The booming success of LLMs initiates rapid development in LLM agents. Though the foundation of an LLM agent is the generative model, it is critical to devise the optimal reasoning strategies and agent architectures. Accordingly, LLM agent research advances from the simple chain-of-thought prompting to more complex ReAct and Reflection reasoning strategy; agent architecture also evolves from single agent generation to multi-agent conversation, as well as multi-LLM multi-agent group chat. However, with the existing intricate frameworks and libraries, creating and evaluating new reasoning strategies and agent architectures has become a complex challenge, which hinders research investigation into LLM agents. Thus, we open-source a new AI agent library, AgentLite, which simplifies this process by offering a lightweight, user-friendly platform for innovating LLM agent reasoning, architectures, and applications with ease. AgentLite is a task-oriented framework designed to enhance the ability of agents to break down tasks and facilitate the development of multi-agent systems. Furthermore, we introduce multiple practical applications developed with AgentLite to demonstrate its convenience and flexibility. Get started now at: \url{https://github.com/SalesforceAIResearch/AgentLite}.
In the multi-modal domain, the dependence of various models on specific input formats leads to user confusion and hinders progress. To address this challenge, we introduce \textbf{LLMBind}, a novel framework designed to unify a diverse array of multi-modal tasks. By harnessing a Mixture-of-Experts (MoE) Large Language Model (LLM), LLMBind processes multi-modal inputs and generates task-specific tokens, enabling the invocation of corresponding models to accomplish tasks. This unique approach empowers LLMBind to interpret inputs and generate outputs across various modalities, including image, text, video, and audio. Furthermore, we have constructed an interaction dataset comprising 400k instructions, which unlocks the ability of LLMBind for interactive visual generation and editing tasks. Extensive experimentation demonstrates that LLMBind achieves very superior performance across diverse tasks and outperforms existing models in user evaluations conducted in real-world scenarios. Moreover, the adaptability of LLMBind allows for seamless integration with the latest models and extension to new modality tasks, highlighting its potential to serve as a unified AI agent for modeling universal modalities.
We introduce Brain-Artificial Intelligence Interfaces (BAIs) as a new class of Brain-Computer Interfaces (BCIs). Unlike conventional BCIs, which rely on intact cognitive capabilities, BAIs leverage the power of artificial intelligence to replace parts of the neuro-cognitive processing pipeline. BAIs allow users to accomplish complex tasks by providing high-level intentions, while a pre-trained AI agent determines low-level details. This approach enlarges the target audience of BCIs to individuals with cognitive impairments, a population often excluded from the benefits of conventional BCIs. We present the general concept of BAIs and illustrate the potential of this new approach with a Conversational BAI based on EEG. In particular, we show in an experiment with simulated phone conversations that the Conversational BAI enables complex communication without the need to generate language. Our work thus demonstrates, for the first time, the ability of a speech neuroprosthesis to enable fluent communication in realistic scenarios with non-invasive technologies.
Chemical synthesis, which is crucial for advancing material synthesis and drug discovery, impacts various sectors including environmental science and healthcare. The rise of technology in chemistry has generated extensive chemical data, challenging researchers to discern patterns and refine synthesis processes. Artificial intelligence (AI) helps by analyzing data to optimize synthesis and increase yields. However, AI faces challenges in processing literature data due to the unstructured format and diverse writing style of chemical literature. To overcome these difficulties, we introduce an end-to-end AI agent framework capable of high-fidelity extraction from extensive chemical literature. This AI agent employs large language models (LLMs) for prompt generation and iterative optimization. It functions as a chemistry assistant, automating data collection and analysis, thereby saving manpower and enhancing performance. Our framework's efficacy is evaluated using accuracy, recall, and F1 score of reaction condition data, and we compared our method with human experts in terms of content correctness and time efficiency. The proposed approach marks a significant advancement in automating chemical literature extraction and demonstrates the potential for AI to revolutionize data management and utilization in chemistry.
