“Societal bias is the type of bias that imposes a system’s values on others. Societal bias is a type of bias that reflects the assumptions, norms, or values of a specific society or culture. Societal bias can affect the fairness and ethics of AI systems, as they may affect how different groups or domains are perceived, treated, or represented by AI systems. For example, societal bias can occur when AI systems impose a system’s values on others, such as using Western standards of beauty or success to judge or rank people from other cultures.”

“Poor data quality is a potential reason for not seeing accurate results from an AI model. Poor data quality means that the data is inaccurate, incomplete, inconsistent, irrelevant, or outdated for the AI task. Poor data quality can affect the performance and reliability of AI models, as they may not have enough or correct information to learn from or make accurate predictions.”

The significance of the explainability of trusted AI systems is that it describes how AI models make decisions. Explainability is crucial for building trust and accountability in AI systems, ensuring that users and stakeholders understand the decision-making processes and outcomes generated by AI. This is particularly important in scenarios where AI decisions impact personal or financial status, such as in credit scoring or healthcare diagnostics. Salesforce emphasizes the importance of explainable AI through its ethical AI practices, aiming to make AI systems more transparent and understandable. More details about Salesforce’s approach to ethical and explainable AI can be found in Salesforce AI ethics resources at Salesforce AI Ethics.

Natural language processing (NLP) is the type of AI that can enhance customer service agents' email responses by analyzing the written content of previous emails. NLP technologies interpret and generate human language, allowing AI systems to understand, respond to, and even anticipate customer needs based on email interactions. This capability helps in crafting more relevant, accurate, and personalized email responses, improving customer service quality. Salesforce utilizes NLP in its Einstein AI platform to augment various customer service functions. More about Salesforce Einstein’s NLP capabilities can be found on the Salesforce Einstein page at Salesforce Einstein NLP.

“Creating guardrails that mitigate toxicity and protect PII is an action that should be taken to develop and implement trusted generative AI with Salesforce’s safety guideline in mind. Salesforce’s safety guideline is one of the Trusted AI Principles that states that AI systems should be designed and developed with respect for the safety and well-being of humans and the environment. Creating guardrails means implementing measures or mechanisms that can prevent or limit the potential harm or risk caused by AI systems. For example, creating guardrails can help mitigate toxicity by filtering out inappropriate or offensive content generated by AI systems. Creating guardrails can also help protect PII by masking or anonymizing personal or sensitive information generated by AI systems.”

“The most likely impact that high-quality data will have on customer relationships is improved customer trust and satisfaction. High-quality data means that the data is accurate, complete, consistent, relevant, and timely for the AI task. High-quality data can improve customer relationships by enabling AI systems to provide personalized and relevant products, services, or solutions that meet the customers’ expectations, needs, and interests. High-quality data can also improve customer trust and satisfaction by reducing errors, delays, or waste in customer interactions.”

When inconsistencies in data are identified, the most appropriate action is to investigate these inconsistencies and apply data quality techniques. Adjusting the AI model to accommodate poor quality data or simply increasing the quantity of data without addressing the underlying issues does not solve the problem and can lead to less reliable AI outputs. Proper data cleaning, normalization, and validation are necessary steps to ensure that the data fed into an AI model is accurate and reliable, thus enhancing the model’s performance. Salesforce provides guidelines on how to manage and improve data quality, including practical steps for addressing data inconsistencies, detailed at Improving Data Quality in Salesforce.