Intelligent machines have reached capabilities that go beyond a level that a human being can fully comprehend without sufficiently detailed understanding of the underlying mechanisms. The choice of moves in the game Go (generated by Deep Mind?s Alpha Go Zero [1]) are an impressive example of an artificial intelligence system calculating results that even a human expert for the game can hardly retrace [2]. But this is, quite literally, a toy example. In reality, intelligent algorithms are encroaching more and more into our everyday lives, be it through algorithms that recommend products for us to buy, or whole systems such as driverless vehicles. We are delegating ever more aspects of our daily routines to machines, and this trend looks set to continue in the future. Indeed, continued economic growth is set to depend on it. The nature of human-computer interaction in the world that the digital transformation is creating will require (mutual) trust between humans and intelligent, or seemingly intelligent, machines. But what does it mean to trust an intelligent machine? How can trust be established between human societies and intelligent machines?

Designing and constructing pedagogical tests that contain items (i.e. questions) which measure various types of skills for different levels of students equitably is a challenging task. Teachers and item writers alike need to ensure that the quality of assessment materials is consistent, if student evaluations are to be objective and effective. Assessment quality and validity are therefore heavily reliant on the quality of the items included in the test. Moreover, the notion of difficulty is an essential factor that can determine the overall quality of the items and the resulting tests.Thus, item difficulty prediction is extremely important in any pedagogical learning environment. Although difficulty is traditionally estimated either by experts or through pre-testing, such methods are criticised for being costly, time-consuming, subjective and difficult to scale, and consequently, the use of automatic approaches as proxies for these traditional methods is gaining more and more traction. In this paper, we provide a comprehensive and systematic review of methods for the priori prediction of question difficulty. The aims of this review are to: 1) provide an overview of the research community regarding the publication landscape; 2) explore the use of automatic, text-based prediction models; 3) summarise influential difficulty features; and 4) examine the performance of the prediction models. Supervised machine learning prediction models were found to be mostly used to overcome the limitations of traditional item calibration methods. Moreover, linguistic features were found to play a major role in the determination of item difficulty levels, and several syntactic and semantic features were explored by researchers in this area to explain the difficulty of pedagogical assessments. Based on these findings, a number of challenges to the item difficulty prediction community are posed, including the need for a publicly available repository of standardised data-sets and further investigation into alternative feature elicitation and prediction models.

In recent years interface agents have been developed to assist users with various tasks. Some systems employ machine learning techniques to allow the agent to adapt to the user's changing requirements. With the increase in the volume of data on the Internet agents have emerged that are able to monitor and learn from their users to identify topics of interest. One such agent described here has been developed to filter mail messages. We examine the issues involved in constructing an autonomous interface agent that employs a learning component and explore the use of two different learning techniques in this context.

In order to support semantic interoperation in open environments, where agents can dynamically join or leave and no prior assumption can be made on the ontologies to align, the different agents involved need to agree on the semantics of the terms used during the interoperation. Reaching this agreement can only come through some sort of negotiation process. Indeed, agents will differ in the domain ontologies they commit to; and their perception of the world, and hence the choice of vocabulary used to represent concepts. We propose an approach for supporting the creation and exchange of different arguments, that support or reject possible correspondences. Each agent can decide, according to its preferences, whether to accept or refuse a candidate correspondence. The proposed framework considers arguments and propositions that are specific to the matching task and are based on the ontology semantics. This argumentation framework relies on a formal argument manipulation schema and on an encoding of the agents' preferences between particular kinds of arguments.

Multiagent systems evolved from a need for knowledge-aware, distributed, problem-solving mechanisms. These systems are formally grounded using theoretical approaches, including those that assume mentalistic notions. As a result, much of this research into multiagent systems has provided formal proofs or proof-of-concept demonstrators (such as example systems or prototypes). It has provided only limited, pragmatic support (systems, software, and tools) for the user community. Research into Web services, in contrast, has focused on the user community, resulting in a pragmatic, bottom-up enabling technology that readily facilitates the robust construction of service-oriented systems. Much of the focus of Web services research has been on developing declarative descriptions that application developers can share and that their tools can use to construct and develop large-scale distributed software. Despite these differing approaches, the inherent component-based structure underlying both agents and Web services raises questions about how exactly they differ and whether they can coexist.