Trusted AI

IBM Research is building and enabling AI solutions people can trust

As AI advances, and humans and AI systems increasingly work together, it is essential that we trust the output of these systems to inform our decisions. Alongside policy considerations and business efforts, science has a central role to play: developing and applying tools to wire AI systems for trust. IBM Research’s comprehensive strategy addresses multiple dimensions of trust to enable AI solutions that inspire confidence.

Robustness

We are working to ensure the security and reliability of AI systems by exposing and fixing their vulnerabilities: identifying new attacks and defense, designing new adversarial training methods to strengthen against attack, and developing new metric to evaluate robustness.

View publications

Fairness

To encourage the adoption of AI, we must ensure it does not take on and amplify our biases. We are creating methodologies to detect and mitigate bias through the life cycle of AI applications.

View publications

Explainability

Knowing how an AI system arrives at an outcome is key to trust, particularly for enterprise AI. To improve transparency, we are researching local and global interpretability of models and their output, training for interpretable models and visualization of information flow within models, and teaching explanations.

View publications

Lineage

Lineage services can infuse trust in AI systems by ensuring all their components and events are trackable. We are developing services like instrumentation and event generation, scalable event ingestion and management, and efficient lineage query services to manage the complete lifecycle of AI systems.

View publications

IBM Researcher Sashka

AI Factsheets

In a new paper, IBM scientists suggest that AI services be accompanied with a factsheet outlining the details about how it operates, how it was trained and tested, its performance metrics, fairness and robustness checks, intended uses, maintenance, and other critical details.

IBM Researcher Kush

Reducing discrimination in AI with new methodology

IBM Research introduces a methodolgy to reduce the bias in a training dataset, such that any AI algorithm that later learns from that dataset will perpetuate as little inequity as possible.

IBM Researcher Sashka

Reducing discrimination in AI with new methodology

IBM Research introduces a methodolgy to reduce the bias in a training dataset, such that any AI algorithm that later learns from that dataset will perpetuate as little inequity as possible.

IBM Researcher Sashka

AI bias will explode. But only unbiased AI will survive

IBM researcher Francesca Rossi discusses bias and the importance of building AI systems free of it.

IBM Researcher Sashka

Explanations based on what’s missing

An algorithm that can identify missing elements to improve the explanations generated by AI technologies.

IBM Researcher Sashka

Robustness metric for AI models

Scoring the robustness of different neural networks against adversarial attack to help build more reliable AI systems.

IBM Researcher Sashka

Try it yourself: How models effect the US court outcomes

Learn how models are deciding the future of our country's defendants through this interactive experience.

IBM Researcher Sashka

Mitigating Bias in AI Models

IBM has been working to substantially increase the accuracy of its Watson Visual Recognition service for facial analysis.

Projects and resources

AI bias presentation

IBM researcher Francesca Rossi discusses bias and the importance of building AI systems free of it.

Comparison of our CEM versus LRP and LIME on MNIST.

An algorithm that can identify missing elements to improve the explanations generated by AI technologies.

A CLEVER Way to Resist Adversarial Attack

Scoring the robustness of different neural networks against adversarial attack to help build more reliable AI systems.

Experiment Reducing Bias in AI

Learn how models are deciding the future of our country's defendants through this interactive experience.

Mitigating Bias in AI Models

IBM has been working to substantially increase the accuracy of its Watson Visual Recognition service for facial analysis.

Increasing Trust in AI Services through Supplier's Declarations of Conformity

M. Hind, S. Mehta, A. Mojsilović, R. Nair, K. N. Ramamurthy, A. Olteanu and K. R. Varshney

Preferences and Ethical Principles in Decision Making

A. Loreggia, N. Mattei, F. Rossi and K. B. Venable

A Notion of Distance Between CP-nets

A. Loreggia, N. Mattei, F. Rossi and K. B. Venable

On the Distance Between CP-nets

A. Loreggia, N. Mattei, F. Rossi and K. B. Venable

On Building Efficient Temporal Indexes on Hyperledger Fabric

H. Gupta, S. Hans, S. Mehta and P. Jayachandran

Efficiently Processing Temporal Queries on Hyperledger Fabric

H. Gupta, S. Hans, K. Aggarwal, S. Mehta, B. Chatterjee and P. Jayachandran

An End-To-End Machine Learning Pipeline That Ensures Fairness Policies

S. Shaikh, H. Vishwakarma, S. Mehta, K. R. Varshney, K. N. Ramamurthy and D. Wei

Provenance in Context of Hadoop as a Service (HaaS) - State of the Art and Research Directions

H. Gupta, S. Mehta, S. Hans, B. Chatterjee, P. Lohia and C. Rajmohan

Collaborative Human-AI (CHAI): Evidence-Based Interpretable Melanoma Classification in Dermoscopic Images

N. C. F. Codella, C.-C. Lin, A. Halpern, M. Hind, R. Feris and J. R. Smith

Seq2Seq-Vis: A Visual Debugging Tool for Sequence-to-Sequence Models

H. Strobelt, S. Gehrmann, M. Behrisch, A. Perer, H. Pfister and A. M. Rush

Interpretable to Whom? A Role-based Model for Analyzing Interpretable Machine Learning Systems

R. Tomsett, D. Braines, D. Harborne, A. Preece and S. Chakraborty

Improving Simple Models with Confidence Profiles

A. Dhurandhar, K. Shanmugam, R. Luss and P. Olsen

EAD: Elastic-Net Attacks to Deep Neural Networks via Adversarial Examples

Pin-Yu Chen, Y. Sharma, H. Zhang, J. Yi, Cho-Jui Hsieh

Why Interpretability in Machine Learning? An Answer Using Distributed Detection and Data Fusion Theory

K. R. Varshney, P. Khanduri, S. Zhang, P. Sharma and P. K. Varshney

Explanations based on the Missing: Towards Contrastive Explanations with Pertinent Negatives

A. Dhurandhar, P.-Y. Chen, R. Luss, C.-C. Tu, P. Ting, K. Shanmugam and P. Das

Analyze, Detect and Remove Gender Stereotyping from Bollywood Movies

N. Madaan, S. Mehta, T. S. Agrawaal, V. Malhotra, A. Aggarwal, Y. Gupta and M. Saxena

Fairness GAN

P. Sattigeri, S. C. Hoffman, V. Chenthamarakshan and K. R. Varshney

Protecting Intellectual Property of Deep Neural Networks with Watermarking

J. Zhang, Z. Gu, J. Jang, H. Wu, M. Ph. Stoecklin, H. Huang and I. Molloy

Variational Inference of Disentangled Latent Concepts from Unlabeled Observations

Abhishek Kumar, Prasanna Sattigeri, Avinash Balakrishnan

Evaluating the Robustness of Neural Networks: An Extreme Value Theory Approach

Tsui-Wei Weng, Huan Zhang, Pin-Yu Chen, Jinfeng Yi, Dong Su, Yupeng Gao, Cho-Jui Hsieh, Luca Daniel

Optimized Pre-Processing for Discrimination Prevention

F. P. Calmon, D. Wei, B. Vinzamuri, K. N. Ramamurthy, and K. R. Varshney

Teaching Meaningful Explanations

N. C. F. Codella, M. Hind, K. N. Ramamurthy, M. Campbell, A. Dhurandhar, K. R. Varshney, D. Wei and A. Mojsilovic

 

Science for Social Good

Learn how IBM and its partners are applying AI, cloud and deep science toward societal challenges.

Learn more