Tools for trust in AI

Share this post:

AI is becoming a pervasive technology affecting an increasing share of humanity’s critical decisions.  Whether it is a doctor asking an AI for targeted treatments, an autopilot in a car or a teacher for potential learning strategies for her class.

However there have also been flaws in the application of AI, luckily most of them with few consequences, but some with fatal outcomes.

AI applications have for many years been a black box, and are still in most cases a black box.  They are mostly based on large training data sets,  often applied in many layers of neural and deep neural networks making up the machine learning models. They are neither transparent, explainable and may contain bias from both ‘the humans’, that created them and trained them. Also the underlying training data may be skew and biased for a number of reasons, and may result in unfair outcomes.

In IBM we support transparency and data governance policies that will ensure people understand how an AI system came to a given conclusion or recommendation. Companies must be able to explain their algorithm’s recommendations. If they cannot, their systems shouldn’t be on the market.

In IBM we have developed several tools to assist AI designers  and developers in ensuring the fairness and explain-ability, which are readily accessible. Also we have developed a guide to assist AI designers and developers in making sure their AI applications are aligned with the users values, are explainable, fair and respect users data rights.

If you are concerned about, whether your AI is trustworthy, I can highly recommend that you check out these tools and the ‘Everyday Ethics for Artificial Intelligence’.

Because trust in AI is key……

AI OpenScale is an enterprise-grade environment that provide visibility into how AI is making decisions and give recommendations on how to mitigate any potentially damaging bias. It features a visually clear dashboard that line-of-business users can easily understand, reducing the burden of accountability from data scientists and empowering business users.

The AI Fairness 360 toolkit is a free open source software toolkit that can help detect and remove bias in machine learning models. It enables developers to use state-of-the-art algorithms to regularly check for unwanted biases from entering their machine learning pipeline and to mitigate any biases that are discovered.

The Adversarial Robustness Toolbox is an open-source software library to support researchers and developers in defending neural networks against adversarial attacks poisoning and “backdoor” attacks in machine learning models.

Guide:  Everyday Ethics for Artificial Intelligence.

Don’t hesitate to contact me at andersq@dk.ibm.com if you have any questions.