Consumers and policymakers have concerns about how artificial intelligence (AI), machine learning (ML) and the biases contained in algorithms might impact society.
Advertisers and publishers are concerned about exposure to risk, the ability to reach audiences, and the potential negative impact on reputation, sales and revenue, as well as the possibility of legal action due to non-compliance with laws and regulations, data privacy issues, or lack of governance.
Former Google Chairman Eric Schmidt earlier this month warned that AI in the future will be much more powerful than today, and will have the ability to manipulate people in ways that some might not have imagined.
Becoming aware of bias in advertising and marketing will become the first step to overcoming this challenge in technology.
Understanding Bias in AI for Marketing: A Comprehensive Guide to Avoiding Negative Consequences with Artificial Intelligence -- the second editorial release from The Interactive Advertising Bureau (IAB) AI Standards Working Group -- maps out artificial intelligence and machine-learning practices.
Thought leaders and expert practitioners from top publishers, advertising agencies, and ad tech companies have collaborated to produce this paper to provide insight for business executives, technologists, legal and compliance officers, and platform users. Some of them include Dentsu, IBM Watson Advertising, and ViacomCBS.
The report acknowledges that “bias is generally introduced into AI systems unintentionally by humans, but the duality of humans and machines makes bias detectable — and the risk mitigated helps companies do the right thing for their businesses and society.”
The online experience of billions of people worldwide is shaped by algorithms relying on AI and ML. In fact, some form of AI and ML is present for every search for content, video stream, or product shopping. These technologies increase the efficiency and accuracy of consumption, but also use behavioral data from historic sessions to determine the next move.
Bias “is a human cognitive condition earned over a lifetime of experiences,” but AI is “not inherently biased,” per the report. Still, the advertising industry must take responsibility for developing, deploying and managing platforms and technology built on AI and ML, especially when there’s a tendency for bias to creep into messages and campaigns.
The guide's insights come from conversations about real-world challenges faced daily by top-tier companies. Today’s future-thinking marketing and advertising technology leaders should be leaning into the development of their own processes and approaches.
As part of the best practices, the report suggests that companies pay attention to specific categories of implicit bias:
1. Unknown Unknowns: You may not be aware of a model’s bias until it presents itself in outputs, which is why your project team should view all proof-of-concept AI solutions.
2. Imperfect Processes: Test training sets are the same training sets used to develop.
3. Lack of Social Context: The concept and definition of the problem will vary across users requiring training and social alignment which reinforces the need for continuous collective alignment.
4. Define Fairness: You need to define what a fair world looks like and assign a quotient against that ideal state based on real-world production level variables.