Last Friday, I participated in a Legal Webinar covering Artificial Intelligence from a ´not so common´ angle, namely: protecting your intellectual property (IP) rights, when developing artificial intelligence (AI) solutions. The webinar was sponsored by Lexology, and taught by a partner and an associate at the law firm Bereskin & Parr LLP.
AI is the ability of a computer program or machine to think and learn from experience and become better in predicting outcomes, based on large amounts of data made available and/or collected on an ongoing basis. Because it requires large amounts of data, it becomes inevitable to encounter complex privacy and data protection concerns, such as retention periods, accuracy risks (e.g., it may become outdated or lead to bias) and required consents and the required transparency practices that this entails.
These privacy and data protection risks are more often discussed in different contexts. In the case of AI development, the best recommendation to have in mind is to employ a ‘privacy by design’ approach, which means analyzing privacy considerations at every step of your AI design and development. Other – more common – but still useful practices include:
1) Seeking express consent from data owner whenever possible
2) Perform privacy impact assessments periodically and seek transparency at all times.
Specifically for IP, the webinar speakers emphasized the importance of using IP to protect your AI technology, data and brands, but filing patent applications at an early stage, with sufficient technical details, and update it frequently to account for future developments, improvements and uses.
Other IP-related issues that were covered included:
- Patenting your trade secrets if viable and, in any event, maintaining strict confidentiality protocols to protect them, both internally (employees) and externally (third party vendors).
- Entering into written collaboration agreements with third parties, especially if you are planning to share algorithms and data, making sure to plan ahead contractually for both success and failure (exist strategy) scenarios.
- Cleansing the data to avoid biased and distressed/ inaccurate data (remember, if you get garbage in, you will get garbage out) and get the best out of your AI solution.