Classical AI Research Articles

AI in Finance: Challenges, Techniques and Opportunities

AI in finance broadly refers to the applications of AI techniques in financial businesses. This area has attracted attention for decades with both classic and modern AI techniques applied to increasingly broader areas of finance, economy and society. In contrast to either discussing the problems, aspects and opportunities of finance that have benefited from specific AI techniques and in particular some new-generation AI and data science (AIDS) areas or reviewing the progress of applying specific techniques to resolving certain financial problems, this review offers a comprehensive and dense roadmap of the overwhelming challenges, techniques and opportunities of AI research in finance over the past decades. The landscapes and challenges of financial businesses and data are firstly outlined, followed by a comprehensive categorization and a dense overview of the decades of AI research in finance. We then structure and illustrate the data-driven analytics and learning of financial businesses and data. A comparison, criticism and discussion of classic vs. modern AI techniques for finance follows. Finally, the open issues and opportunities to address future AI-empowered finance and finance-motivated AI research are discussed.

Artificial Intelligence: What it Was, and What it Should Be?

Artificial Intelligence was embraced as an idea of simulating unique abilities of humans, such as thinking, self-improvement, and expressing their feelings using different languages. The idea of “Programs with Common Sense was the main and central goal of Classical AI; it was, mainly built around an internal, updatable cognitive model of the world. But, now almost all the proposed models and approaches lacked reasoning and cognitive models and have been transferred to be more data driven. In this paper, different approaches and techniques of AI are reviewed, specifying how these approaches strayed from the main goal of Classical AI, and emphasizing how to return to its main objective. Additionally, most of the terms and concepts used in this field such as Machine Learning, Neural Networks and Deep Learning are highlighted. Moreover, the relations among these terms are determined, trying to remove mysterious and ambiguities around them. The transition from the Classical AI to Neuro-Symbolic AI and the need for new Cognitive-based models are also explained and discussed.

Classical AI linguistic understanding and the insoluble Cartesian problem

This paper examines an insoluble Cartesian problem for classical AI, namely, how linguistic understanding involves knowledge and awareness of u’s meaning, a cognitive process that is irreducible to algorithms. As analyzed, Descartes’ view about reason and intelligence has paradoxically encouraged certain classical AI researchers to suppose that linguistic understanding suffices for machine intelligence. Several advocates of the Turing Test, for example, assume that linguistic understanding only comprises computational processes which can be recursively decomposed into algorithmic mechanisms. Against this background, in the first section, I explain Descartes’ view about language and mind. To show that Turing bites the bullet with his imitation game and in the second section I analyze this method to assess intelligence. Then, in the third section, I elaborate on Schank and Abelsons’ Script Applier Mechanism (SAM, hereby), which supposedly casts doubt on Descartes’ denial that machines can think. Finally, in the fourth section, I explore a challenge that any algorithmic decomposition of linguistic understanding faces. This challenge, I argue, is the core of the Cartesian problem: knowledge and awareness of meaning require a first-person viewpoint which is irreducible to the decomposition of algorithmic mechanisms.

On the simplicity of simple heuristics

Recent evidence suggests that the take-the-best heuristic—flagship of “fast and frugal heuristics” research program—might in fact not be as frugal as tallying, which is considered to be a more complex strategy. Characterizing a simple decision strategy has always seemed straightforward, and the debate around the simplicity of the take-the-best heuristic is mostly concerned with a proper specification of the heuristic. I argue that the predominate conceptions of “simplicity” and “frugality” need to be revised. To this end, a number of recent behavioral and neuroscientific results are discussed. The example of take-the-best heuristic serves as an entry point to a foundational debate on bounded agency. I argue that the fast and frugal heuristics needs to question some of its legacy from the classical AI research. For example, the assumption that the bottleneck of decision-making process is information processing due to its serial nature. These commitments are hard to reconcile with the modern neuroscientific view of a human decision-maker. In addition, I discuss an overlooked source of uncertainty, namely neural noise, and compare a generic heuristic model to a similar neural algorithm.

