AAAI/EAAI Patrick Henry Winston Outstanding Educator Award: T(w)eaching AI in the Age of LLMs

Mechanistic interpretability is one of the most exciting and important research programs in current AI. My aim is to build some philosophical foundations for the program, along with setting out some concrete challenges and assessing progress to date. I will argue for the importance of propositional interpretability, which involves interpreting a system’s mechanisms and behavior in terms of propositional attitudes: attitudes (such as belief, desire, or subjective probability) to propositions (e.g. the proposition that it is hot outside). Propositional attitudes are the central way that we interpret and explain human beings and they are likely to be central in AI too. A central challenge is what I call thought logging: creating systems log all of the relevant propositional attitudes in an AI system over time. I will examine currently popular methods of interpretability (such as probing, sparse auto-encoders, and chain of thought methods) as well as philosophical methods of interpretation (including psychosemantics and representation theorems) to assess their strengths and weaknesses as methods of propositional interpretability.

Propositional Interpretability in Humans and AI Systems

Can AI Benefit Humanity?

Spatial reasoning is a core component of an agent’s ability to operate in or reason about the physical world. LLMs are widely promoted as having abilities to reason about a wide variety of domains, including commonsense. In this talk I will discuss the ability of state-of-the-art LLMs to perform commonsense reasoning, particularly with regard to spatial information. Across a wide range of LLMs, although they show abilities rather better than chance, they still struggle with many questions and tasks, for example when reasoning about directions, or topological relations.

Can Large Language Models Reason about Spatial Information?

In this keynote, Andrew will explore how current technologies, particularly agentic workflow, are revolutionizing the development of AI products. He will also delve into the role of AI in prototyping, showcasing how AI lowers the cost of software development and accelerates the prototyping process.

AI, Agents and Applications

This talk focuses on the use of foundation models to study problems in labor economics, in particular problems relating to the progression of worker careers and wages. In Vafa et al. (2024a), we introduced a transformer-based predictive model, CAREER, that predicts a worker’s next job as a function of career history (an “occupation model”). CAREER was initially estimated (“pre-trained”) using a large, unrepresentative resume dataset, which served as a “foundation model,” and parameter estimation was continued (“fine-tuned”) using data from a representative survey. CAREER had better predictive performance than benchmarks. Athey et al (2024) considers an alternative where the resume-based foundation model is replaced by a large language model (LLM). We convert tabular data from the survey into text files that resemble resumes and fine-tune the LLMs using these text files with the objective to predict the next token (word). The resulting fine-tuned LLM is used as an input to an occupation model. Its predictive performance surpasses all prior models. We demonstrate the value of fine-tuning and further show that by adding more career data from a different population, fine-tuning smaller LLMs surpasses the performance of fine-tuning larger models. In Vafa et al. (2024b), we apply and adapt this framework to the problem of gender wage decompositions, which require estimating the portion of the gender wage gap explained by career histories of workers. Classical methods for decomposing the wage gap employ simple linear models, and the resulting decompositions thus suffer from omitted variable bias (OVB), where covariates that are correlated with both gender and wages are not included in the model. We explore an alternative methodology for wage gap decomposition that employs CAREER as a foundation model. We prove that the way foundation models are usually trained might still lead to OVB, but develop fine-tuning algorithms that empirically mitigate this issue. We first provide a novel set of conditions under which an estimator of the wage gap based on a fine-tuned foundation model is root-n-consistent. Building on the theory, we then propose methods for fine-tuning foundation models that minimize OVB. Using data from the Panel Study of Income Dynamics, we find that history explains more of the gender wage gap than standard econometric models can measure, and we identify elements of history that are important for reducing OVB.

Predicting Career Transitions and Estimating Wage Disparities Using Foundation Models

Natural language serves as the primary mode of communication when an intelligent agent with a physical presence engages with human beings. While a plethora of research focuses on natural language understanding (NLU), encompassing endeavors such as sentiment analysis, intent prediction, question answering, and summarization, the scope of NLU directed at situations necessitating tangible actions by an embodied agent remains limited. The inherent ambiguity and incompleteness inherent in natural language present challenges for intelligent agents striving to decipher human intention. To tackle this predicament head-on, we introduce a novel system known as \textbf{t}ask and \textbf{a}rgument \textbf{g}rounding for \textbf{E}mbodied agents (\system). At its core, our system employs an inventive neural network model designed to extract a series of tasks from complex task instructions expressed in natural language. Our proposed model adopts an encoder-decoder framework enriched with nested decoding to effectively extract tasks and their corresponding arguments from these intricate instructions. These extracted tasks are then mapped (or grounded) to the robot's established collection of skills, while the arguments find grounding in objects present within the environment. To facilitate the training and evaluation of our system, we have curated a dataset featuring complex instructions. The results of our experiments underscore the prowess of our approach, as it outperforms robust baseline models.

tagE: Enabling an Embodied Agent to Understand Human Instructions​

A key contribution to being a successful researcher in natural language processing, as in any area, is having a clear overarching vision of what your body of research is trying to accomplish. Using my own 40-year career as an example, I will attempt to provide general advice on formulating and pursuing a coherent research vision. In particular, I will focus on formulating a unique, personal objective that exploits your specific talents, knowledge, and passions, and that is distinct from the current popular trends in the field. I will also focus on formulating a vision that bridges existing fields of study to produce an overarching agenda that unifies previously disparate ideas.

