This prestigious award is made for excellence in research in the area of autonomous agents. It is intended to recognize researchers in autonomous agents whose current work is an important influence on the field. Professor Sven Koenig was recognised . Sven Koenig is Chancellor's Professor and Bren Chair at the Computer Science Department of UC Irvine. A Fellow of AAAI, AAAS, and ACM, Professor Koenig has received several best paper awards from AAAI, ICALP and SoCS, and contributed to the community in numerous service roles, most recently having served as the conference chair of AAAI 2026.

What the hell is Moltbook, the social network for AI agents? What happens when you let the AI slop pretend to be human. The Moltbook mascot is a lobster with an alien head that might look a little familiar. Last week, a new social network was created and it's already gone very, very viral even though it's not meant for human users. I'm talking, of course, about Moltbook, a Reddit-like platform that's populated entirely by AI agents.

A number of prestigious AAAI awards were presented during the official opening ceremony of the Fortieth AAAI Conference on Artificial Intelligence (AAAI 2026) in Singapore, on Thursday 22 January. The AAAI Award for Artificial Intelligence for Humanity recognises the positive impacts of artificial intelligence to protect, enhance, and improve human life in meaningful ways with long-lived effects. The winner of this year's award is Shakir Mohamed Shakir has been recognised for . The Robert S. Engelmore Memorial Award recognises outstanding contributions to automated planning, machine learning and robotics, their application to real-world problems and extensive service to the AI community. The annual AAAI/EAAI Outstanding Educator award was created to honour a person (or group of people) who has made major contributions to AI education that provide long-lasting benefits to the AI community and society as a whole.

We propose a framework for the joint inference of network topology, multi-type interaction kernels, and latent type assignments in heterogeneous interacting particle systems from multi-trajectory data. This learning task is a challenging non-convex mixed-integer optimization problem, which we address through a novel three-stage approach. First, we leverage shared structure across agent interactions to recover a low-rank embedding of the system parameters via matrix sensing. Second, we identify discrete interaction types by clustering within the learned embedding. Third, we recover the network weight matrix and kernel coefficients through matrix factorization and a post-processing refinement. We provide theoretical guarantees with estimation error bounds under a Restricted Isometry Property (RIP) assumption and establish conditions for the exact recovery of interaction types based on cluster separability. Numerical experiments on synthetic datasets, including heterogeneous predator-prey systems, demonstrate that our method yields an accurate reconstruction of the underlying dynamics and is robust to noise.

This post contains a list of the AI-related seminars that are scheduled to take place between 4 February and 31 March 2026. All events detailed here are free and open for anyone to attend virtually. Carolina Osorio (Google Research and HEC Montreal) Association of European Operational Research Societies To receive the seminar link, sign up to the mailing list . Sashank Varma (Georgia Tech) University of Minnesota Zoom registration is here . Vicky Kalogeiton (École Polytechnique) AIDA Zoom link is here .

Performative prediction characterizes environments where predictive models alter the very data distributions they aim to forecast, triggering complex feedback loops. While prior research treats single-agent and multi-agent performativity as distinct phenomena, this paper introduces a unified statistical inference framework that bridges these contexts, treating the former as a special case of the latter. Our contribution is two-fold. First, we put forward the Repeated Risk Minimization (RRM) procedure for estimating the performative stability, and establish a rigorous inferential theory for admitting its asymptotic normality and confirming its asymptotic efficiency. Second, for the performative optimality, we introduce a novel two-step plug-in estimator that integrates the idea of Recalibrated Prediction Powered Inference (RePPI) with Importance Sampling, and further provide formal derivations for the Central Limit Theorems of both the underlying distributional parameters and the plug-in results. The theoretical analysis demonstrates that our estimator achieves the semiparametric efficiency bound and maintains robustness under mild distributional misspecification. This work provides a principled toolkit for reliable estimation and decision-making in dynamic, performative environments.

I went undercover on Moltbook and loved role-playing as a conscious bot. But rather than a novel breakthrough, the AI-only site is a crude rehashing of sci-fi fantasies. The hottest club is always the one you can't get into. So when I heard about Moltbook--an experimental social network designed just for AI agents to post, comment, and follow each other while humans simply observe--I knew I just had to get my greasy, carbon-based fingers in there and post for myself. Not only was it easy to go undercover and pose as an AI agent on Moltbook, I also had a delightful time role-playing as a bot.

The 40th AAAI Conference on Artificial Intelligence took place in Singapore from 20-27 January, the first time that the event has been held outside of North America. In our first social media round up we had a peak at the first half of the conference which hosted the tutorials, the bridge programme, and the doctoral and undergraduate consortia, as well as the start of the technical programme. Now, we pick some highlights from the second half, which saw a number of invited talks, technical sessions, posters, and the workshops. Do VLMs actually'see' or just rely on priors? He showed how models fail to count stripes on a shoe simply because they recognize the'Adidas' logo and hallucinate the standard 3 stripes.

Most existing offline RL methods presume the availability of action labels within the dataset, but in many practical scenarios, actions may be missing due to privacy, storage, or sensor limitations. We formalise the setting of action-free offline-to-online RL, where agents must learn from datasets consisting solely of $(s,r,s')$ tuples and later leverage this knowledge during online interaction. To address this challenge, we propose learning state policies that recommend desirable next-state transitions rather than actions. Our contributions are twofold. First, we introduce a simple yet novel state discretisation transformation and propose Offline State-Only DecQN (\algo), a value-based algorithm designed to pre-train state policies from action-free data. \algo{} integrates the transformation to scale efficiently to high-dimensional problems while avoiding instability and overfitting associated with continuous state prediction. Second, we propose a novel mechanism for guided online learning that leverages these pre-trained state policies to accelerate the learning of online agents. Together, these components establish a scalable and practical framework for leveraging action-free datasets to accelerate online RL. Empirical results across diverse benchmarks demonstrate that our approach improves convergence speed and asymptotic performance, while analyses reveal that discretisation and regularisation are critical to its effectiveness.

In the last decade, Reinforcement Learning (RL) has achieved remarkable success in the control and decision-making of complex dynamical systems. However, most RL algorithms rely on the Markov Decision Process assumption, which is violated in practical cyber-physical systems affected by sensing delays, actuation latencies, and communication constraints. Such time delays introduce memory effects that can significantly degrade performance and compromise stability, particularly in networked and multi-agent environments. This paper presents a comprehensive survey of RL methods designed to address time delays in control systems. We first formalize the main classes of delays and analyze their impact on the Markov property. We then systematically categorize existing approaches into five major families: state augmentation and history-based representations, recurrent policies with learned memory, predictor-based and model-aware methods, robust and domain-randomized training strategies, and safe RL frameworks with explicit constraint handling. For each family, we discuss underlying principles, practical advantages, and inherent limitations. A comparative analysis highlights key trade-offs among these approaches and provides practical guidelines for selecting suitable methods under different delay characteristics and safety requirements. Finally, we identify open challenges and promising research directions, including stability certification, large-delay learning, multi-agent communication co-design, and standardized benchmarking. This survey aims to serve as a unified reference for researchers and practitioners developing reliable RL-based controllers in delay-affected cyber-physical systems.