Strong Stubborn Set Pruning for Star-Topology Decoupled State Space Search

Analyzing reachability in large discrete transition systems is an important sub-problem in several areas of AI, and of CS in general. State space search is a basic method for conducting such an analysis. A wealth of techniques have been proposed to reduce the search space without affecting the existence of (optimal) solution paths. In particular, strong stubborn set (SSS) pruning is a prominent such method, analyzing action dependencies to prune commutative parts of the search space. We herein show how to apply this idea to star-topology decoupled state space search, a recent search reformulation method invented in the context of classical AI planning.
 Star-topology decoupled state space search, short decoupled search, addresses planning tasks where a single center component interacts with several leaf components. The search exploits a form of conditional independence arising in this setting: given a fixed path p of transitions by the center, the possible leaf moves compliant with p are independent across the leaves. Decoupled search thus searches over center paths only, maintaining the compliant paths for each leaf separately. This avoids the enumeration of combined states across leaves.
 Just like standard search, decoupled search is adversely affected by commutative parts of its search space. The adaptation of strong stubborn set pruning is challenging due to the more complex structure of the search space, and the resulting ways in which action dependencies may affect the search. We spell out how to address this challenge, designing optimality-preserving decoupled strong stubborn set (DSSS) pruning methods. We introduce a design for star topologies in full generality, as well as simpler design variants for the practically relevant fork and inverted fork special cases. We show that there are cases where DSSS pruning is exponentially more effective than both, decoupled search and SSS pruning, exhibiting true synergy where the whole is more than the sum of its parts. Empirically, DSSS pruning reliably inherits the best of its components, and sometimes outperforms both.

Extending Classical Planning with State Constraints: Heuristics and Search for Optimal Planning


 
 
 We present a principled way of extending a classical AI planning formalism with systems of state constraints, which relate - sometimes determine - the values of variables in each state traversed by the plan. This extension occupies an attractive middle ground between expressivity and complexity. It enables modelling a new range of problems, as well as formulating more efficient models of classical planning problems. An example of the former is planning-based control of networked physical systems - power networks, for example - in which a local, discrete control action can have global effects on continuous quantities, such as altering flows across the entire network. At the same time, our extension remains decidable as long as the satisfiability of sets of state constraints is decidable, including in the presence of numeric state variables, and we demonstrate that effective techniques for cost-optimal planning known in the classical setting - in particular, relaxation-based admissible heuristics - can be adapted to the extended formalism. In this paper, we apply our approach to constraints in the form of linear or non-linear equations over numeric state variables, but the approach is independent of the type of state constraints, as long as there exists a procedure that decides their consistency. The planner and the constraint solver interact through a well-defined, narrow interface, in which the solver requires no specialisation to the planning context.
 
 

Letter to Turing

This personal, yet scientific, letter to Alan Turing, reflects on Turing's personality in order to better understand his scientific quest. It then focuses on the impact of his work today. By joining human attitude and particular scientific method, Turing is able to “immerse himself” into the phenomena on which he works. This peculiar blend justifies the epistolary style. Turing makes himself a “human computer”, he lives the dramatic quest for an undetectable imitation of a man, a woman, a machine. He makes us see the continuous deformations of a material action/reaction/diffusion dynamics of hardware with no software. Each of these investigations opens the way to new scientific paths with major consequences for contemporary live and for knowledge. The uses and the effects of these investigations will be discussed: the passage from classical AI to today's neural nets, the relevance of non-linearity in biological dynamics, but also their abuses, such as the myth of a computational world, from a Turing-machine like universe to an encoded homunculus in the DNA. It is shown that these latter ideas, which are sometimes even made in Turing's name, contradict his views.

