February 7, 2018 | Written by: Beth Smith
Categorized: News and Updates
Share this post:
Bringing transformative technology into the world is what IBM has done for more than a century. We have experienced first-hand how new innovations have fundamentally reshaped society, and recognize that powerful new technologies must be ushered into the world responsibly. People must trust that new innovations will have positive impact, and that trust must be earned through responsible stewardship.
AI is the latest revolutionary technology bringing responsibility to the forefront. IBM’s work on AI began more than 50 years ago, and today we are working with clients and partners to apply AI’s transformative potential across industries, from healthcare and finance to transportation and cybersecurity.
As an example, my team has been working for nine months to substantially increase the accuracy of the Watson Visual Recognition service. AI models improve with broader training data sets, advances in neural network learning techniques, and finer tuned algorithms. Based on a set of such improvements, our new model recognizes gender and skin tone with up to 10 times decrease in error-rates. On February 23, we will make this updated model available in the Watson Visual Recognition Service – with more robust facial recognition.
As we were improving the accuracy of this service, our team was contacted by academic researchers who had identified an issue with our Watson Visual Recognition service. Given our commitment to responsible use of AI technology including fairness, my team engaged with these researchers to fully understand and address their findings. See here for a technical analysis of the issue.
First, let me say – my IBM colleagues and I appreciate the contribution of these researchers to address the challenge of bias – not just in AI, but also in the data sets used to train AI systems.
IBM is committed to ensuring the quality, completeness and appropriateness of our AI training data and models. We recognize that bias can never be fully eliminated, and our work to mitigate it will never be complete. We therefore continually test our systems and find new data sets to better align their output with human values and expectations.
We have articulated a clear set of principles to guide the advancement of AI because we take responsibility seriously. That’s why we’ve dedicated months of time and expertise to improving Watson’s ability to recognize faces, and why we will continue working every day to enhance IBM’s full suite of AI technologies. Our clients, and society, deserve nothing less.
Proud to be an IBMer.