Steps Toward Robust Artificial Intelligence

By Thomas G. Dietterich. 2017

175 citations recorded by the source corpus.

Recent advances in artificial intelligence are encouraging governments and corporations to deploy AI in high-stakes settings including driving cars autonomously, managing the power grid, trading on stock exchanges, and controlling autonomous weapons systems. Such applications require AI methods to be robust to both the known unknowns (those uncertain aspects of the world about which the computer can reason explicitly) and the unknown unknowns (those aspects of the world that are not captured by the system’s models). This article discusses recent progress in AI and then describes eight ideas related to robustness that are being pursued within the AI research community. While these ideas are a start, we need to devote more attention to the challenges of dealing with the known and unknown unknowns. These issues are fascinating, because they touch on the fundamental question of how finite systems can survive and thrive in a complex and dangerous world

Subfields: Artificial Intelligence, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety, Robustness & Interpretability

Citations in the corpus

The Next Decade in AI: Four Steps Towards Robust Artificial Intelligence (2020, background): “Davis and I discuss this briefly in Chapter X of Rebooting AI, and Tom Dietterich has an excellent discussion in his AAAI Presidential Address (Dietterich, 2017) that I discovered belatedly, after Rebooting AI came out.”
Wild Patterns: Ten Years After the Rise of Adversarial Machine Learning (2017, background): “…of the ﬁeld of adversarial machine learning and the emerging ﬁelds of robust artiﬁcial intelligence and inter-pretability of machine learning [101, 102], and we believe that these directions will help our society to get a more conscious understanding of the potential and limits of modern…”
Wireless Network Intelligence at the Edge (2018, background): “Unfortunately, such a one-time training process is vulnerable to initially unmodeled phenomena, as it is practically impossible to enumerate all preconditions and ensuing consequences, referred to as qualification and ramification problems, respectively [82].”
The challenge of crafting intelligible intelligence (2018, background): “In order to trust deployed AI systems, we must not only improve their robustness [5], but also develop ways to make their reasoning intelligible.”
Recent Advances in Open Set Recognition: A Survey (2018, background): “As discussed above, in real-world recognition/classiﬁcation, it is usually impossible to model everything [141], thus the OSR scenario is ubiquitous.”
Distribution-Free One-Pass Learning (2017, background): “This is also a crucial step towards robust artificial intelligence [12], [13].”
Stakeholders in Explainable AI (2018, methodology): “Unknown unknowns : these have recently been high-lighted as a key concern in AI system robustness (Dietterich 2017), with a variety of methods being proposed to deal with them, including employing a portfolio of models to mitigate against weaknesses in individual models, and creating AI systems…”
Heterogeneous Few-Shot Model Rectification With Semantic Mapping (2020, background): “AS MACHINE learning has been successfully applied to many real-world applications, whether a learner is robust or not has been attracting attentions [2], [3].”
Asking 'Why' in AI: Explainability of intelligent systems - perspectives and challenges (2018, methodology): “4 The ultimate goal of this work is to allow rapid incorporation of emergingML and interpretation techniques in a system that supports deployment of portfolio methods towards achieving robust AI decisionmaking (Dietterich, 2017), where users can tailor the system in terms of performance / explainabi”
Generalized Out-of-Distribution Detection: A Survey (2021, unknown): “A trustworthy visual recognition system should not only produce accurate predictions on known context, but also detect unknown examples and reject them (or hand them over to human users for safe handling) [1], [2], [3], [4], [5], [6], [7], [8].”
Open Set Learning with Counterfactual Images (2018, background): “However, many practical classification tasks may involve testing in the presence of “unknown unknown” classes not encountered during training [1].”
Assessing Generalization in Deep Reinforcement Learning (2018, background): “Generalization is often regarded as an essential characteristic of advanced intelligent systems and a central issue in AI research (Lake et al., 2017; Marcus, 2018; Dietterich, 2017).”
Learning and the Unknown: Surveying Steps toward Open World Recognition (2019, background): “As highlighted in a recent AI Magazine piece (Dietterich 2017), OSR is a growing subfield that that is critical for robust systems.”
Learning with Feature and Distribution Evolvable Streams (2020, background): “As the data are usually collected from open and dynamic environments, it is of great importance to facilitate the learning system with the capability of dealing with the environmental changes (Dietterich, 2017), which is also one of the key requirements of learnware (Zhou, 2016).”
SIN: Semantic Inference Network for Few-Shot Streaming Label Learning (2022, background): “Similar scenarios occur frequently in practice, such as new event detection [4], emerging disease surveillance [5], and tagging pictures on social networks [6], which need systems to be updated accurately according to the incoming new labels [7].”
Intelligible Artificial Intelligence (2018, methodology): “Yet it’s crucial that we be able to trust deployed systems, and this requires improving robustness and developing ways to make their reasoning intelligible [7].”
Rectify Heterogeneous Models with Semantic Mapping (2018, background): “As machine learning has been successfully applied in many real-world applications, the robustness of the learner is attracting more attention (Dietterich, 2017).”