CognitiveScale, an enterprise AI software vendor, released a beta version of its new product Cortex Certifai. It’s aimed at helping customers to identify, quantify, and manage risks inherent in AI applications. All without in-depth technical knowledge of AI and machine learning (ML). This is an industry-first product that addresses the need to increase transparency and trustworthiness of ML models and the AI applications they power.
ML models are, for the most part, inscrutable black boxes offering no explanation as to how the model has arrived at a particular recommendation. ML models are also inherently biased, and care should be taken to understand which groups of individuals can be disparately impacted before deploying such models into operation, as they can catastrophically affect individuals’ health and well-being, access to economic opportunities and quality of life (loans, education, housing, jobs), as well as liberty, and in some cases even life.
Cortex Certifai offers insights into these black boxes and increases transparency and trust by, among other things, evaluating key attributes that contribute to model accuracy, testing the edge cases, and asking questions such as:
It even suggests changes to the model profile, so that a person who has received an unfavorable outcome would know what to do to change that.
In addition to bias and explainability, Cortex Certifai evaluates the following areas of AI risk:
Image source: Cortex Certify product description. Accessed 16 Sept. 2019.
It generates a FICO-like “AI Trust Index” score consisting of six key elements of trust: fairness/bias, robustness, explainability, accuracy, compliance, and auditability. This score allows comparison of the model externally to the industry or internally for measurement and go-live criteria.
Cortex Certifai is integrated into the AI DevOps lifecycle and systems and is available as a stand-alone product or as a container-based Kubernetes application on all major cloud providers.
CognitiveScale is a privately-held company, and its investors include Norwest Venture Partners, Intel Capital, IBM Watson, M12 (Microsoft Ventures), and USAA. It is ranked #1 in AI patents among privately held companies and #4 overall behind IBM, Google, and Microsoft. The company’s award-winning Cortex Cognitive Platform is deployed by leading global financial services, as well as healthcare and digital commerce companies.
If you have or are developing ML applications, sign up for Certifai beta list and take it out for a ride. Keep in mind, though, that machine bias is not a technical problem, and that it should not be addressed via technology only. Engage humans as well as code to identify and mitigate machine biases in your systems.
If you are just pondering AI use cases for your organization, keep this product in mind for future use.
Cortex Certifai is a welcome development, as trust is the foundation of business, especially with new technologies such as ML and AI. And with this new product, CognitiveScale joins Google, IBM, and Microsoft who have already released or are developing solutions to check for and mitigate unwanted biases in ML models and the datasets used to train them. (This is also a very hot and active area of academic and industry research, so more products and capabilities are to come.)
It remains to be seen, however, whether Cortex Certifai will live up to its name – both the intelligence implied in the “cortex” and the assurance in the “certify”– but this is a tremendous step forward.