Human-level Artificial Intelligence Research Articles

A "thoughtful" Local Friendliness no-go theorem: a prospective experiment with new assumptions to suit

A recent paper by two of us and co-workers \cite{CQD20}, based on an extended Wigner's friend scenario, demonstrated that certain empirical correlations predicted by quantum theory (QT) violate inequalities derived from a set of metaphysical assumptions we called "Local Friendliness" (LF). These assumptions are strictly weaker than those used for deriving Bell inequalities. Crucial to the theorem was the premise that a quantum system with reversible evolution could be an observer (colloquially, a "friend"). However, that paper was noncommittal on what would constitute an observer for the purpose of an experiment. Here, we present a new LF no-go theorem which takes seriously the idea that a system's having thoughts is a sufficient condition for it to be an observer. Our new derivation of the LF inequalities uses four metaphysical assumptions, three of which are thought-related, including one that is explicitly called "Friendliness". These four assumptions, in conjunction, allow one to derive LF inequalities for experiments involving the type of system that "Friendliness" refers to. In addition to these four metaphysical assumptions, this new no-go theorem requires two assumptions about what is technologically feasible: Human-Level Artificial Intelligence, and Universal Quantum Computing which is fast and large scale. The latter is often motivated by the belief that QT is universal, but this is not an assumption of the theorem. The intent of the new theorem is to give a clear goal for future experimentalists, and a clear motivation for trying to achieve that goal. We review various approaches to QT in light of our theorem. The popular stance that "quantum theory needs no interpretation" does not question any of our assumptions and so is ruled out. Finally, we quantitatively discuss how difficult the experiment we envisage would be, and briefly discuss milestones on the paths towards it.

Everything and More: The Prospects of Whole Brain Emulation

Whole Brain Emulation (WBE) has been championed as the most promising, well-defined route to achieving both human-level artificial intelligence and superintelligence. It has even been touted as a viable route to achieving immortality through brain uploading. WBE is not a fringe theory: the doctrine of Computationalism in philosophy of mind lends credence to the in-principle feasibility of the idea, and the standing of the Human Connectome Project makes it appear to be feasible in practice. Computationalism is a popular, independently plausible theory, and Connectomics a well-funded empirical research program, so optimism about WBE is understandable. However, this optimism may be misplaced. This article argues that WBE is, at best, no more compelling than any of the other far-flung routes to achieving superintelligence. Similarly skeptical conclusions are found regarding immortality. The essay concludes with some positive considerations in favor of the Biological Theory of consciousness, as well as morals about the limits of Computationalism in both artificial intelligence and the philosophy of mind more generally.

Key Advanced Research Initiative: A Manifesto for the New-Generation Artificial Intelligence

The goal here is to identify key directions for the future advanced research initiatives in Artificial Intelligence (AI) and beyond. The following areas are identified as having particular importance: (1) socially emotional, ethical, and moral AI, (2) self-developing and self-sustainable AI, and (3) human-analogous AI, inspired by the human psychology. As a result, a general concept is formulated with the intent to clarify and unify the currently popular slogans, including Artificial General Intelligence (AGI), Strong AI, Human-Level or Humanlike AI (HLAI), Brain-Inspired or Biologically Inspired Cognitive Architectures (BICA), and more. The key idea of the proposed concept is that future AI must open a new angle of view and new perspectives to humans, thereby enriching and transforming the society, helping it to solve its problems and taking the civilization to a new level. While being created by humans, for humans, and fully compatible with humans at the social level, it will not be “a human in silicon”, but rather an “alien”: intelligent, friendly, and welcome. Its principles will combine preprogrammed basic functions and its own natural ontogeny in a virtual social environment. Forms of implementation will range from virtual entities to wearable electronics and autonomous robots. The expected impact on the society will be immense and crucial for its survival.

