AI is trained on a large amount of data that has been collected over time. If that data collected shows bias, or is not representative of the people the system will impact, it can amplify those biases. For example, the recently launched BlenderBot 3<\/a>, a conversational AI, perpetuated negative bias by generating unsafe and offensive remarks during a public demo. Research has also shown that many popular open-source benchmark training datasets \u2014 ones that many new machine learning models are measured against \u2014 are either not valid for the context<\/a> in which they have been widely reused or contain data that is inaccurate or mislabeled<\/a>.<\/p>\n\n\n\n Einstein AI helps to empower your teams with built-in intelligence to engage with empathy, increase productivity, and scale customer experiences.<\/p>\n\t\t\t\n\t\t\t\n\t\t\t\t\t\t\t <\/p>\n\n\n\n As a result, governments around the world have begun drafting and implementing AI regulations. Singapore has a law focused on accountable and responsible development of AI<\/a>, while Thailand has most AI initiatives embedded within policies and strategies to strengthen the development of AI-related technologies.<\/p>\n\n\n\n Regulation, when well-crafted and appropriately applied, ensures an ethical AI system that is inclusive and unbiased. <\/p>\n\n\n\n In the meantime, business leaders need to pave the way to a more equitable AI infrastructure. Here\u2019s how:<\/p>\n\n\n\n To increase trustworthiness, many regulations require businesses to be transparent about how they trained their AI model (a program trained on a set of data to recognise certain types of patterns), the factors used in the model, its intended and unintended uses, and any known bias. Policymakers usually request this in the form of data sheets<\/a> or model cards, which act like nutrition labels for AI models. Salesforce, for example, publishes its model cards<\/a> so customers and prospects can learn how the models were trained to make predictions. <\/p>\n\n\n\n Why does an AI system make the recommendation or prediction it does? You might shrug when AI recommends you watch a new movie, but you\u2019ll definitely want to know how AI weighed the pros and cons of your loan application. Those explanations need to be understood by the person receiving the information \u2014 such as a lender or loan officer \u2014 who then must decide how to act upon the recommendation an AI system is making.<\/p>\n\n\n\n That said, one study<\/a> conducted by researchers at IBM Research AI, Cornell University, and Georgia Institute of Technology found that even people well versed in AI systems often over-relied on and misinterpreted the system\u2019s results. Misunderstanding how the AI systems work can result in disastrous consequences in situations that require more human attention. The bottom line? More real-life testing needs to occur with the people using the AI to ensure they understand the explanations.<\/p>\n\n\n\n Some regulations call for a human to make the final decision about anything with legal or similarly significant effects, such as hiring, loans, school acceptance, and criminal justice recommendations. By requiring human review rather than automating the decision, regulators expect bias and harm to be more easily caught and mitigated. <\/p>\n\n\n\n In some cases, humans may defer to AI recommendations rather than rely on their own judgment. Combine this tendency with the difficulty of grasping the rationale behind an AI decision, humans don\u2019t actually provide that safety mechanism against bias.<\/p>\n\n\n\n That doesn\u2019t mean high-risk, critical decisions should simply be automated. It means we need to ensure that people can interpret AI explanations and are incentivised to flag a decision for bias or harm. For instance, the AI governance framework<\/a> in Singapore, along with its personal protection data policy, allows AI to make recommendations for decisions with legal impact (for example, a loan approval or rejection) but requires a human to make the final decision. This not only promotes AI adoption, but also builds customer confidence and trust.<\/p>\n\n\n\n It\u2019s understandable, then, that some executives may be reluctant to collect sensitive data \u2014 age, race, and gender, for example \u2014 to begin with. Some worry about inadvertently biasing their models or being unable to properly comply with privacy regulations. However, you cannot achieve fairness through inaction. Companies need to collect this data in order to analyse if there is disparate impact for different subpopulations. This sensitive data can be stored in a separate system where ethicists or auditors can access it for the purpose of bias and fairness analysis. <\/p>\n\n\n\n In a perfect world, AI systems would be built without bias, but this is not a perfect world. While some guardrails can minimise the effects of AI bias on society, no dataset or model can ever be truly bias free. Statistically speaking, bias means \u201cerror,\u201d so a bias-free model must make perfect predictions or recommendations every time \u2014 and that just isn\u2019t possible. <\/p>\n\n\n\n There is a lot to understand to ensure companies are creating and implementing AI responsibly and in compliance with regulations. This requires business leaders to build ethical AI practices.<\/p>\n\n\n\n Executives can start by hiring a diverse group of experts from many backgrounds in ethics, psychology, ethnography, critical race theory, and computer science. They can build on that by creating a company culture that rewards employees for flagging risks \u2014 empowering them to not only ask, \u201cCan we do this?\u201d but also, \u201cShould we do this?\u201d \u2014 and by implementing consequences when harms are ignored.<\/p>\n\n\n\n Prepare for the future of AI. Find out how.<\/i><\/p>\n\n\n\n DOWNLOAD REPORT<\/a><\/i><\/p>\n\n\n\n This post originally appeared on the U.S.-version of the Salesforce blog.<\/a><\/i><\/p>\n\n\n\nInnovate with AI <\/span> <\/h2>\n\t\t\t\t\t\t\t
![\"\"](\"https:\/\/www.salesforce.com\/ap\/blog\/wp-content\/themes\/salesforce-blog\/dist\/images\/offer-block\/offer-illustration-layout-one.png\")
\n\n\t\t\n\t\t\t\t\t![\"\"](\"https:\/\/www.salesforce.com\/ap\/blog\/wp-content\/themes\/salesforce-blog\/dist\/images\/offer-block\/offer-accent-layout-one.png\")
\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\t\t\t![\"\"](\"https:\/\/www.salesforce.com\/ap\/blog\/wp-content\/themes\/salesforce-blog\/dist\/images\/offer-block\/offer-cloud-layout-one.png\")
\n\t\t\n\t<\/div>\n\n\n\nWays to combat bias in AI <\/span> <\/h3>\n\n\n\n
Be transparent <\/span> <\/h2>\n\n\n\n
Make your AI interpretable <\/span> <\/h2>\n\n\n\n
Keep a human in the loop <\/span> <\/h2>\n\n\n\n
How to add ethical AI practices to your business <\/span> <\/h2>\n\n\n\n