S. A. Rodionov

Sergey Rodionov

We evaluate an initial coding-agent system for ARC-AGI-3 in which the agent maintains an executable Python world model, verifies it against previous observations, refactors it toward simpler abstractions as a practical proxy for an MDL-like simplicity bias, and plans through the model before acting. The system is intentionally direct: it uses a scripted controller, predefined world-model interfaces, verifier programs, and a plan executor, but no hand-coded game-specific logic. We report results on the 25 public ARC-AGI-3 games. Each recorded playthrough uses a fresh agent instance with no access to previous playthrough-specific files or conversation state. Most games have a single recorded playthrough; for a few games, we report multiple independent fresh-agent playthroughs to expose run-to-run variability. The agent fully solved 7 games, achieved a Relative Human Action Efficiency greater than 75%, on 6 games, and obtained a mean per-game RHAE of 32.58%. Because the system uses no game-specific code, it can serve as a game-general baseline for ARC-AGI-3. Performance on the private validation set remains to be tested. Overall, the results provide preliminary evidence that verifier-driven executable world models are a promising approach for ARC-AGI-3 agents.

Sergey Rodionov, Zarathustra Amadeus Goertzel, Ben Goertzel

This report summarizes a short study of the performance of GPT-4 on the ETHICS dataset. The ETHICS dataset consists of five sub-datasets covering different fields of ethics: Justice, Deontology, Virtue Ethics, Utilitarianism, and Commonsense Ethics. The moral judgments were curated so as to have a high degree of agreement with the aim of representing shared human values rather than moral dilemmas. GPT-4's performance is much better than that of previous models and suggests that learning to work with common human values is not the hard problem for AI ethics.

Alexey Potapov, Sergey Rodionov

Universal induction relies on some general search procedure that is doomed to be inefficient. One possibility to achieve both generality and efficiency is to specialize this procedure w.r.t. any given narrow task. However, complete specialization that implies direct mapping from the task parameters to solutions (discriminative models) without search is not always possible. In this paper, partial specialization of general search is considered in the form of genetic algorithms (GAs) with a specialized crossover operator. We perform a feasibility study of this idea implementing such an operator in the form of a deep feedforward neural network. GAs with trainable crossover operators are compared with the result of complete specialization, which is also represented as a deep neural network. Experimental results show that specialized GAs can be more efficient than both general GAs and discriminative models.

Alexey Potapov, Sergey Rodionov, Hugo Latapie et al.

Cross-dataset transfer learning is an important problem in person re-identification (Re-ID). Unfortunately, not too many deep transfer Re-ID models exist for realistic settings of practical Re-ID systems. We propose a purely deep transfer Re-ID model consisting of a deep convolutional neural network and an autoencoder. The latent code is divided into metric embedding and nuisance variables. We then utilize an unsupervised training method that does not rely on co-training with non-deep models. Our experiments show improvements over both the baseline and competitors' transfer learning models.

Alexey Potapov, Oleg Shcherbakov, Innokentii Zhdanov et al.

In this paper we propose a conceptual framework for higher-order artificial neural networks. The idea of higher-order networks arises naturally when a model is required to learn some group of transformations, every element of which is well-approximated by a traditional feedforward network. Thus the group as a whole can be represented as a hyper network. One of typical examples of such groups is spatial transformations. We show that the proposed framework, which we call HyperNets, is able to deal with at least two basic spatial transformations of images: rotation and affine transformation. We show that HyperNets are able not only to generalize rotation and affine transformation, but also to compensate the rotation of images bringing them into canonical forms.

Alexey Potapov, Sergey Rodionov, Maxim Peterson et al.

What frameworks and architectures are necessary to create a vision system for AGI? In this paper, we propose a formal model that states the task of perception within AGI. We show the role of discriminative and generative models in achieving efficient and general solution of this task, thus specifying the task in more detail. We discuss some existing generative and discriminative models and demonstrate their insufficiency for our purposes. Finally, we discuss some architectural dilemmas and open questions.

Alexey Potapov, Sergey Rodionov

Rational agents are usually built to maximize rewards. However, AGI agents can find undesirable ways of maximizing any prior reward function. Therefore value learning is crucial for safe AGI. We assume that generalized states of the world are valuable - not rewards themselves, and propose an extension of AIXI, in which rewards are used only to bootstrap hierarchical value learning. The modified AIXI agent is considered in the multi-agent environment, where other agents can be either humans or other "mature" agents, which values should be revealed and adopted by the "infant" AGI agent. General framework for designing such empathic agent with ethical bias is proposed also as an extension of the universal intelligence model. Moreover, we perform experiments in the simple Markov environment, which demonstrate feasibility of our approach to value learning in safe AGI.

Sergey Rodionov, Alexey Potapov, Yurii Vinogradov

Optimal probabilistic approach in reinforcement learning is computationally infeasible. Its simplification consisting in neglecting difference between true environment and its model estimated using limited number of observations causes exploration vs exploitation problem. Uncertainty can be expressed in terms of a probability distribution over the space of environment models, and this uncertainty can be propagated to the action-value function via Bellman iterations, which are computationally insufficiently efficient though. We consider possibility of directly measuring uncertainty of the action-value function, and analyze sufficiency of this facilitated approach.

Alexey Potapov, Sergey Rodionov, Andrew Myasnikov et al.

Existing theoretical universal algorithmic intelligence models are not practically realizable. More pragmatic approach to artificial general intelligence is based on cognitive architectures, which are, however, non-universal in sense that they can construct and use models of the environment only from Turing-incomplete model spaces. We believe that the way to the real AGI consists in bridging the gap between these two approaches. This is possible if one considers cognitive functions as a "cognitive bias" (priors and search heuristics) that should be incorporated into the models of universal algorithmic intelligence without violating their universality. Earlier reported results suiting this approach and its overall feasibility are discussed on the example of perception, planning, knowledge representation, attention, theory of mind, language, and some others.