Imago Dei in the Age of Artificial Intelligence: Challenges and Opportunities for a Science-Engaged Theology

Modern developments in evolutionary and cognitive science have increasingly challenged the view that humans are distinctive creatures. In theological anthropology, this view is germane to the doctrine of the image of God. To address these challenges, imago Dei theology has shifted from substantial toward functional and relational interpretations: the image of God is manifested in our divine mandate to rule the world, or in the unique personal relationships we have with God and with each other. If computers ever attain human-level Artificial Intelligence, such imago Dei interpretations could be seriously contested. This article reviews the recent shifts in theological anthropology and reflects theologically on the questions raised by the potential scenario of human-level AI. It argues that a positive outcome of this interdisciplinary dialogue is possible: theological anthropology has much to gain from engaging with AI. Comparing ourselves to intelligent machines, far from endangering our uniqueness, might instead lead to a better understanding of what makes humans genuinely distinctive and in the image of God.

After 40 years of search for the promised land BICA*AI designers come to a point of disillusion and discord. Why? For what sins and misdeeds?

The ultimate goal of research in cognitive architectures is to model the human mind, eventually enabling the recreation of human-level artificial intelligence. However current wave of research in cognitive architectures is centered only on the study of human mind modeling occupied with the human behavior and the underlying cognitive mechanisms necessary for implementation of this behavior. In doing so, human mind research totally derails its own ability to study other operating mechanisms of biological brains, such as how the brain processes information, evaluates its semantics, makes decisions and interacts with the surrounding world. The prime intention of this short comment on the current trends in the human brain research is to draw people's attention to the harmful consequences of some of such trends.

The transformative potential of artificial intelligence

The terms ‘human-level artificial intelligence’ and ‘artificial general intelligence’ are widely used to refer to the possibility of advanced artificial intelligence (AI) with potentially extreme impacts on society. These terms are poorly defined and do not necessarily indicate what is most important with respect to future societal impacts. We suggest that the term ‘transformative AI’ is a helpful alternative, reflecting the possibility that advanced AI systems could have very large impacts on society without reaching human-level cognitive abilities. To be most useful, however, more analysis of what it means for AI to be ‘transformative’ is needed. In this paper, we propose three different levels on which AI might be said to be transformative, associated with different levels of societal change. We suggest that these distinctions would improve conversations between policy makers and decision makers concerning the mid- to long-term impacts of advances in AI. Further, we feel this would have a positive effect on strategic foresight efforts involving advanced AI, which we expect to illuminate paths to alternative futures. We conclude with a discussion of the benefits of our new framework and by highlighting directions for future work in this area.

NeuroLISP: High-level symbolic programming with attractor neural networks

Despite significant improvements in contemporary machine learning, symbolic methods currently outperform artificial neural networks on tasks that involve compositional reasoning, such as goal-directed planning and logical inference. This illustrates a computational explanatory gap between cognitive and neurocomputational algorithms that obscures the neurobiological mechanisms underlying cognition and impedes progress toward human-level artificial intelligence. Because of the strong relationship between cognition and working memory control, we suggest that the cognitive abilities of contemporary neural networks are limited by biologically-implausible working memory systems that rely on persistent activity maintenance and/or temporal nonlocality. Here we present NeuroLISP, an attractor neural network that can represent and execute programs written in the LISP programming language. Unlike previous approaches to high-level programming with neural networks, NeuroLISP features a temporally-local working memory based on itinerant attractor dynamics, top-down gating, and fast associative learning, and implements several high-level programming constructs such as compositional data structures, scoped variable binding, and the ability to manipulate and execute programmatic expressions in working memory (i.e., programs can be treated as data). Our computational experiments demonstrate the correctness of the NeuroLISP interpreter, and show that it can learn non-trivial programs that manipulate complex derived data structures (multiway trees), perform compositional string manipulation operations (PCFG SET task), and implement high-level symbolic AI algorithms (first-order unification). We conclude that NeuroLISP is an effective neurocognitive controller that can replace the symbolic components of hybrid models, and serves as a proof of concept for further development of high-level symbolic programming in neural networks.

Compositional memory in attractor neural networks with one-step learning

Compositionality refers to the ability of an intelligent system to construct models out of reusable parts. This is critical for the productivity and generalization of human reasoning, and is considered a necessary ingredient for human-level artificial intelligence. While traditional symbolic methods have proven effective for modeling compositionality, artificial neural networks struggle to learn systematic rules for encoding generalizable structured models. We suggest that this is due in part to short-term memory that is based on persistent maintenance of activity patterns without fast weight changes. We present a recurrent neural network that encodes structured representations as systems of contextually-gated dynamical attractors called attractor graphs. This network implements a functionally compositional working memory that is manipulated using top-down gating and fast local learning. We evaluate this approach with empirical experiments on storage and retrieval of graph-based data structures, as well as an automated hierarchical planning task. Our results demonstrate that compositional structures can be stored in and retrieved from neural working memory without persistent maintenance of multiple activity patterns. Further, memory capacity is improved by the use of a fast store-erase learning rule that permits controlled erasure and mutation of previously learned associations. We conclude that the combination of top-down gating and fast associative learning provides recurrent neural networks with a robust functional mechanism for compositional working memory.

