Adrian Weller

Recent advances in Artificial Intelligence (AI) have yielded powerful computational models that, by learning from vast amounts of human-generated data, are increasingly posited as approximate models of human cognition. However, we argue that many current evaluation paradigms for AI are insufficient for assessing human-like cognitive capabilities in these models. We identify a set of key shortcomings: a lack of human-validated labels, inadequate representation of human response variability and uncertainty, and reliance on simplified and ecologically invalid tasks. We support our claims by conducting a human evaluation study on nine existing AI benchmarks, suggesting major limitations in task and label designs. To address these limitations, we propose five concrete recommendations for future AI evaluation efforts that will enable more rigorous and meaningful understanding of human-like cognitive capacities in AI models.

AI systems have historically been used as tools that execute narrowly defined tasks. Yet recent advances in AI have unlocked possibilities for a new class of models that genuinely collaborate with humans in complex reasoning, from conceptualizing problems to brainstorming solutions. Such AI thought partners enable novel forms of collaboration and extended cognition, yet they also pose major risks—including and beyond risks of typical AI tools and agents. In this Commentary, we systematically characterize risks of AI thought partners through a novel framework that identifies risks at multiple levels of analysis, including Real-time, Individual, and Societal risks arising from collaborative cognition (RISc). We leverage this framework to propose concrete metrics for risk evaluation and finally suggest specific mitigation strategies for developers and policymakers. As AI thought partners continue to proliferate, these strategies can help prevent major harms and ensure that humans actively benefit from productive thought partnerships.

Inertial measurement units (IMUs) have become attractive for monitoring joint kinematics due to their portability and versatility. However, their limited accuracy, inability to analyze data in real-time, and complex data fusion algorithms requiring precise sensor-to-segment calibrations hinder their clinical and daily use. This paper introduces KEEN (KinEmatics Estimation Network), an innovative framework that exploits lightweight artificial neural networks (ANNs) to provide real-time predictions of multi-plane ankle kinematics using a minimal number of IMUs, without calibration requirements. Five ANN algorithms were developed and evaluated using 42 inputs derived from four IMUs in both intra-subject and inter-subject tasks. Extensive experimental results yielded exciting findings: even a single IMU located at the heel can provide clinically acceptable estimations of ankle kinematics, implying significant potential for cost and energy savings. Statistical analysis demonstrated the superiority of the developed Long Short-Term Memory (LSTM) network over the other models in intra-subject tasks, achieving impressive accuracy (RMSE: 1.88 ${}^{\circ } \pm 0.02{}^{\circ }$ , MAE: 1.41 ${}^{\circ }\pm 0.01{}^{\circ }$ , and r2 score: 0.93 $\pm$ 0.01), indicating strong generalization within the same subject. In inter-subject tasks, the convolutional neural network (CNN) and the CNN-LSTM models showed comparable performance but statistically outperformed the other models in terms of estimation accuracy across various inputs. When using a single IMU, the CNN model achieved the lowest error (RMSE: 4.13 ${}^{\circ }\pm 0.55{}^{\circ }$ , MAE: 3.33 ${}^{\circ }\pm 0.48{}^{\circ }$ , and r2 score: 0.50 $\pm$ 0.21), showcasing its effective generalization to new subjects. Furthermore, deploying the CNN into a microcontroller, with a sinlge IMU at the heel, resulted in promising real-time ankle kinematics estimations (RMSE: 3.34 ${}^{\circ }\pm 0.48{}^{\circ }$ , MAE: 2.68 ${}^{\circ }\pm 0.46{}^{\circ }$ and r2 score: 0.63 $\pm$ 0.07). Overall, this research highlights the potential of combining IMUs with ANNs as reliable and practical tools for early prevention and rehabilitation of ankle injuries.

What do we want from machine intelligence? We envision machines that are not just tools for thought but partners in thought: reasonable, insightful, knowledgeable, reliable and trustworthy systems that think with us. Current artificial intelligence systems satisfy some of these criteria, some of the time. In this Perspective, we show how the science of collaborative cognition can be put to work to engineer systems that really can be called ‘thought partners’, systems built to meet our expectations and complement our limitations. We lay out several modes of collaborative thought in which humans and artificial intelligence thought partners can engage, and we propose desiderata for human-compatible thought partnerships. Drawing on motifs from computational cognitive science, we motivate an alternative scaling path for the design of thought partners and ecosystems around their use through a Bayesian lens, whereby the partners we construct actively build and reason over models of the human and world.