Star-topology decoupled state space search

State space search is a basic method for analyzing reachability in discrete transition systems. To tackle large compactly described transition systems – the state space explosion – a wealth of techniques (e.g., partial-order reduction) have been developed that reduce the search space without affecting the existence of (optimal) solution paths. Focusing on classical AI planning, where the compact description is in terms of a vector of state variables, an initial state, a goal condition, and a set of actions, we add another technique, that we baptize star-topology decoupling, into this arsenal. A star topology partitions the state variables into components so that a single center component directly interacts with several leaf components, but the leaves interact only via the center. Many applications explicitly come with such structure; any classical planning task can be viewed in this way by selecting the center as a subset of state variables separating connected leaf components.Our key observation is that, given such a star topology, the leaves are conditionally independent given the center, in the sense that, given a fixed path of transitions by the center, the possible center-compliant paths are independent across the leaves. Our decoupled search hence branches over center transitions only, and maintains the center-compliant paths for each leaf separately. As we show, this method has exponential separations to all previous search reduction techniques, i.e., examples where it results in exponentially less effort. One can, in principle, prune duplicates in a way so that the decoupled state space can never be larger than the original one. Standard search algorithms remain applicable using simple transformations. Our experiments exhibit large improvements on standard AI planning benchmarks with a pronounced star topology.1

BioPlan: Classical Planning with Crowd simulation

Crowd simulation for evacuation situations often assumes that all agents are trying to reach a single point within an environment. Although such an assumption is not entirely wrong, human agents often exhibit more complex behaviors, even if deviations from the standard behavior are not particularly frequent. Classical planning is far from the best way to achieve the minimal path or correct behavior for agents, but adds a deeper level of reasoning about complex goal-achievement and about actions that are more complex than simply moving about. In this paper, we describe a crowd simulation experiment that uses classical AI planning to enrich the behavior of the agents in the scenario. Using this approach, we can express not only the target destination of the agents, but also (sub)goals and path preferences.

Generation of Sokoban Stages using Recurrent Neural Networks

Puzzles and board games represent several important classes of AI problems, but also represent difficult complexity classes. In this paper, we propose a deep learning based alternative to train a neural network model to find solution states of the popular puzzle game Sokoban. The network trains against a classical solver that uses theorem proving as the oracle of valid and invalid games states, in a setup that is similar to the popular adversarial training framework. Using our approach, we have been able to verify the validity of a Sokoban puzzle up to an accuracy of 99% on the test set. We have also been able to train our network to generate the next possible state of the puzzle board up to an accuracy of 99% on the validation set. We hope that through this approach, a trained neural network will be able to replace human experts and classical rule-based AI in generating new instances and solutions for such games.

Interactive Cognitive Systems and Social Intelligence

Research on cognitive systems adopts the aims and assumptions of classical AI research, emphasizing the construction of intelligent agents that exhibit complex behavior. This article reviews the cognitive systems paradigm and two widely adopted hypotheses—physical symbol systems and heuristic search—that underpin it. The author introduces a third claim—the social cognition hypothesis—that states intelligence requires the ability to represent and reason about others’ mental states. The article also examines a number of computational artifacts, both historical and recent, that focus on interaction and exhibit this capacity. Examples include dialogue systems, synthetic experts, believable agents, intelligent tutors, interactive robots, and instructable game players. In closing, the author identifies issues in social cognition that deserve greater attention and poses challenges that can drive future research on interactive cognitive systems.

Multi-agent systems: theory and applications survey

Research in multi-agent systems MAS aims to understand how a set of autonomous entities called agents can organise themselves to solve problems, achieve tasks and produce global phenomena that agent cannot realise alone. The MAS represent a credible alternative to conventional approaches. Agent-based modelling used to represent computationally complex systems that can not be expressed easily in the form of mathematical equations. MAS draw its ideas from three areas: methods of logical reasoning of classical artificial intelligence AI, process management work of distributed AI and the behavioural approach of artificial life. The first applications of multi-agent simulation appeared in the mid-1980s, and gradually covered a variety of domains, ranging from social science to ethology, ecology, biology, traffic and transport and manufacturing. The goal of this paper is to explore clearly the agent concept and MAS, and furthermore to show the interest of the use and the applications of multi-agent simulation.

