Business leaders, academics, policy makers, and countless others are exploring ways to harness generative AI technologies that have the potential to transform how we learn, work, and more. In business, generative AI has the potential to transform the way companies interact with their customers and drive business growth. A new survey shows that 67% of senior IT leaders are prioritizing generative AI in their business in the next 18 months, with a third (33%) citing it as a top priority. Companies are looking to see how it impacts every part of their business including sales, customer service, marketing, commerce, IT, legal, HR and more.
But senior IT leaders want a reliable and data-protected way for their employees to use these technologies. His 79% of senior IT leaders reported being concerned that these technologies pose potential security risks, and an additional 73% were concerned about biased results. More broadly, organizations should recognize the need to ensure the ethical, transparent and responsible use of these technologies.
A business using generative AI technology in an enterprise environment is different than a consumer using it for personal purposes. Companies must comply with regulations relevant to their respective industries (e.g. healthcare) and have legal, financial and ethical implications if the content generated is inaccurate, inaccessible or offensive. There is a minefield to receive. For example, if a generative AI chatbot instructs the wrong steps to cook a recipe, the risk of harm is much lower than if a field service worker instructs a worker to repair heavy equipment. If not designed and deployed with clear ethical guidelines, generative AI can have unintended consequences and cause real harm.
Organizations need to understand how generative AI is used and how generative AI impacts sales, marketing, commerce, service, and IT work, aligning generative AI goals with the business “to do”. We need a clear and workable framework for how.
In 2019, we published Trusted AI Principles (Transparency, Fairness, Responsibility, Accountability, Trustworthiness) to guide the development of ethical AI tools. These apply to any organization investing in AI. However, these principles are only valid if an organization lacks ethical AI practices to practice AI technology development and deployment. Mature and ethical AI practices integrate disciplines such as product management, data science, engineering, privacy, legal, user research, design and accessibility to reduce potential harm and maximize societal benefit. and operationalize its principles and values through responsible product development and deployment. Benefits of AI. There are models for how organizations can start, mature, and scale these practices, providing a clear roadmap for how to build an infrastructure for ethical AI development.
However, with the mainstream emergence and accessibility of generative AI, we realized that organizations needed guidelines specific to the risks posed by this particular technology. These guidelines are not a substitute for our principles, but serve as a north star for how companies can operate and practice them as they develop products and services that use this new technology.
Guidelines for Ethical Development of Generative AI
Our new set of guidelines will help organizations assess the risks and considerations of generative AI as generative AI tools become mainstream adoption. These cover five focus areas.
Accuracy
Organizations can run AI models on their own data to provide verifiable results that balance accuracy, precision, and recall (the model’s ability to correctly identify positive cases within a given dataset). Must be able to train. It’s important to communicate any uncertainty about the AI’s generative response so that people can validate it. It cites the sources from which the model draws information to create content, explains why the AI made such responses, highlights uncertainties, and explains how some tasks are fully automated. This can be achieved by creating guardrails that prevent
safety
Making every effort to mitigate bias, toxicity, and harmful output by conducting bias, explainability, and robustness assessments is always a priority in AI. Organizations must protect the privacy of personally identifiable information contained in data used for training to prevent potential harm. In addition, security assessments help organizations identify vulnerabilities that could be exploited by malicious parties (e.g., the ‘do anything now’ used to override ChatGPT’s guardrails). prompt injection attack).
honest
When collecting data for model training and evaluation, respect the origin of the data and make sure you have consent for its use.. This can be achieved by leveraging open-source and user-provided data. Also, when delivering output autonomously, it should be transparent that an AI created the content. This can be done through a watermark on the content or an in-app message.
empowerment
Sometimes it’s best to fully automate the process, but AI often plays a supporting role. Generative AI is great today. assistant. In industries where trust-building is a top priority, such as finance and healthcare, AI models can potentially provide data-driven insights that can be used by humans to build trust and maintain transparency. Involvement in decision making is important. Additionally, make sure the model’s output is accessible to everyone (e.g. generate ALT text to accompany an image, and the text output is accessible to the screen her reader). And of course, content contributors, authors, and data labelers must be treated with respect (e.g., fair wages, consent to use their work, etc.).
sustainability
Language models are described as “large” based on the number of values or parameters they use. Some of these Large Language Models (LLMs) have hundreds of billions of parameters and use a lot of energy and water to train them. For example, GPT3 required 1.287 gigawatt hours, or about the same amount of electricity and 700,000 liters of clean fresh water as he would power 120 US homes for a year.