Many important problems require modelling large-scale spatio-temporal datasets, with one prevalent example being weather forecasting. Recently, transformer-based approaches have shown great promise in a range of weather forecasting problems. However, these have mostly focused on gridded data sources, neglecting the wealth of unstructured, off-the-grid data from observational measurements such as those at weather stations. A promising family of models suitable for such tasks are neural processes (NPs), notably the family of transformer neural processes (TNPs). Although TNPs have shown promise on small spatio-temporal datasets, they are unable to scale to the quantities of data used by state-of-the-art weather and climate models. This limitation stems from their lack of efficient attention mechanisms. We address this shortcoming through the introduction of gridded pseudo-token TNPs which employ specialised encoders and decoders to handle unstructured observations and utilise a processor containing gridded pseudo-tokens that leverage efficient attention mechanisms. Our method consistently outperforms a range of strong baselines on various synthetic and real-world regression tasks involving large-scale data, while maintaining competitive computational efficiency. The real-life experiments are performed on weather data, demonstrating the potential of our approach to bring performance and computational benefits when applied at scale in a weather modelling pipeline.

Safeguarding the intellectual property of machine learning models has emerged as a pressing concern in AI security. Model watermarking is a powerful technique for protecting ownership of machine learning models, yet its reliability has been recently challenged by recent watermark removal attacks. In this work, we investigate why existing watermark embedding techniques particularly those based on backdooring are vulnerable. Through an information-theoretic analysis, we show that the resilience of watermarking against erasure attacks hinges on the choice of trigger-set samples, where current uses of out-distribution trigger-set are inherently vulnerable to white-box adversaries. Based on this discovery, we propose a novel model watermarking scheme, In-distribution Watermark Embedding (IWE), to overcome the limitations of existing method. To further minimise the gap to clean models, we analyze the role of logits as watermark information carriers and propose a new approach to better conceal watermark information within the logits. Experiments on real-world datasets including CIFAR-100 and Caltech-101 demonstrate that our method robustly defends against various adversaries with negligible accuracy loss (< 0.1%).

From science to law enforcement, many research questions are answerable only by poring over a large amount of unstructured text documents. While people can extract information from such documents with high accuracy, this is often too time-consuming to be practical. On the other hand, automated approaches produce nearly-immediate results, but are not reliable enough for applications where near-perfect precision is essential. Motivated by two use cases from criminal justice, we consider the benefits and drawbacks of various human-only, human–machine, and machine-only approaches. Finding no tool well suited for our use cases, we develop a human-in-the-loop method for fast but accurate extraction of structured data from unstructured text. The tool is based on automated extraction followed by human validation, and is particularly useful in cases where purely manual extraction is not practical. Testing on three criminal justice datasets, we find that the combination of the computer speed and human understanding yields precision comparable to manual annotation while requiring only a fraction of time, and significantly outperforms the precision of all fully automated baselines.

There is much excitement about the opportunity to harness the power of large language models (LLMs) when building problem-solving assistants. However, the standard methodology of evaluating LLMs relies on static pairs of inputs and outputs; this is insufficient for making an informed decision about which LLMs are best to use in an interactive setting, and how that varies by setting. Static assessment therefore limits how we understand language model capabilities. We introduce CheckMate, an adaptable prototype platform for humans to interact with and evaluate LLMs. We conduct a study with CheckMate to evaluate three language models (InstructGPT, ChatGPT, and GPT-4) as assistants in proving undergraduate-level mathematics, with a mixed cohort of participants from undergraduate students to professors of mathematics. We release the resulting interaction and rating dataset, MathConverse. By analyzing MathConverse, we derive a taxonomy of human query behaviors and uncover that despite a generally positive correlation, there are notable instances of divergence between correctness and perceived helpfulness in LLM generations, among other findings. Further, we garner a more granular understanding of GPT-4 mathematical problem-solving through a series of case studies, contributed by experienced mathematicians. We conclude with actionable takeaways for ML practitioners and mathematicians: models that communicate uncertainty, respond well to user corrections, and can provide a concise rationale for their recommendations, may constitute better assistants. Humans should inspect LLM output carefully given their current shortcomings and potential for surprising fallibility.

Concept-based models perform prediction using a set of concepts that are interpretable to stakeholders. However, such models often involve a fixed, large number of concepts, which may place a substantial cognitive load on stakeholders. We propose Selective COncept Models (SCOMs) which make predictions using only a subset of concepts and can be customised by stakeholders at test-time according to their preferences. We show that SCOMs only require a fraction of the total concepts to achieve optimal accuracy on multiple real-world datasets. Further, we collect and release a new dataset, CUB-Sel, consisting of human concept set selections for 900 bird images from the popular CUB dataset. Using CUB-Sel, we show that humans have unique individual preferences for the choice of concepts they prefer to reason about, and struggle to identify the most theoretically informative concepts. The customisation and concept selection provided by SCOM improves the efficiency of interpretation and intervention for stakeholders.