Human Strategic Steering Improves Performance of Interactive Optimization

A central concern in an interactive intelligent system is optimization of its actions, to be maximally helpful to its human user. In recommender systems for instance, the action is to choose what to recommend, and the optimization task is to recommend items the user prefers. The optimization is done based on earlier user's feedback (e.g. "likes" and "dislikes"), and the algorithms assume the feedback to be faithful. That is, when the user clicks "like," they actually prefer the item. We argue that this fundamental assumption can be extensively violated by human users, who are not passive feedback sources. Instead, they are in control, actively steering the system towards their goal. To verify this hypothesis, that humans steer and are able to improve performance by steering, we designed a function optimization task where a human and an optimization algorithm collaborate to find the maximum of a 1-dimensional function. At each iteration, the optimization algorithm queries the user for the value of a hidden function f at a point x, and the user, who sees the hidden function, provides an answer about f(x). Our study on 21 participants shows that users who understand how the optimization works, strategically provide biased answers (answers not equal to f(x)), which results in the algorithm finding the optimum significantly faster. Our work highlights that next-generation intelligent systems will need user models capable of helping users who steer systems to pursue their goals.

[1]  Paulo S. C. Alencar,et al.  The use of machine learning algorithms in recommender systems: A systematic review , 2015, Expert Syst. Appl..

[2]  Jukka Corander,et al.  Bayesian inference of atomistic structure in functional materials , 2017, npj Computational Materials.

[3]  Devi Parikh,et al.  It Takes Two to Tango: Towards Theory of AI's Mind , 2017, ArXiv.

[4]  Daniel Szafir,et al.  Building Second-Order Mental Models for Human-Robot Interaction , 2019, ArXiv.

[5]  Nando de Freitas,et al.  Taking the Human Out of the Loop: A Review of Bayesian Optimization , 2016, Proceedings of the IEEE.

[6]  Samuel Kaski,et al.  Human-in-the-loop Active Covariance Learning for Improving Prediction in Small Data Sets , 2019, IJCAI.

[7]  Desney S. Tan,et al.  Interactive optimization for steering machine classification , 2010, CHI.

[8]  Cynthia Breazeal,et al.  A is for Artificial Intelligence: The Impact of Artificial Intelligence Activities on Young Children's Perceptions of Robots , 2019, CHI.

[9]  Pedram Daee,et al.  Interactive AI with a Theory of Mind , 2019, ArXiv.

[10]  Peter I. Frazier,et al.  A Tutorial on Bayesian Optimization , 2018, ArXiv.

[11]  D. Bates,et al.  Fitting Linear Mixed-Effects Models Using lme4 , 2014, 1406.5823.

[12]  Krishna Rajan,et al.  Information Science for Materials Discovery and Design , 2016 .

[13]  Peter I. Frazier,et al.  Bayesian optimization for materials design , 2015, 1506.01349.

[14]  Nando de Freitas,et al.  On correlation and budget constraints in model-based bandit optimization with application to automatic machine learning , 2014, AISTATS.

[15]  Samuel Kaski,et al.  Improving genomics-based predictions for precision medicine through active elicitation of expert knowledge , 2017, Bioinform..

[16]  Daniel A. Keim,et al.  What you see is what you can change: Human-centered machine learning by interactive visualization , 2017, Neurocomputing.

[17]  Andreas Holzinger,et al.  Interactive machine learning for health informatics: when do we need the human-in-the-loop? , 2016, Brain Informatics.

[18]  Francesco Ricci,et al.  A survey of active learning in collaborative filtering recommender systems , 2016, Comput. Sci. Rev..

[19]  Jonathan D. Nelson,et al.  Generalization guides human exploration in vast decision spaces , 2017, Nature Human Behaviour.

[20]  Andreas Krause,et al.  Information-Theoretic Regret Bounds for Gaussian Process Optimization in the Bandit Setting , 2009, IEEE Transactions on Information Theory.

[21]  Nando de Freitas,et al.  A Tutorial on Bayesian Optimization of Expensive Cost Functions, with Application to Active User Modeling and Hierarchical Reinforcement Learning , 2010, ArXiv.

[22]  Dorota Glowacka,et al.  Interactive Intent Modeling from Multiple Feedback Domains , 2016, IUI.

[23]  Maya Cakmak,et al.  Power to the People: The Role of Humans in Interactive Machine Learning , 2014, AI Mag..

[24]  Samuel Kaski,et al.  Interactive intent modeling , 2014, Commun. ACM.

[25]  Pedram Daee,et al.  Machine Teaching of Active Sequential Learners , 2019, NeurIPS.

[26]  Ali Borji,et al.  Bayesian optimization explains human active search , 2013, NIPS.

[27]  Carl E. Rasmussen,et al.  Gaussian processes for machine learning , 2005, Adaptive computation and machine learning.

[28]  Thomas L. Griffiths,et al.  Modeling human function learning with Gaussian processes , 2008, NIPS.

[29]  Yu Zhang,et al.  Plan Explanations as Model Reconciliation: Moving Beyond Explanation as Soliloquy , 2017, IJCAI.

[30]  Samuel Kaski,et al.  Knowledge elicitation via sequential probabilistic inference for high-dimensional prediction , 2016, Machine Learning.