Ad Bias And Artificial Intelligence

Combatting Ad Bias

Demographics are typically a key setting for targeting.  Advertisers want to target people who are in their "target market" for their product. However, sometimes this targeting can lead to bias in advertising.  Advertisers might target ads based on age, gender, race, or other factors. This can result in some people seeing ads that are not relevant to them or that are offensive. It can result in people falling into "buckets" that they don't belong in.

Is AI to Blame for Ad Bias?

Although AI is a hot topic, AI isn't the reason that bias exists. The bias exists as the baseline of the training of algorithms used for targeting and advertising. Fortunately, by using forms of AI and algorithms, we are also able to measure where bias occurs.

"Ethics and bias are an important issue. I think our team (Pinterest) is driving this - we've launched additional filters for diversification in our algorithm training data. One of the first steps there is understanding and measuring those aspects of your training data, and understanding what types of unintended biases there are in your algorithms due to your training data." Said Chuck Rosenberg, Head of Advanced Technology Group at Pinterest

There are a few ways to measure bias:

-Remove protected groups from the data set and see if performance changes

-Add in random noise to the features to see if the model's output changes

-Compare the distribution of outcomes for different groups. If there is a difference, that could be evidence of bias

- Check for demographic bias by segmentation

How Can We Fix Ad Bias

There are a few things that companies can do to combat bias in advertising:

1. Diversify your data: Make sure that your data represents a diverse range of people. This will help to ensure that your algorithms are not biased against any particular group.

2. Be aware of the types of bias: There are many different types of bias. Be aware of the types of bias that could affect your algorithms and take steps to mitigate them.

3. Evaluate your algorithms: Regularly evaluate your algorithms to ensure that they are not biased. This can be done using a variety of methods, such as auditing or testing with diverse data sets.

4. Take action:  If you find that your algorithms are biased, take action to fix the problem. This might involve changing your training data, altering your algorithms, or both.

Discriminatory advertising is a big problem because it can result in people being treated unfairly and can cause harm. However, there are things that companies can do to combat this problem. By diversifying their data, being aware of different types of bias, and taking action to fix the problem, companies can help to ensure that their algorithms are not discriminatory.

“We have to eliminate this bias, because it’s no good for society,” said WPP boss Mark Read, in an AdWeek article. “There’s also a business case, if you’re biased you’re missing out on [commercial opportunities] and we don’t want to miss out on anything.”

What Else Can We Do?

One key way we could help to eliminate demographic-based biases would be to make the switch entirely to contextual advertising. By eliminating the targeting settings that are largely biased (age, gender, race, socio-economic status), advertisers are left to choose content categories or keywords that are related to their products. This would allow the algorithms to be trained for content vs. people.

At CatapultX, we're doing just that.