The project's aim is to create an AI agent capable of selecting good actions in a game-playing domain called Battlespace. Sequential domains like Battlespace are important testbeds for planning problems, as such, the Department of Defense uses such domains for wargaming exercises. The agents we developed combine Monte Carlo Tree Search (MCTS) and Deep Q-Network (DQN) techniques in an effort to navigate the game environment, avoid obstacles, interact with adversaries, and capture the flag. This paper will focus on the encoding techniques we explored to present complex structured data stored in a Python class, a necessary precursor to an agent.
In many collaborative settings, artificial intelligence (AI) agents must be able to adapt to new teammates that use unknown or previously unobserved strategies. While often simple for humans, this can be challenging for AI agents. For example, if an AI agent learns to drive alongside others (a training set) that only drive on one side of the road, it may struggle to adapt this experience to coordinate with drivers on the opposite side, even if their behaviours are simply flipped along the left-right symmetry. To address this we introduce symmetry-breaking augmentations (SBA), which increases diversity in the behaviour of training teammates by applying a symmetry-flipping operation. By learning a best-response to the augmented set of teammates, our agent is exposed to a wider range of behavioural conventions, improving performance when deployed with novel teammates. We demonstrate this experimentally in two settings, and show that our approach improves upon previous ad hoc teamwork results in the challenging card game Hanabi. We also propose a general metric for estimating symmetry-dependency amongst a given set of policies.
The human-like automatic deductive reasoning has always been one of the most challenging open problems in the interdiscipline of mathematics and artificial intelligence. This paper is the third in a series of our works. We built a neural-symbolic system, called FGeoDRL, to automatically perform human-like geometric deductive reasoning. The neural part is an AI agent based on reinforcement learning, capable of autonomously learning problem-solving methods from the feedback of a formalized environment, without the need for human supervision. It leverages a pre-trained natural language model to establish a policy network for theorem selection and employ Monte Carlo Tree Search for heuristic exploration. The symbolic part is a reinforcement learning environment based on geometry formalization theory and FormalGeo, which models GPS as a Markov Decision Process. In this formal symbolic system, the known conditions and objectives of the problem form the state space, while the set of theorems forms the action space. Leveraging FGeoDRL, we have achieved readable and verifiable automated solutions to geometric problems. Experiments conducted on the formalgeo7k dataset have achieved a problem-solving success rate of 86.40%. The project is available at https://github.com/PersonNoName/FGeoDRL.
Achieving human-AI alignment in complex multi-agent games is crucial for creating trustworthy AI agents that enhance gameplay. We propose a method to evaluate this alignment using an interpretable task-sets framework, focusing on high-level behavioral tasks instead of low-level policies. Our approach has three components. First, we analyze extensive human gameplay data from Xbox's Bleeding Edge (100K+ games), uncovering behavioral patterns in a complex task space. This task space serves as a basis set for a behavior manifold capturing interpretable axes: fight-flight, explore-exploit, and solo-multi-agent. Second, we train an AI agent to play Bleeding Edge using a Generative Pretrained Causal Transformer and measure its behavior. Third, we project human and AI gameplay to the proposed behavior manifold to compare and contrast. This allows us to interpret differences in policy as higher-level behavioral concepts, e.g., we find that while human players exhibit variability in fight-flight and explore-exploit behavior, AI players tend towards uniformity. Furthermore, AI agents predominantly engage in solo play, while humans often engage in cooperative and competitive multi-agent patterns. These stark differences underscore the need for interpretable evaluation, design, and integration of AI in human-aligned applications. Our study advances the alignment discussion in AI and especially generative AI research, offering a measurable framework for interpretable human-agent alignment in multiplayer gaming.
As AI systems quickly improve in both breadth and depth of performance, they lend themselves to creating increasingly powerful and realistic agents, including the possibility of agents modeled on specific people. We anticipate that within our lifetimes it may become common practice for people to create custom AI agents to interact with loved ones and/or the broader world after death; indeed, the past year has seen a boom in startups purporting to offer such services. We call these generative ghosts, since such agents will be capable of generating novel content rather than merely parroting content produced by their creator while living. In this paper, we reflect on the history of technologies for AI afterlives, including current early attempts by individual enthusiasts and startup companies to create generative ghosts. We then introduce a novel design space detailing potential implementations of generative ghosts, and use this analytic framework to ground discussion of the practical and ethical implications of various approaches to designing generative ghosts, including potential positive and negative impacts on individuals and society. Based on these considerations, we lay out a research agenda for the AI and HCI research communities to better understand the risk/benefit landscape of this novel technology so as to ultimately empower people who wish to create and interact with AI afterlives to do so in a beneficial manner.