“Paradigms of AI Programming” in Python

Norvig’s (1992) Paradigms of AI Programming is an important book for learning about AI programming. However, the book uses Common Lisp as the programming language, which is less popular now than in 1992. Thus, we have translated many classical AI programs described in the book into Python, a more commonly used language. We have also documented the programs and offered them as a resource in a course on knowledge-based AI.

An architecture for observational learning and decision making based on internal models

We present a cognitive architecture whose main constituents are allowed to grow through a situated experience in the world. Such an architectural growth is bootstrapped from a minimal initial knowledge and the architecture itself is built around the biologically-inspired notion of internal models. The key idea, supported by findings in cognitive neuroscience, is that the same internal models used in overt goal-directed action execution can be covertly re-enacted in simulation to provide a unifying explanation to a number of apparently unrelated individual and social phenomena, such as state estimation, action and intention understanding, imitation learning and mindreading. Thus, rather than reasoning over abstract symbols, we rely on the biologically plausible processes firmly grounded in the actual sensorimotor experience of the agent. The article describes how such internal models are learned in the first place, either through individual experience or by observing and imitating other skilled agents, and how they are used in action planning and execution. Furthermore, we explain how the architecture continuously adapts its internal agency and how increasingly complex cognitive phenomena, such as continuous learning, prediction and anticipation, result from an interplay of simpler principles. We describe an early evaluation of our approach in a classical AI problem-solving domain: the Sokoban puzzle.

Introduction: Philosophy and Theory of Artificial Intelligence

The theory and philosophy of artificial intelligence has come to a crucial point where the agenda for the forthcoming years is in the air. This special volume of Minds and Machines presents leading invited papers from a conference on the ‘‘Philosophy and Theory of Artificial Intelligence’’ that was held in October 2011 in Thessaloniki (www.pt-ai.org). Artificial Intelligence is perhaps unique among engineering subjects in that it has raised very basic questions about the nature of computing, perception, reasoning, learning, language, action, interaction, consciousness, humankind, life etc. etc.— and at the same time it has contributed substantially to answering these questions (in fact, it is sometimes seen as a form of empirical research). There is thus a substantial tradition of work, both on AI by philosophers and of theory within AI itself. The classical theoretical debates have centred on the issues whether AI is possible at all (often put as ‘‘Can machines think?’’) or whether it can solve certain problems (‘‘Can a machine do x?’’). In the meantime, technical AI systems have progressed massively and are now present in many aspects of our environment. Despite this development, there is a sense that classical AI is inherently limited, and must be replaced by (or supplanted with) other methods, especially neural networks, embodied cognitive science, statistical methods, universal algorithms, emergence,

Anytime heuristic search in temporal HTN planning for developing incident action plans

Extreme events challenge emergency managers' decision making capabilities of developing unified action plans to ensure effective coordination for multiple responding agencies. In this paper, we present a novel HTN planner, called XEPlanner, aimed at working in dynamic emergency response environment with critical time constraints. By understanding incident action plans developing process and characteristics of flood controlling, this paper proposes a set of planning requirements as the guidance for designing a domain specific HTN planner, which challenge the current known HTN planners. Based on the classical AI planning technologies, our planner is designed and implemented for adapting to the application domain. It follows the anytime principles: the first feasible plan can be quickly produced and the quality of the plans is improved as more time is available. Additionally, a set of heuristics for selecting search nodes is proposed for taking into account preferences of emergency managers during the planning process. Moreover, the priorities of incident objectives can be handled effectively. Our experimental results demonstrate that our planner satisfies all the proposed planning requirements and has advantages when it's applied in flood evacuation domain.

DESCARTES: LAS INTUICIONES MODALES Y LA INTELIGENCIA ARTIFICIAL CLÁSICA

Descartes denies that any machine can be intelligent, since machines are predictable, inflexible and limited. Advocates of AI argue the contrary. Even so, both Descartes and classical AI envisage the possibility that thought and intelligence need not be instantiated by physical properties. Such a possibility is entertained by Descartes through a modal intuition according to which mind can exist without body, an idea which seems to have been endorsed by classical AI when it reduces mind to a Turing Machine whose physical realization is irrelevant. Even though both arguments presuppose different theories and consequences, Functionalism turns out to be compatible with a form of Dualism, which discards the Materialism that originally inspired classical Artificial Intelligence.

