This voice is auto-generated. Please let us know if you have any feedback.
Editor’s Note: The following is a guest post by Caroline O’Brien, Afiniti’s Chief Data Officer and Product Officer, and Kristin Johnston, Afiniti’s General Counsel for AI, Privacy and Security.
Generative AI appears to promise faster innovation, greater efficiency, and greater employee productivity.
Amid macroeconomic concerns in the US and Europe, tools like ChatGPT, Google’s Bard, Microsoft’s Bing and Salesforce’s Einstein GPT are under pressure to do more with less. an attractive way forward for
As the responsible AI leader for a global contact center AI company, we recognize the business benefits of using generative AI for rapid process improvement and content creation. But we believe it can also do harm.
Here are four things businesses can do to use generative AI responsibly.
1. Develop frameworks and policies for responsible AI governance
IBM data shows that more than one-third of global companies report using AI within their organizations, but most companies have implemented processes to ensure the trustworthiness and accountability of AI. do not have.
When responsibly integrating generative AI into your business, we recommend that you start by creating a governance framework to ensure that you are using AI ethically and legally.
The Responsible AI Governance Framework outlines the functions, processes, and groups involved in developing and managing an effective and responsible AI program. This framework requires cross-functional teams to ensure key stakeholders such as data governance, privacy, law, and information security are at the table.
This governance effort requires establishing and overseeing a company responsible AI policy.
The Responsible AI Policy is a formal document outlining the company’s ethical principles that guide the development, procurement, and deployment of AI services.
In addition to applicable regulatory requirements, policies must adhere to at least three core principles: fairness, transparency, and explainability.
- Fairness means proactively identifying areas where AI models may exhibit potential bias or discrimination. If found, it should be fixed.
- Transparency means being able to see enough data and processes to show how the AI model makes decisions.
- Explainability means that AI decisions should not only be visible, but also comprehensible and supported by evidence.
Transparency and explainability do not necessarily require disclosure of the provider’s intellectual property to the world. These principles are about providing stakeholders with the right amount of information to understand the inputs and outputs of AI models, including their impact. This is important in building trust.
2. Understand both inputs and outputs
Responsible AI principles can impact every stage of a company’s data supply chain, starting with the data that generative AI models are leveraging.
One key risk is that AI-generated content trained on historical data could infringe on the intellectual property of others.
Generative AI is very efficient in automating time-consuming processes such as creating marketing copy and creating images for websites and product packaging, but what inputs are used to generate such content? It is important to understand whether
It is also important to take steps to properly supervise the input and output of such content to avoid potential violations of copyright law.
Using chatbots to respond to customer inquiries is also a function that requires responsible input/output control. Because generative AI creates new content, the output of that model can have accuracy and veracity issues.
Each time a new model is put into production, the data used should be documented, categorized, and assigned a risk category. It should also be continuously monitored for discriminatory output.
3. Measure output
Tracking the data used to train a generative AI model is not enough. You should also periodically evaluate your output to ensure that it works as intended and that potential risks such as bias are controlled.
One way to address this is to build measurability into your product or service from the beginning. Be sure to develop well-defined metrics to understand biases and controls that allow you to identify when a particular AI algorithm may be creating bias.
For example, Afiniti tracks how its proprietary AI algorithms are serving customers by turning the technology on and off intermittently.
This benchmarking feature not only allows us to demonstrate how our AI is performing for our users. Using a randomized control group can also detect and mitigate if, for example, the AI exhibits racial or gender bias when it is powered on.
4. Stay up-to-date with human information
The last step may sound obvious, but it’s probably the most important. Humans need to be involved in overseeing generative AI. Develop a comprehensive review process for AI-generated output and empower employees to speak up if something isn’t right.
Generative AI systems are still new, but it has already been shown that the output is not always perfectly accurate or unbiased. The human eye is essential in ensuring that technology is implemented responsibly.
If humans pay close attention to data input and a critical eye to output, the technology we use is more likely to work with the principles rather than destroy them. .