Step-by-step decision planning with large language models (LLMs) is gaining attention in AI agent development. This paper focuses on decision planning with uncertainty estimation to address the hallucination problem in language models. Existing approaches are either white-box or computationally demanding, limiting use of black-box proprietary LLMs within budgets. The paper's first contribution is a non-parametric uncertainty quantification method for LLMs, efficiently estimating point-wise dependencies between input-decision on the fly with a single inference, without access to token logits. This estimator informs the statistical interpretation of decision trustworthiness. The second contribution outlines a systematic design for a decision-making agent, generating actions like ``turn on the bathroom light'' based on user prompts such as ``take a bath''. Users will be asked to provide preferences when more than one action has high estimated point-wise dependencies. In conclusion, our uncertainty estimation and decision-making agent design offer a cost-efficient approach for AI agent development.
Large language models (LLM) are perceived to offer promising potentials for automating security tasks, such as those found in security operation centers (SOCs). As a first step towards evaluating this perceived potential, we investigate the use of LLMs in software pentesting, where the main task is to automatically identify software security vulnerabilities in source code. We hypothesize that an LLM-based AI agent can be improved over time for a specific security task as human operators interact with it. Such improvement can be made, as a first step, by engineering prompts fed to the LLM based on the responses produced, to include relevant contexts and structures so that the model provides more accurate results. Such engineering efforts become sustainable if the prompts that are engineered to produce better results on current tasks, also produce better results on future unknown tasks. To examine this hypothesis, we utilize the OWASP Benchmark Project 1.2 which contains 2,740 hand-crafted source code test cases containing various types of vulnerabilities. We divide the test cases into training and testing data, where we engineer the prompts based on the training data (only), and evaluate the final system on the testing data. We compare the AI agent's performance on the testing data against the performance of the agent without the prompt engineering. We also compare the AI agent's results against those from SonarQube, a widely used static code analyzer for security testing. We built and tested multiple versions of the AI agent using different off-the-shelf LLMs -- Google's Gemini-pro, as well as OpenAI's GPT-3.5-Turbo and GPT-4-Turbo (with both chat completion and assistant APIs). The results show that using LLMs is a viable approach to build an AI agent for software pentesting that can improve through repeated use and prompt engineering.
Many important behavior changes are frictionful; they require individuals to expend effort over a long period with little immediate gratification. Here, an artificial intelligence (AI) agent can provide personalized interventions to help individuals stick to their goals. In these settings, the AI agent must personalize rapidly (before the individual disengages) and interpretably, to help us understand the behavioral interventions. In this paper, we introduce Behavior Model Reinforcement Learning (BMRL), a framework in which an AI agent intervenes on the parameters of a Markov Decision Process (MDP) belonging to a boundedly rational human agent. Our formulation of the human decision-maker as a planning agent allows us to attribute undesirable human policies (ones that do not lead to the goal) to their maladapted MDP parameters, such as an extremely low discount factor. Furthermore, we propose a class of tractable human models that captures fundamental behaviors in frictionful tasks. Introducing a notion of MDP equivalence specific to BMRL, we theoretically and empirically show that AI planning with our human models can lead to helpful policies on a wide range of more complex, ground-truth humans.