Computers and Thought—The Back Story

This issue of Anecdotes gives a some of the back story behind the classic computer science and AI text Computers and Thought. One of the book's coeditors, Julian Feldman, describes some of their reasons for publishing the book and some of the challenges involved in its actually publication. This Anecdotes installment also gives a brief history of the first business-to-business and business-to-consumer applications in the UK in the 1980s, which presaged today's Internet-based home shopping industry.

Distributed artificial intelligence, diversity and information literacy

My proposal is based on my doctoral dissertation On the Shoulders of AI-technology : Sociology of Knowledge and Strong Artificial Intelligence which I succesfully defended on May 29th 2009. E-published http://www.lu.se/o.o.i.s?id=12588&postid=1389611 The dissertation is concerned with Sociology’s stance in the debate on Strong Artificial Intelligence,.i.e. AI-systems that is able to shape knowledge on their own. There is a need for sociologists to realize the difference between two approaches to constructing AI systems: Symbolic AI (or Classic AI) and Connectionistic AI in a distributed model – DAI. Sociological literature shows a largely critical attitude towards Symbolic AI, an attitude that is justified. The main theme of the dissertation is that DAI is not only compatible with Sociology’s approach to what is social, but also constitutes an apt model of how a social system functions. This is consolidated with help from german sociologist Niklas Luhmann’s social systems theory. A lot of sociologists criticize AI because they think that diversity is important and can only be comprehended in informal circumstances that only humans interacting together can handle. They mean that social intelligence is needed to make something out of diversity and informalism. Luhmann´s systems theory gives the opposite perspective. It tells us that it is social systems that communicate and produce new knowledge structures out of contincency. Psychological systems, i.e. humans, can only think within the circumstances the social system offer. In that way human thoughts are bound by formalism. Diversity is constructed when the social systems interact with complexity in their environments. They reduce the complexity and try to present it as meaningful diversity. Today when most of academic literature is electronically stored and is accessible through the Internet from al over the world, DAI can help social systems to observe and reduce complexity in this global dimension. It is pointed out that human consciousness is limited in handling this global dimension. Therefore is it reasonable to argue that DAI in at least this dimension has a stronger intelligence than humans have. I will argue that Luhmann´s social theory and DAI give a god model to analyze the conditions for diversity in the Internet society. Further, the discussion about strong AI gives a lot of opportunities to discuss what sort of information literacy is needed and it also gives some perspective to discuss the concept of IL I have observed that the concept has evolved from something that coined some formal capacities, to something that has to do with a capacity to observe informal relations. That discussion can easily be compared to a parallel discussion within the debate about strong AI.

NATURE–INSPIRED INTELLIGENCE: A REVIEW OF SELECTED METHODS AND APPLICATIONS

The successful handling of numerous real–world complex problems has increased the popularity of nature–inspired intelligent (NII) algorithms and techniques. Their successful implementation primarily on difficult and complicated optimization problems, stresses their upcoming importance in the broader area of artificial intelligence. NII techniques take advantage of the way that biological systems deal with real–world situations. Specifically, they simulate the way real biological systems, such as the human brain, ant colonies and human immune system work, when solving complex real–world situations. In this survey paper, we briefly present a number of selected NII approaches and we point particular suitable areas of application for each of them. Specifically, five major categories of nature inspired approaches are presented, namely, Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), DNA computing, artificial immune systems and membrane computing. Applications include problems related to optimization (financial, industrial and medical), task scheduling, system design (optimization of the system's parameters), image processing and data processing (feature selection and classification). We also refer to collaboration between NII techniques and classical AI methodologies, such as neural networks, genetic algorithms, fuzzy logic, etc. The current survey states that NII techniques are likely to become the next step in the rapid evolution of artificial intelligence tools.