AI Ethics and Transparency for Business

Digital Marketing - Study Notes:

Biases in AI

Artificial intelligence (or AI) can only be as effective as the data it draws on. In other words, AI is not inherently neutral. Biases can be embedded during data collection, algorithmic design, or even in the interpretation of results. Organizations need to be aware of the potential for racial, gender, or socioeconomic biases in AI-driven strategies. If the data is infected with these biases, its output will be similarly infected. This can lead to the unintended (or even deliberate) perpetuation of harmful biases, such as gender or racial biases.

Origins of gender and racial biases in AI

Historical data

One of the primary sources of bias is historical data. If the data used to train an AI system includes historical gender or racial biases, the AI is likely to perpetuate these biases. For example, if a job recommendation algorithm is trained on data from an industry that has historically been male-dominated, the algorithm may continue to recommend men over women for job positions in that field.

Human prejudices

Biases can also be introduced by the humans who design and deploy these systems. If the team behind an AI project lacks diversity, there's a higher chance that the system will reflect the biases of that team.

Societal norms

Societal norms and stereotypes can also find their way into AI systems. For instance, facial recognition technologies have been found to misidentify people of certain ethnic backgrounds at a higher rate than others. This is often because the data sets used for training did not have a diverse range of faces.

These biases can lead to concerns around:

• Ethics
• Transparency
• Privacy, and…
• Accountability

Let’s explore each of these in more detail.

Ethics

The use of AI in business brings up several ethical concerns. This includes the manipulation of user data and behavior, consent, and the ‘creepiness factor’ in hyper-targeted ads. At what point do highly personalized ads take on a whiff of digital stalking?

Business professionals need to uphold the highest ethical standards when employing AI. In particular, AI tools should never be used to mislead, misinform, or harm people.

Transparency

As AI-generated content becomes more common, questions about transparency, plagiarism, and authorship and IP ownership become increasingly relevant. Businesses should disclose the use of AI where appropriate and be very cautious about inadvertently plagiarizing content generated by AI. For example, The Guardian and Bloomberg use AI to generate news articles, but they are transparent about it.

Being transparent brings practical benefits to brands. Customers will appreciate being told that content hasn’t been generated by AI and that brands are still making the effort to deliver unique, high-quality, and personalized content. Creating content with the use of AI can become a value proposition for organizations. On the other hand, if brands do use AI to generate content, customers will appreciate being told this.

Privacy

The use of AI often involves the collection and analysis of large sets of customer data. Organizations must adhere to privacy regulations such as GDPR, CCPA, and new AI-focused laws such as the Artificial Intelligence Act in the EU. Knowing what data can be legally collected and how to protect it is crucial. They should also adhere to any sector-specific regulations – such as HIPAA in the healthcare sector.

Accountability

When employing AI in business, brands must establish a system of accountability and governance. This involves setting up clear protocols for data handling, algorithmic decision-making, and auditing. Business professionals should know who is responsible for each aspect of an AI project and have a framework in place for ethical review and oversight.

AI issues in action

Keeping ethical and legal concerns front and center is imperative when implementing AI in digital business.

Here are some key guidelines that you should follow.

• Disclosure: Always disclose the use of AI to consumers where applicable.
• Compliance: Ensure compliance with data protection regulations like GDPR.
• Audit: Audit AI algorithms for any inherent biases and correct them.

How might organizations address these ethical concerns when deploying AI?

Let’s say you are setting up a generative AI chatbot to manage queries on an ecommerce site. You could follow these steps:

• Documentation: You document the type of data the chatbot will collect and how it will be used.
• Review: The internal ethics board reviews and approves the chatbot deployment, provided that users are informed clearly that they are interacting with a machine and have an option to opt out.
• Audit: An audit trail is set up to log all interactions the chatbot has with users.
• Monitoring: The chatbot is programmed to flag conversations where users express strong emotions like frustration or anger, to be reviewed manually by a human agent.
• Disclosure: A disclaimer is added on the website, transparently stating how the chatbot will use data.
• Compliance: Finally, the data storage and processing steps are reviewed to make sure they comply with GDPR (or local equivalent) regulations.

For example, Microsoft has set up its own AI, Ethics, and Effects in Engineering and Research (AETHER) Committee to handle these questions.