This report outlines a transformative initiative in the financial investment industry, where the conventional decision-making process, laden with labor-intensive tasks such as sifting through voluminous documents, is being reimagined. Leveraging language models, our experiments aim to automate information summarization and investment idea generation. We seek to evaluate the effectiveness of fine-tuning methods on a base model (Llama2) to achieve specific application-level goals, including providing insights into the impact of events on companies and sectors, understanding market condition relationships, generating investor-aligned investment ideas, and formatting results with stock recommendations and detailed explanations. Through state-of-the-art generative modeling techniques, the ultimate objective is to develop an AI agent prototype, liberating human investors from repetitive tasks and allowing a focus on high-level strategic thinking. The project encompasses a diverse corpus dataset, including research reports, investment memos, market news, and extensive time-series market data. We conducted three experiments applying unsupervised and supervised LoRA fine-tuning on the llama2_7b_hf_chat as the base model, as well as instruction fine-tuning on the GPT3.5 model. Statistical and human evaluations both show that the fine-tuned versions perform better in solving text modeling, summarization, reasoning, and finance domain questions, demonstrating a pivotal step towards enhancing decision-making processes in the financial domain. Code implementation for the project can be found on GitHub: https://github.com/Firenze11/finance_lm.
We investigate the question: if an AI agent is known to be safe in one setting, is it also safe in a new setting similar to the first? This is a core question of AI alignment--we train and test models in a certain environment, but deploy them in another, and we need to guarantee that models that seem safe in testing remain so in deployment. Our notion of safety is based on power-seeking--an agent which seeks power is not safe. In particular, we focus on a crucial type of power-seeking: resisting shutdown. We model agents as policies for Markov decision processes, and show (in two cases of interest) that not resisting shutdown is "stable": if an MDP has certain policies which don't avoid shutdown, the corresponding policies for a similar MDP also don't avoid shutdown. We also show that there are natural cases where safety is _not_ stable--arbitrarily small perturbations may result in policies which never shut down. In our first case of interest--near-optimal policies--we use a bisimulation metric on MDPs to prove that small perturbations won't make the agent take longer to shut down. Our second case of interest is policies for MDPs satisfying certain constraints which hold for various models (including language models). Here, we demonstrate a quantitative bound on how fast the probability of not shutting down can increase: by defining a metric on MDPs; proving that the probability of not shutting down, as a function on MDPs, is lower semicontinuous; and bounding how quickly this function decreases.
Despite significant progress in autonomous vehicles (AVs), the development of driving policies that ensure both the safety of AVs and traffic flow efficiency has not yet been fully explored. In this paper, we propose an enhanced human-in-the-loop reinforcement learning method, termed the Human as AI mentor-based deep reinforcement learning (HAIM-DRL) framework, which facilitates safe and efficient autonomous driving in mixed traffic platoon. Drawing inspiration from the human learning process, we first introduce an innovative learning paradigm that effectively injects human intelligence into AI, termed Human as AI mentor (HAIM). In this paradigm, the human expert serves as a mentor to the AI agent. While allowing the agent to sufficiently explore uncertain environments, the human expert can take control in dangerous situations and demonstrate correct actions to avoid potential accidents. On the other hand, the agent could be guided to minimize traffic flow disturbance, thereby optimizing traffic flow efficiency. In detail, HAIM-DRL leverages data collected from free exploration and partial human demonstrations as its two training sources. Remarkably, we circumvent the intricate process of manually designing reward functions; instead, we directly derive proxy state-action values from partial human demonstrations to guide the agents' policy learning. Additionally, we employ a minimal intervention technique to reduce the human mentor's cognitive load. Comparative results show that HAIM-DRL outperforms traditional methods in driving safety, sampling efficiency, mitigation of traffic flow disturbance, and generalizability to unseen traffic scenarios. The code and demo videos for this paper can be accessed at: https://zilin-huang.github.io/HAIM-DRL-website/
In proof assistants, the physical proximity between two formal mathematical concepts is a strong predictor of their mutual relevance. Furthermore, lemmas with close proximity regularly exhibit similar proof structures. We show that this locality property can be exploited through online learning techniques to obtain solving agents that far surpass offline learners when asked to prove theorems in an unseen mathematical setting. We extensively benchmark two such online solvers implemented in the Tactician platform for the Coq proof assistant: First, Tactician's online $k$-nearest neighbor solver, which can learn from recent proofs, shows a $1.72\times$ improvement in theorems proved over an offline equivalent. Second, we introduce a graph neural network, Graph2Tac, with a novel approach to build hierarchical representations for new definitions. Graph2Tac's online definition task realizes a $1.5\times$ improvement in theorems solved over an offline baseline. The $k$-NN and Graph2Tac solvers rely on orthogonal online data, making them highly complementary. Their combination improves $1.27\times$ over their individual performances. Both solvers outperform all other general-purpose provers for Coq, including CoqHammer, Proverbot9001, and a transformer baseline by at least $1.48\times$ and are available for practical use by end-users.
