Creating a Strong Foundation for Responsible AI

Recently, the Office of the Director of National Intelligence (DNI) released a guide on the Principles of Artificial Intelligence (AI) Ethics for the Intelligence Community which set forward the high-level commitments that guide the Intelligence Community’s (IC) design, development, and use of AI.

The guide encourages the Intelligence Community to employ AI in a manner that:

• Follows the law and respects human dignity
• Is appropriately transparent and accountable
• Ensures that the augmentation of intelligence work by AI does not lead to the replacement of human judgment
• Is objective and equitable, with affirmative steps taken to identify and mitigate potential bias

In short, these Principles and the supporting Artificial Intelligence Ethics Framework for the Intelligence Community calls for the responsible use of AI through several guidelines to assist personnel within the IC in procuring, designing, building, using, protecting, consuming, and managing AI, machine learning (ML), and analytics efforts and related data.

While these act as a framework in guiding the IC, it should be noted that these principles also serve as a critical step towards ensuring ethical outcomes and should be considerations for any organization looking to drive the responsible deployment of AI, ML, and analytics.

Understand Goals and Risks

Any successful endeavor — from house building to digital transformation — requires the establishment of a strong foundation, and this holds true when deploying AI and ML. According to a report by Gartner, a key factor for any successful AI deployment is a strong level of data management and analytical maturity since there is a high dependency on reliable, high-quality data.¹

 However, this is dependent on an understanding of your current analytic situation. For instance, do you know what your strengths and weaknesses are regarding analytics? Are your processes hostage to legacy systems (i.e., spreadsheets), technology, data silos, or team alignment? If these are areas of concern, then some focus needs to be paid to building an analytics culture that focuses on breaking down traditional barriers between data scientists, IT, citizen data scientists, analysts, and domain experts. The emergence of unified analytic platforms, like the Alteryx APA Platform™, is helping organizations overcome these barriers to creating a strong analytics culture.

APA enables organizations to democratize data, automate processes, and upskill resources

Self-service analytic platforms, like Alteryx, include drag-and-drop capabilities, allowing you to deploy geospatial analysis, natural language processing, and predictive analytics into repeatable workflows. This gives your data science teams more time to focus on building and deploying AI and ML models and to address any risks across the lifecycle of these models. The Alteryx APA Platform enables organizations to democratize data, automate processes, and upskill resources with enhanced analytic capabilities, creating a natural and robust foundation for the responsible use of AI, ML, analytics, and related data.

Incorporate Human Judgment and Accountability

One key principle in the responsible AI framework is the concept of keeping humans in the loop, incorporating human judgment and accountability, and informing decisions appropriately. Since the deployment of AI, there has been a significant delineation between “black-box” and "clear-box” AI. While AI and ML can be trained to perform many tasks without humans in the loop, these systems can often operate in a black-box fashion, leaving it unclear as to how these machine-based decisions are made. The converse of this black-box approach is a “clear-box” approach that enables insight into how machine learning makes predictions.

The Alteryx APA Platform is built specifically with the concept of being human-centered, augmenting human capability regardless of one’s technical acumen. In other words, with Alteryx everyone can participate and benefit from a collaborative advanced-analytics environment. Even those who are not proficient in R or Python nor able to write their own models can take advantage of geo spatial, predictive, and ML-based analytic capabilities to collaborate, innovate, and solve. Specifically, with the Alteryx Intelligence Suite, an assisted modeling capability provides documented “clear-box” approach to understanding how predictive models interpret and analyze data, giving users of the platform a level of insight and confidence into the results that machine learning models are producing.

Maintain Transparency and Testing

Another key factor in deploying responsible AI is the requirement to maintain accountability and transparency for iterations, versions, and changes made to models. The Alteryx APA Platform is built upon enabling the open documentation of workflows and models. This includes understanding the source, quality, and lineage of data, and the certification and reliability of deployed models and the flexibility to share insights across multiple reporting, visualization, and BI platforms. With the Alteryx APA Platform, organizations within the IC have a unified platform to create actionable insights that can be shared and help propel outcomes.

One of the significant promises of AI and ML is the ability to harness and drive insights within unstructured data, from determining express sentiment in a social media post to modeling topics within textual-based information. Within the Alteryx Intelligence Suite, users get the ability to leverage automated building blocks that deliver capabilities for working with semi-structured and unstructured data through OCR recognition, sentiment analysis, and topic modeling. With these elevated capabilities, unstructured data agencies within the IC are going to have better access to actionable insights.

Mitigate Undesired Bias and Ensure Objectivity

Although not specifically stated as a guiding principle for responsible AI deployment, one key element that needs to be considered is time. When analytic teams are slowed down by manual, mundane, and repetitive processes, they can become stressed, overworked, or behind schedule. The Alteryx APA Platform enables your data teams to automate basic data gathering, cleaning, joining, and analysis functions.

The more time spent on prepping data for advanced analytics, AI, or ML, the less time is spent dedicated to the due diligence needed to ensure responsible analysis, including understanding possible inherent bias contained in data or previous analyses.

In a recent blog by Alan Jacobson, chief data and analytics officer at Alteryx, he writes that "When leveraging artificial intelligence and advanced analytic methods, we must be careful that inputs don’t bias the outcomes. In many cases, models are built based on historical data, and if these data include biases, they can propagate into future decision-making. In one now-famous ‘AI fail’ example, a tech company looking to automatically perform initial resume screening built a model based on the characteristics of historically ‘high-achieving’ employees. However, the inputs were flawed: the tech industry is heavily male-dominated, and their high-achieving employees were thus more likely to be male as a percentage.” As Jacobson explains, “Artificial intelligence doesn’t make moral judgements. It is not inherently biased, but historical data and the creators of the model could be.”

With a strong foundation and ethical framework for responsible AI and an Analytic Process Automation capability, the IC — and your organization — will be well-positioned to build the actionable insights needed to accelerate mission outcomes.  

To find out more read our white paper on “The Quest for Actionable Intelligence”. Connect directly on LinkedIn or Twitter.