Cognitive Vulnerability, Artificial Intelligence, and the Image of God in Humans

Recent progress in artificial intelligence (AI) opens up the possibility that one day machines could do anything that a human being can do, raising thus serious questions regarding human distinctiveness. For theological anthropology, the prospect of human-level AI brings a fresh opportunity to clarify the definition of the image of God. Comparing human and artificial intelligence leads to replacing the Aristotelian-like interpretation of the image of God as rationality with a relational model. Instead of regarding our cognitive biases as vulnerabilities, they should be seen as instrumental in bringing about our unique type of intelligence, one marked by relationality.

Hybrid Bionic Cognitive Architecture for Artificial General Intelligence Agents

The article describes the author’s proposal on cognitive architecture for the development of a general-level artificial intelligent agent («strong» artificial intelligence). New principles for the development of such an architecture are proposed — a hybrid approach in artificial intelligence and bionics. The architecture diagram of the proposed solution is given and descriptions of possible areas of application are described. Strong artificial intelligence is a technical solution that can solve arbitrary cognitive tasks available to humans (human-level artificial intelligence) and even surpass the capabilities of human intelligence (artificial superintelligence). The fields of application of strong artificial intelligence are limitless — from solving current problems facing the human to completely new problems that are not yet available to human civilization or are still waiting for their discoverer. The novelty of the work lies in the author’s approach to the construction of cognitive architecture, which has absorbed the results of many years of research in the field of artificial intelligence and the results of the analysis of cognitive architectures of other researchers.

Compositional Generalization via Parsing Tree Annotation

Humans can understand a novel sentence by parsing it into known components like phrases and clauses. To achieve human-level artificial intelligence, compositional generalization tasks are suggested and used to assess machine learning models. Among those tasks, the SCAN tasks are challenging for the standard deep learning models, such as RNN sequence-to-sequence models and Transformers, that show great success across many natural language processing tasks. Even though a long line of deep learning research has developed memory augmented neural networks aimed at the SCAN tasks, their generalities remain questionable for more complex and realistic applications where the standard seq2seq models dominate. Hence, one needs to propose a method that helps the standard models to discover compositional rules. To this end, we propose a data augmentation technique using paring trees. Our technique annotates targets by inserting a new delimiter token in between them according to their parsing trees. For the training stage, the technique needs prior knowledge about the targets' semantic or syntactic compositionality. On the other hand, for the test stage, the technique uses no such knowledge. Experiments show that our technique enables the standard models to achieve compositional generalization on the SCAN tasks. Furthermore, we validate our technique on a synthetic task and confirm the standard models' strong performance gains without using prior knowledge about semantic compositionality. As one way to infuse parsing tree information into sequences, our technique can be used for tasks with structured targets like program code generation tasks.

On achieving human-level knowledge representation by developing a natural language of thought

What is the nature of knowledge representation needed for human-level artificial intelligence? This position paper contends that to achieve human-level AI, a system architecture for human-level knowledge representation would benefit from a neuro-symbolic approach combining deep neural networks with a ‘natural language of thought’ and would be greatly handicapped if relying only on formal logic systems.

Artificial Interdisciplinarity: Artificial Intelligence for Research on Complex Societal Problems

This paper considers the question: In what ways can artificial intelligence assist with interdisciplinary research for addressing complex societal problems and advancing the social good? Problems such as environmental protection, public health, and emerging technology governance do not fit neatly within traditional academic disciplines and therefore require an interdisciplinary approach. However, interdisciplinary research poses large cognitive challenges for human researchers that go beyond the substantial challenges of narrow disciplinary research. The challenges include epistemic divides between disciplines, the massive bodies of relevant literature, the peer review of work that integrates an eclectic mix of topics, and the transfer of interdisciplinary research insights from one problem to another. Artificial interdisciplinarity already helps with these challenges via search engines, recommendation engines, and automated content analysis. Future “strong artificial interdisciplinarity” based on human-level artificial general intelligence could excel at interdisciplinary research, but it may take a long time to develop and could pose major safety and ethical issues. Therefore, there is an important role for intermediate-term artificial interdisciplinarity systems that could make major contributions to addressing societal problems without the concerns associated with artificial general intelligence.