Constructing a universal moral code for artificial intelligence (AI) is difficult or even impossible, given that different human cultures have different definitions of morality and different societal norms. We therefore argue that the value system of an AI should be culturally attuned: just as a child raised in a particular culture learns the specific values and norms of that culture, we propose that an AI agent operating in a particular human community should acquire that community's moral, ethical, and cultural codes. How AI systems might acquire such codes from human observation and interaction has remained an open question. Here, we propose using inverse reinforcement learning (IRL) as a method for AI agents to acquire a culturally-attuned value system implicitly. We test our approach using an experimental paradigm in which AI agents use IRL to learn different reward functions, which govern the agents' moral values, by observing the behavior of different cultural groups in an online virtual world requiring real-time decision making. We show that an AI agent learning from the average behavior of a particular cultural group can acquire altruistic characteristics reflective of that group's behavior, and this learned value system can generalize to new scenarios requiring altruistic judgments. Our results provide, to our knowledge, the first demonstration that AI agents could potentially be endowed with the ability to continually learn their values and norms from observing and interacting with humans, thereby becoming attuned to the culture they are operating in.
In recent years, AI has demonstrated remarkable capabilities in simulating human behaviors, particularly those implemented with large language models (LLMs). However, due to the lack of systematic evaluation of LLMs' simulated behaviors, the believability of LLMs among humans remains ambiguous, i.e., it is unclear which behaviors of LLMs are convincingly human-like and which need further improvements. In this work, we design SimulateBench to evaluate the believability of LLMs when simulating human behaviors. In specific, we evaluate the believability of LLMs based on two critical dimensions: 1) consistency: the extent to which LLMs can behave consistently with the given information of a human to simulate; and 2) robustness: the ability of LLMs' simulated behaviors to remain robust when faced with perturbations. SimulateBench includes 65 character profiles and a total of 8,400 questions to examine LLMs' simulated behaviors. Based on SimulateBench, we evaluate the performances of 10 widely used LLMs when simulating characters. The experimental results reveal that current LLMs struggle to align their behaviors with assigned characters and are vulnerable to perturbations in certain factors.
A key method for creating Artificial Intelligence (AI) agents is Reinforcement Learning (RL). However, constructing a standalone RL policy that maps perception to action directly encounters severe problems, chief among them being its lack of generality across multiple tasks and the need for a large amount of training data. The leading cause is that it cannot effectively integrate prior information into the perception-action cycle when devising the policy. Large language models (LLMs) emerged as a fundamental way to incorporate cross-domain knowledge into AI agents but lack crucial learning and adaptation toward specific decision problems. This paper presents a general framework model for integrating and learning structured reasoning into AI agents' policies. Our methodology is motivated by the modularity found in the human brain. The framework utilises the construction of intrinsic and extrinsic functions to add previous understandings of reasoning structures. It also provides the adaptive ability to learn models inside every module or function, consistent with the modular structure of cognitive processes. We describe the framework in-depth and compare it with other AI pipelines and existing frameworks. The paper explores practical applications, covering experiments that show the effectiveness of our method. Our results indicate that AI agents perform and adapt far better when organised reasoning and prior knowledge are embedded. This opens the door to more resilient and general AI agent systems.
Artificial Intelligence (AI) systems, trained in controlled environments, often struggle in real-world complexities. We propose a general framework for estimating domain complexity across diverse environments, like open-world learning and real-world applications. This framework distinguishes between intrinsic complexity (inherent to the domain) and extrinsic complexity (dependent on the AI agent). By analyzing dimensionality, sparsity, and diversity within these categories, we offer a comprehensive view of domain challenges. This approach enables quantitative predictions of AI difficulty during environment transitions, avoids bias in novel situations, and helps navigate the vast search spaces of open-world domains.