Decision-making in the presence of human and analytic bias

AI and the Intelligence Cycle, Part I

A decision maker’s greatest fear?

Bias.

Geospark Analytics set out to achieve two goals: to reduce the time spent sifting through information, and to empower leaders to make smarter, more impartial decisions.

Conclusions reached through subjective analytics bear dangerous implications. As humans, we weigh information more or less depending on our past experience with the subject matter, our level of confidence, and our personal beliefs, to name a few. The human mind seeks efficiency and predisposes us to subjective processing. By reducing the human biases that may perpetuate cognitive errors and influence decision-making, we can come to more rational conclusions. Today, one of the leaders in automating decision-making is artificial intelligence. Algorithms are coded to achieve goals by subtracting subjectivity. In the transient scene of foreign affairs and security, the capability to remove baseline partiality through automation is insurmountable.

Geospark Analytics created our machine learning platform, Hyperion, to reduce the impact of human and analytic biases in a number of ways.

Foremost, the platform does not distinguish among media and political ideologies. Geospark data scientists tuned the algorithms to analyze all sides of a story, from pro-western to anti-western outlets, conservative to liberal media, verified to fabricated sources. Principally, this is because any information source — regardless of its ideological stance and/or validity — drives human perception. Perception drives action, and action is reality. Neglecting information, consciously or not, can lead to blind spots in intelligence products and unknowingly force decision makers to act on incomplete information. Hyperion developers chose to incorporate diverse data sources to supply end users with the same information consumers see from their local newspapers and Twitter feeds.

An example of a warning message placed by Twitter on a Tweet during the U.S. 2020 Presidential Election. Misinformative data, flagged or not, has the power to influence political beliefs and unrest events.

Second, Hyperion reduces the time users spend filtering through information, thus reducing bias in the filtering process. At Geospark Analytics, we understand how one uses information differs according to individual objectives. End users leverage Hyperion’s tools, refining data through layers of analytical netting. They have more time, and more precise data, to use in subsequent tasks.

Event search results for April 26 through April 28, 2021 in Southern Chad. The user utilized Boolian Logic to narrow incoming media and news data to Unrest, Conflict, and Terrorism-related events that include the keywords ‘Rebel,’ ‘Militant,’ or ‘Invasion.’

Geospark Analytics’ tools improve five of the six stages of intelligence cycle that proceed decision-making. Given that each layer builds upon its predecessor, if any one stage is biased, the decision will also be biased. Hyperion tackles the Collection stage by incorporating more information than can be processed by the human brain, between 150,000 to 200,000 pieces of social and news media every day. Users can apply Hyperion’s analytical tools to filter out the extraneous, as well as to amalgamate the data into quantified, regional stability assessments. As a stability indicator model and platform for media filtering, Hyperion empowers users by giving them the raw information, the layers of analytical netting to refine that data, and an understanding of the risk and stability dynamics of any given region. Information collection becomes less biased and also less time-consuming. Decreasing the partiality in the gathering and initial analysis stages is imperative to achieve rational, less partial analyses down the road and eventually promote a well-informed decision.

Finally, Geospark Analytics’ data scientists are aware of the dangers of bias within machine learning models. Despite the many benefits of using machine learning in place of subjective analysis, algorithms too are created by human developers. Any application that leverages machine learning is susceptible to some level of inherent bias built into the algorithms, unintentionally giving biased recommendations to the end user. While no algorithm is can be bereft of bias, our scientists actively seek solutions that will minimize bias when constructing data sets and training the system. They incorporate a diverse range of data, compare outputs against statistical metrics for partiality, and constantly tweak the algorithms. We are committed to providing an objective platform for analysis, and we will continue to advance the algorithms towards a goal of impartial automation.

In a world overflowing with data, analysts do not have enough time — or objectivity — to incorporate every detail into an assessment. Leaders need to know that they can make a decision based off of the most impartial information and analyses available. Geospark Analytics provides users with extensive data and automated analytical tools to weaken subjectivity and promote smarter decision-making.

This blog is the first in a series that will dive into how Hyperion AI empowers end users and organization leaders to make smarter decisions, faster. In our next installment, you will hear from one of our data scientists about how we utilize the power of data and machine learning to produce on regional stability assessments and forecasts.