This update is part of our AI FAQ Series. Learn more at our AI Law Center.
1. Can I use this data to train my AI models?
Whether or not you can use a particular dataset to train an AI model depends on a few different factors including: (i) the licenses or terms that govern the dataset and the restrictions within such licenses/terms; (ii) the privacy notice under which the data was collected; and (iii) the organization’s AI policies. As such, you should consult with your legal team or the team in charge of AI compliance to learn more about the process for determining whether a particular dataset may be used to train an AI model and under what conditions.
2. What should I think about when I am developing my AI product?
When developing an AI product, start by clearly defining the problem it’s meant to solve and the value it will bring to users. Choose the most appropriate AI technologies and methods that match your data and objectives. Design your AI solution with clear metrics for success and integrate it into a user-friendly product. Finally, rigorously evaluate the impact of your AI, using feedback to make iterative improvements to ensure it meets user needs and performs reliably in real-world scenarios.
At each stage, an AI product developer should think about the range of issues likely to be encountered with the product. This could include the initial licensing of any technology and the potential patentability of any novel technology; legal implications related to use of any data relied upon by the technology; and privacy and data security issues, particularly ensuring that data used in AI systems is collected, processed and stored in compliance with applicable privacy laws and regulations. AI systems should be designed to uphold principles of data minimization, consent and individual rights to access and delete personal information. Depending on the type of data involved, consider appropriate protections and consents that would need to be sought, particularly where it may be protected data (e.g., health) or customer data.
These concerns and risks should then be factored into the relevant contracts and other third-party agreements, ensuring that customers or parties using the product are put on appropriate notice and are appropriately responsible for their own data. And when marketing the product, the organization should take care that its representations about the product and data use are fair and not misleading.
3. How do I monetize my AI tool/product?
There are a variety of different direct sales, licensing and data monetization strategies to monetize AI tools and products. Additionally, there may be opportunities for partnerships and integrations with other platforms to expand your market reach. The right technique for your product will depend on many factors, including your industry and how your tool was trained. Consider engaging counsel to learn more about the legal considerations for the different monetization options and to determine which technique may be right for your product.
4. How do I measure AI bias in my models?
Bias can be measured by looking at the disparate impact of model outcomes across different population subsets. A similar method for measuring disparate impact is often used in public health studies. Log odds or odds ratios may be used to measure the amount of bias by measuring the difference in model outcomes between different groups. This is a rapidly evolving area, and companies should consult subject matter experts to determine whether their AI models have been sufficiently vetted for bias issues prior to releasing the models publicly.
[View source.]