The Vision Thing: Finding and Pursuing your Research Passion

Transliteration is very important in the Indian
language context due to the usage of multi-
ple scripts and the widespread use of roman-
ized inputs. However, few training and evalu-
ation sets are publicly available. We introduce
Aksharantar1, the largest publicly available
transliteration dataset for Indian languages cre-
ated by mining from monolingual and paral-
lel corpora, as well as collecting data from hu-
man annotators. The dataset contains 26 mil-
lion transliteration pairs for 21 Indic lan-
guages from 3 language families using 12
scripts. Aksharantar is 21 times larger than ex-
isting datasets and is the first publicly available
dataset for 7 languages and 1 language fam-
ily. We also introduce the Aksharantar testset
comprising 103k word pairs spanning 19 lan-
guages that enables a fine-grained analysis of
transliteration models on native origin words,
foreign words, frequent words, and rare words.
Using the training set, we trained IndicXlit,
a multilingual transliteration model that im-
proves accuracy by 15% on the Dakshina test
set, and establishes strong baselines on the Ak-
sharantar testset introduced in this work. The
models, mining scripts, transliteration guide-
lines, and datasets are available at https://
github.com/AI4Bharat/IndicXlit under
open-source licenses.

Aksharantar: Open Indic-language Transliteration datasets and models for the Next Billion Users

The prolific use of Large Language Models (LLMs) as an alternate knowledge base requires them to be factually consistent, necessitating both correctness and consistency traits for paraphrased queries. Recently, significant attempts have been made to benchmark datasets and metrics to evaluate LLMs for these traits. However, structural simplicity (subject-relation-object) and contemporary association in their query formulation limit the broader definition of factuality and consistency. In this study, we introduce TeCFaP, a novel Temporally Consistent Factuality Probe task to expand the consistent factuality probe in the temporal dimension. To this end, we propose TEMP-COFAC, a high-quality dataset of prefix-style English query paraphrases. Subsequently, we extend the definitions of existing metrics to represent consistent factuality across temporal dimension. We experiment with a diverse set of LLMs and find most of them performing poorly on TeCFaP. Next, we propose a novel solution CoTSeLF (Consistent-Time-Sensitive Learning Framework) combining multi-task instruction tuning (MT-IT) with consistent-time-sensitive reinforcement learning (CTSRL) to improve temporally consistent factuality in LLMs. Our experiments demonstrate the efficacy of CoTSeLF over several baselines.

Temporally Consistent Factuality Probing for Large Language Models

Language Model Programs, i.e. sophisticated pipelines of modular language model (LM) calls, are increasingly advancing NLP tasks, but they require crafting prompts that are jointly effective for all modules. We study prompt optimization for LM programs, i.e. how to update these prompts to maximize a downstream metric without access to module-level labels or gradients. To make this tractable, we factorize our problem into optimizing the free-form instructions and few-shot demonstrations of every module and introduce several strategies to craft task-grounded instructions and navigate credit assignment across modules. Our strategies include (i) program- and data-aware techniques for proposing effective instructions, (ii) a stochastic mini-batch evaluation function for learning a surrogate model of our objective, and (iii) a meta-optimization procedure in which we refine how LMs construct proposals over time. Using these insights we develop MIPRO, a novel algorithm for optimizing LM programs. MIPRO outperforms baseline optimizers on five of seven diverse multi-stage LM programs using a best-in-class open-source model (Llama-3-8B), by as high as 13% accuracy. We have released our new optimizers and benchmark in DSPy at [http://dspy.ai](http://dspy.ai).

Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs

Recent models for natural language understanding are inclined to exploit simple patterns in datasets, commonly known as shortcuts. These shortcuts hinge on spurious correlations between labels and latent features existing in the training data. At inference time, shortcut-dependent models are likely to generate erroneous predictions under distribution shifts, particularly when some latent features are no longer correlated with the labels. To avoid this, previous studies have trained models to eliminate the reliance on shortcuts. In this study, we explore a different direction: pessimistically aggregating the predictions of a mixture-of-experts, assuming each expert captures relatively different latent features. The experimental results demonstrate that our post-hoc control over the experts significantly enhances the model's robustness to the distribution shift in shortcuts. Besides, we show that our approach has some practical advantages. We also analyze our model and provide results to support the assumption.

Underline digital video library

AAAI/EAAI Patrick Henry Winston Outstanding Educator Award: T(w)eaching AI in the Age of LLMs

Propositional Interpretability in Humans and AI Systems

Can AI Benefit Humanity?

Can Large Language Models Reason about Spatial Information?

AI, Agents and Applications

Predicting Career Transitions and Estimating Wage Disparities Using Foundation Models

tagE: Enabling an Embodied Agent to Understand Human Instructions

The Vision Thing: Finding and Pursuing your Research Passion

Aksharantar: Open Indic-language Transliteration datasets and models for the Next Billion Users

Temporally Consistent Factuality Probing for Large Language Models

Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs

Not Eliminate but Aggregate: Post-Hoc Control over Mixture-of-Experts to Address Shortcut Shifts in Natural Language Understanding

Stay up to date with the latest Underline news!

PRODUCT

COMPANY

RESOURCES

Underline digital video library

AAAI/EAAI Patrick Henry Winston Outstanding Educator Award: T(w)eaching AI in the Age of LLMs

Propositional Interpretability in Humans and AI Systems

Can AI Benefit Humanity?

Can Large Language Models Reason about Spatial Information?

AI, Agents and Applications

Predicting Career Transitions and Estimating Wage Disparities Using Foundation Models

tagE: Enabling an Embodied Agent to Understand Human Instructions​

The Vision Thing: Finding and Pursuing your Research Passion

Aksharantar: Open Indic-language Transliteration datasets and models for the Next Billion Users

Temporally Consistent Factuality Probing for Large Language Models

Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs

Not Eliminate but Aggregate: Post-Hoc Control over Mixture-of-Experts to Address Shortcut Shifts in Natural Language Understanding

Stay up to date with the latest Underline news!

PRODUCT

COMPANY

RESOURCES

tagE: Enabling an Embodied Agent to Understand Human Instructions