When considering AI models, bigger is not necessarily better. When developing our own model, we strive to minimize model size while maximizing accuracy by training the model on large volumes of high-quality CRM data. This helps reduce your carbon footprint by reducing the amount of computation required and reducing energy consumption and carbon footprint from your data center.
Generative AI integration
Most organizations integrate generative AI tools rather than building their own AI tools. Here are some tactical tips for safely integrating generative AI into your business applications to improve business outcomes.
Use zero-party or first-party data
Companies should use zero-party data (data that customers actively share) and first-party data collected directly by customers to train generative AI tools. Strong data provenance is key to ensuring models are accurate, original, and authoritative. Relying on third-party data, or information obtained from external sources, to train an AI tool makes it difficult to ensure that the output is accurate.
For example, data brokers can have outdated data, incorrectly combine data from devices or accounts that don’t belong to the same person, or make inaccurate inferences based on data. This is true when you base your model on your data. So in Marketing Cloud, personalization can be wrong if all the data in a customer’s CRM is from a data broker.
Keep your data fresh and properly labeled
AI performance is determined by the data used to train it. Models that generate responses to customer support queries will produce inaccurate or stale results if the underlying content of the model is outdated, incomplete, or inaccurate. This can lead to hallucinations in which the tool confidently claims that the falsehood is real. Using biased training data creates a tool that propagates the bias.
Firms should review all datasets and documents used to train models and remove biased, harmful, and erroneous elements. This curation process is key to the principles of safety and accuracy.
Make sure humans are involved
Just because something can be automated doesn’t mean it should be automated. Generative AI tools don’t always understand emotions or business situations, or recognize when they’re wrong or damaging.
A human must be involved to check the accuracy of the output, remove bias, and verify that the model is working as intended. More broadly, generative AI should be seen as a way to augment human capabilities and empower communities, not replace or replace humans.
Businesses have a key role to play in adopting generative AI responsibly and integrating these tools in ways that enhance rather than degrade the work experience of their employees and customers. This goes back to ensuring that AI is used responsibly to maintain accuracy, safety, integrity, empowerment and sustainability, reduce risk and eliminate biased results. . And that effort must go beyond the company’s immediate interests to include broader social responsibility and ethical AI practices.
test, test, test
Generative AI cannot operate on a set-and-forget basis. Tools require continuous monitoring. Companies can start by looking for ways to automate the review process by collecting metadata on AI systems and developing standard mitigations for specific risks.
Ultimately, humans will also need to be involved in checking the output for correctness, bias, and hallucinations. Companies can consider investing in ethical AI training for frontline engineers and managers so they are prepared to evaluate AI tools. If you have limited resources, you can prioritize testing the models that are most likely to cause harm.
get a reaction
Listening to your employees, trusted advisors, and affected communities is key to identifying and correcting risks. Companies can create a variety of channels for employees to report concerns, including anonymous hotlines, mailing lists, dedicated girlfriend Slack and social media channels, and focus groups. Creating incentives for employees to report problems can also be effective.
Some organizations form ethics advisory committees made up of internal employees, external experts, or both to consider AI development. Finally, having open lines of communication with community stakeholders is key to avoiding unintended consequences.
• • •
As generative AI becomes mainstream, businesses have a responsibility to use this technology ethically and mitigate potential harm. By adhering to guidelines and having proactive guardrails, businesses can ensure that the tools they deploy are accurate, secure, reliable, and help humanity thrive.
Because generative AI is evolving rapidly, the specific steps companies need to take will evolve over time. But sticking to a solid ethical framework will help your organization navigate this period of rapid change.
