Power to the People: The Role of Humans in Interactive Machine Learning

By Saleema Amershi, Maya Cakmak, W. B. Knox, Todd Kulesza. The AI Magazine, 2014

1,228 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Safety Research, Software

View Paper: https://www.semanticscholar.org/paper/825ca26af5a2a510dbc1a7b97587212bc98ae968

How researchers see this paper

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

In Computer Vision and Pattern Recognition

In Safety Research

In Software

Citations in the corpus

Interactive Model Cards: A Human-Centered Approach to Model Documentation (2022, background): “Prior work has shown that non-expert analysts benefit from interacting with machine learning models and their data [1, 19, 59].”
The challenge of crafting intelligible intelligence (2018, positive): “Here, we expect to build on tools for interactive machine learning [1] and explanatory debugging [20, 21], which have explored interactions for adding new training examples, correcting erroneous labels in existing data, specifying new features, and modifying shape functions.”
Towards Accountable AI: Hybrid Human-Machine Analyses for Characterizing System Failure (2018, background): “Related efforts include research on interactive machine learning (Amershi et al. 2014; Ware et al. 2001) for building systems that interactively learn from their users.”
Learning from Interventions: Human-robot interaction as both explicit and implicit feedback (2020, background): “However, a key difference between our work and most other interactive learning [29] work is that we deal with demonstrations rather than explicit positive/negative labels, instead inferring the ad-”
Toward User-Driven Sound Recognizer Personalization with People Who Are d/Deaf or Hard of Hearing (2021, methodology): “In contrast to AutoML’s black box approach to model creation, interactive machine learning (IML) leverages end-users as “humans-in-the-loop” to iteratively engage in building and refning ML models [3, 19, 61, 70].”
Recent Advances in Robot Learning from Demonstration (2020, background): “In (51), several case studies are presented which illustrate the importance of studying the users of intelligent systems in order improve both user experience and robot performance.”
Shared autonomy via hindsight optimization for teleoperation and teaming (2017, background): “Prior works suggest that treating users as an oracle is frustrating (Amershi et al., 2014; Guillory and Bilmes, 2011), and this method naturally avoids it.”
“Everyone wants to do the model work, not the data work”: Data Cascades in High-Stakes AI (2021, background): “Machine Learning workflows are fundamentally iterative and exploratory in nature [7, 52, 71, 96].”
Deep Reinforcement Learning from Policy-Dependent Human Feedback (2019, neutral): “To address these concerns, we turn to the area of human-in-the-loop reinforcement learning (HRL) (Amershi et al., 2014), which mimics the traditional reinforcement-learning setting in all regards except for the speciﬁcation of learner feedback; in lieu of a hard-coded reward function, HRL…”
Visual Interaction with Deep Learning Models through Collaborative Semantic Inference (2019, background): “While some guidelines have been developed for interaction design for machine learning [2,25,75,86], they focus on designing machine learning methods where the model performs complementary tasks to the user, or where the user interacts with black-box models.”
Algorithmic bias: review, synthesis, and future research directions (2021, background): “In algorithmic environments, Amershi et al. (2014) promote the idea of deploying interactive ML systems to allow individuals to train algorithms by providing demonstrations and examples.”
Explanation-Based Human Debugging of NLP Models: A Survey (2021, background): “data labels after being given explanations (Amershi et al., 2014; Smith-Renner et al., 2020).”
How and What Can Humans Learn from Being in the Loop? (2020, background): “While new patterns of mass-training algorithms such as human-in-the-loop or interactive machine learning [1] have evolved that utilize human judgment to show algorithms where their errors lie, we appear to increasingly accept that algorithms need to make errors to become better.”
iBCM: Interactive Bayesian Case Model Empowering Humans via Intuitive Interaction (2015, background): “…limitations of indirect interaction with machine learning methods, there has been growing interest in the development of interactive machine learning systems [2], ranging from studying the theoretical aspect of interactive clustering [6] to building systems for use in specific domains [31] [23].”
NL-EDIT: Correcting Semantic Parse Errors through Natural Language Interaction (2021, background): “Kim, 2017) and learn interactively from their endusers (Amershi et al., 2014) can result in better user experiences and more effective learning systems.”
LanCon-Learn: Learning with Language to Enable Generalization in Multi-Task Manipulation (2021, background): “Through reinforcement, we offer an alternative in which the robot learns from a language command and a defined reward function, if corrective feedback from a human teacher is impractical [14], achieving over three-fold improvement in zero-shot skill transfer relative to baselines.”
Towards a Science of Human-AI Decision Making: A Survey of Empirical Studies (2021, background): “How can the two communities work better together? How can HCI research drive AI technical development? Such questions have been long contemplated on in other interdisciplinary areas such as interactive machine learning [5] and human-robot interaction [118].”