Ontology Reasoning with Deep Neural Networks


 
 
 The ability to conduct logical reasoning is a fundamental aspect of intelligent human behavior, and thus an important problem along the way to human-level artificial intelligence. Traditionally, logic-based symbolic methods from the field of knowledge representation and reasoning have been used to equip agents with capabilities that resemble human logical reasoning qualities. More recently, however, there has been an increasing interest in using machine learning rather than logic-based symbolic formalisms to tackle these tasks. In this paper, we employ state-of-the-art methods for training deep neural networks to devise a novel model that is able to learn how to effectively perform logical reasoning in the form of basic ontology reasoning. This is an important and at the same time very natural logical reasoning task, which is why the presented approach is applicable to a plethora of important real-world problems. We present the outcomes of several experiments, which show that our model is able to learn to perform highly accurate ontology reasoning on very large, diverse, and challenging benchmarks. Furthermore, it turned out that the suggested approach suffers much less from different obstacles that prohibit logic-based symbolic reasoning, and, at the same time, is surprisingly plausible from a biological point of view.
 
 

Deep learning and the sociology of human-level artificial intelligence

Deep learning and the sociology of human-level artificial intelligence

Understanding understanding and ambiguity in natural language

Abstract How can we define and understand the nature of understanding itself? This paper discusses cognitive processes for understanding the world in general and for understanding natural language. The discussion considers whether and how an artificial cognitive system could use a ‘natural language of thought’, and whether the ambiguities of natural language would be a theoretical barrier or could be a theoretical advantage for such a system, in a research approach toward human-level artificial intelligence.

Toward metascience via human-level AI with metacognition

Abstract Rosenbloom gave reasons why Computing should be considered as a fourth great domain of science, along with the Physical sciences, Life sciences, and Social sciences. This paper considers Metascience as the future, fifth great domain of science, and discusses reasons why metascience may be closely related to metacognition in human intelligence and human-level artificial intelligence, suggesting that the representation and processing which could support an AI system’s metacognition could also support an AI system reasoning metascientifically about domains of science.

AnswerNet: Learning to Answer Questions

Multi-modal tasks like visual question answering (VQA) are an important step towards human-level artificial intelligence. In general, the input of the VQA task consists of an image and a related question. In order to correctly answer the question, a model needs to extract and integrate useful information from both the image and the question. In this paper, we propose a model named AnswerNet to tackle this task. In the proposed model, discriminative features are extracted from both the image and the question. Specifically, high-level image features are extracted by the state-of-the-art convolutional neural network, i.e., Deep Residual Net. For question features, the semantic representations of the question and the term frequencies of the distinct words are captured by long short-term memory network and bag-of-words model, respectively. Then, a hierarchical fusion network is proposed to effectively fuse the image features with the question features. Experimental results on three large-scale datasets, VQA, COCO-QA, and VQA2, demonstrate the effectiveness of the proposed AnswerNet.

Biologically Driven Artificial Intelligence

Artificial Intelligence (AI) based on the computational principles of the brain can overcome current shortcomings and lead to human-level AI.

A Survey on Computational Models for Learning and Memory in Artificial Intelligence

Learning and Memory is the important area of artificial Intelligence to study about the main functionality of human brain to develop a new computational models and algorithms based on cognitive learning and memory. Therefore, here human intelligence encourages a review for function of learning and memory. This paper clarify that without learning the aim of accomplish human-level artificial intelligence is at a great distance from completion. This review paper covers the human learning and memory, hippocampus learning and hidden markov model. This review paper covers mostly work done in the area of learning and memory. In this review we focuses on definitions of learning and memory and development and theories of learning over the past few years . This paper also presents advantages and disadvantages of various computational models and algorithms related to learning and memory.