Ensure the AI training data is accurate and includes all variables to lower bias

New Ideas in MarketingEssential news for marketers, summarised by YouGov
March 28, 2022, 4:59 AM GMT+0

A diverse set of scenarios and outcomes in the training data can aid in developing AI models optimised for rare events as well as promoting new products.

To reduce bias in AI models, marketers should ask specific questions when issuing requests for proposals (RFPs) or requests for information (RFIs) to AI vendors. While AI platforms provide a black box wherein several scenarios are presented, they often include ambiguity and the risk of unchecked bias.

Marketers that want to utilise their audience data to train and test the AI algorithm in DSPs should ensure the data is accurate, as faulty data can result in substandard AI models. Check for consistency by comparing the data sources and validating them against "truth sets."

Brands must run their own tests by using key subgroups as well as inputs and outputs to assess the impact of AI bias. Conduct tests with subgroups like old versus new consumers, as well as people from subgroups to train the AI models without biases. Test inputs and outputs like high-or low-consideration products and conversion-optimised ads to optimise AI models with lower bias.

Discover the top organisations in your market and industry that have customers buzzing using YouGov BrandRankings.

Read the original article

[7 minute read]

Explore more data & articles