The third installment of the quarterly Alation State of Data Culture Report was recently released, highlighting the data challenges enterprises face as they continue investing in artificial intelligence (AI). While AI is the number one data and analytics priority for respondents, nearly 90% are concerned about data quality issues leading to AI failures.

AI fails when it’s fed bad data, resulting in inaccurate or unfair results. Bad data, in turn, can stem from issues such as inconsistent data standards, data non-compliance, and a lack of data democratization, crowdsourcing, and cataloging. Unsurprisingly, survey respondents cited these specific issues as the main culprits behind their AI failures.

This latest report went beyond assessing organizations on the state of their data culture and key business drivers for data and analytics, and included an in-depth look at how organizations are deploying AI and the challenges inhibiting optimal results from those initiatives.

AI is a Top Priority, But Data Concerns Abound

Companies are expected to spend nearly US$23 billion annually on AI by 2024, and are naturally hoping for a substantial return on those investments, while worrying privately about whether those returns will materialize. Look at the latest Gartner Hype Cycle for AI to see why: nearly every point is on the “peak of inflated expectations.”

In this latest State of Data Culture survey, all respondents cite having deployed or planning to deploy AI, but only 8% report having deployed AI throughout the organization. Companies with a top-tier data culture, however, are more than twice as likely to be using AI throughout their organizations (18%) than their peers at less mature institutions. Since an organization likely wouldn’t expand a failing AI initiative, this suggests a correlation between having a good data culture and successfully implementing AI across an organization.

There are clearly some hurdles companies face in deploying AI, with data quality being a major concern, as 87% of respondents report being at least somewhat concerned about data quality impacting the success of their AI implementations. More than half of those respondents were very or extremely concerned.

When it comes to why AI fails, data leaders cite these top issues: inconsistent data collection standards (50%), compliance/privacy issues (48%), lack of data democratization (44%), and inadequate data infrastructure (44%).

Worries of Data Bias Creating Discriminatory AI Results

Skewed or unrepresentative training data leads to biased results. This is a fact, and the world is fast discovering the impact data bias has had in a wide array of machine learning applications, from criminal justice to healthcare to business.

So naturally, the survey found that nearly nine out of ten respondents are somewhat or more concerned about inherent biases being used in AI to produce discriminatory output. Top responses on how to combat bias in AI included better data literacy, cataloging data for visibility, and crowdsourced information (e.g. the tribal knowledge problem).

Those in the Know Recognize that a Solid Data Foundation and a Data Catalog are Critical to AI Success

Enterprises that have successfully deployed AI are more apt to be very or extremely concerned about data quality (50%) versus those who have not yet deployed AI (34%). Those data leaders who have deployed AI also cite incomplete data as the top issue that leads to AI failures. This is because when you go searching for data to create the models—be it for product innovation, operational efficiency, or customer experience—you uncover questions around the accuracy, quality, redundancy, and comprehensiveness of the data. Those who have deployed AI obviously understand this, which is why they cite cataloging data for visibility and access to available data (38%) and the ability to crowdsource information (38%) as two of the top three ways to combat bias, just behind better modeling skills (44%).

Ways to Combat Bias in AI

Collecting data from more varied sources is also a top response by both those who have and have not deployed AI, which reinforces the need for visibility, access, and crowdsourcing. One of the key driving reasons customers buy Alation is because they need to leverage AI and advanced analytics throughout their organization. They know our data intelligence platform helps solve for data literacy, data curation, data cataloging, and collaboration and crowdsourcing. It’s how we’re helping our customers achieve success with their AI initiatives.

Download the latest Alation State of Data Culture Report to learn more about improving the odds of success for your AI initiatives, building a culture of data-driven decision making, and how enterprises with top-tier data cultures are leveraging data to drive business value.

Alation is the Clear Leader in The Forrester Wave™: Machine Learning Data Catalogs, Q4 2020