. A second system manages implicit and explicit user feedback (inside a search retrieval environment) by a backbone mind-map engine. In a third system, an author-centric graph system which is the mind-map for bibliographic entries layer has been implemented to demonstrate affinity between social communities. However, all these mind-map implementations allow an exploited sub-role within a larger software system, focusing on a given purpose. This is, in our opinion a nonoptimal view, because mind-maps satisfy the demand for Figure 2: The explorative Mind-maps with the STM (grey area) and LTM (inner area). a cognitive engine that may work independently and nondeterministically. 2. May we understand Explorative Mind-maps as an Engine of Trust? 2.1. Mind-maps meet the cognitive demands! Generally, cognition is commonly accepted as to be the process of thought. In the wide field of Artificial Intelligence, it refers to the representation of mental processes (thoughts) and functions that state intelligent entities. The acceptability of explorative mind-maps with these matters is that they have already demonstrated cognitive abilities, they are adaptive to the input and work continuously and incrementally (like humans and creatures). Explorative mind-maps disarm the claim for a real-time system by a sliding window of size k and the filter cells. It also address the complexity problem by a dynamic and flexible association affectation. In general, each processing step is done in parallel while having a synchronization step. Mind-maps are associative, dynamic, and fault-tolerant; they share a hybrid structure with a sub-symbolic mind-map core and an user-directed interaction, which is not only related to a symbolic parameter adjustment but also to a verificative retrieve of existing mind-map information. And, whenever a mini-network is sent to the mind-map, some of the existing entity cells change their activation status and their activation to the neighbor entity cells through learning. Even though the existent explorative mind-maps are mostly deterministically organized (as the procedure of what to do is predetermined), we want to find an answer to the question on how to integrate advanced cognitive performances like for example trust. With trust, we understand the assumption that future changes will have a positive and expected progression. Furthermore, we understand trust as the result of an existing reliable, authentic and confident status, which follows from a representation of the current internal knowledge (belief), the state of personal emotion and associated interest. As mention in [2] and [3], the individual decision (to trust or not) is additionally governed by a host of factors like past experiences, history of events, others’ opinion and socio-cognitive behavior. 2.2. When Alice converses with Bob The trust between two persons is considered as the measure of a mutual belief. How to define the appropriate trust opinion during a conversation is a matter of concern. Assume that two natural persons Alice and Bob talk to each other where both may decide if an incoming conversational signal is considered as worth to remember or even not. Also if textual patterns inside the conversational streams are to be extracted, summarized and kept in mind or even handled as noise. If both Alice and Bob store conversational signals inside a knowledge-based representation framework, then Alice will know something (or something more) about Bob and vice versa. After receiving the input signal from Bob, Alice surely develops some certain belief about Bob and his believes at least with respect to the subject of conversation (and vice versa). Moreover, we may follow the idea that both are able to match their own mind-maps (representation of what each of them thinks) against the mind-maps of the conversing partner: both may decide at any time if they trust, and if yes, up to which level.