AI Expert: More Must Be Done to Protect Data Privacy in the AI Age

LinkedIn, the professional networking giant, was recently caught collecting user data to train its generative AI. The controversy was exacerbated by the fact that LinkedIn began this data collection without prior explicit consent from its users. Instead, all users were automatically opted in, meaning their data was being used unless they actively chose not to share it.

In response to the backlash, the company’s general counsel released a blog and an FAQ outlining upcoming changes to the user agreement and privacy policy, effective November 20^th, meant to better explain how user data is collected. However, neither the blog nor the FAQ explain the full extent of what this user data will be used for.

The uncertainty has prompted renewed scrutiny around how much control users truly have over their data and whether companies like LinkedIn should be more transparent about their data usage policy. Should the industry or the government enforce a standard of transparency, like how the food industry is forced to have nutritional labels?

What is LinkedIn really doing with information they are collecting?  The Large Language Models (LLMs) already built utilize a much larger content set than LinkedIn’s data could ever provide, so why is Microsoft going to such lengths to covertly collect it?

The reason is that building a large language model is not the only Generative AI solution that can be built with large amounts of data.  LinkedIn appears to be training a new type of model, the Large Action Model (LAM). Unlike traditional language models that predict the next word or phrase, large action models aim to predict users’ next actions based on their past activities.

LinkedIn doesn’t just have data on what users have written, it also has an extensive dataset on user actions. Analyzing a user’s connections, past jobs, articles read, posts liked, and more puts LinkedIn in a prime position to develop a model that can predict what members will do next in their professional journey.

Imagine the potential: LinkedIn could predict who is hiring, who is looking for a job, or who is seeking specific services, all based on user activity. This capability could revolutionize the job market and professional networking giving LinkedIn a powerful predictive model that many recruiting and business service organizations would pay significant fees to access.

It also raises important ethical questions about data privacy and user consent. Make no mistake, LinkedIn is not alone in this endeavor. Many organizations are exploring similar technologies, using data from facial recognition and wearable devices to train their AI action models. As these technologies become more prevalent, the need for robust privacy protections and transparent data usage policies will only grow.

As AI technology becomes more widespread, the challenge lies in balancing innovation with ethical data use. Platforms like LinkedIn need to be required to ensure that users have complete control over their data, a requirement that LinkedIn, for the most part, does quite well. What needs to be added to that mandate, however, is that users should be proactively and fully informed about how their data is being used. The automatic opt-in approach may benefit AI development, but it leaves users in the dark and creates a sense of lost control over their personal information. To build trust, companies must prioritize transparency and user control, offering clear and accessible options for managing data preferences.

One proposed solution that I believe has potential is a “nutrition label” approach to transparency. While food labels tell you what you are putting in your body, companies that collect data should explicitly state what data they’re taking and what they’re using it for.

Stock analysts on networks like CNBC must disclose certain information about investments. Companies using AI should also be mandated to disclose their data usage practices in a visible and easy to understand format. This could include information on whether they are collecting user data, if that data is being used in AI training models, and whether any recommendations users receive from the software are generated by AI. Such transparency would better equip users to make informed decisions on how they want their data used.

In the case of LinkedIn, existing data privacy regulations in other countries are already exerting a chilling effect on the company’s covert AI training. LinkedIn’s FAQ is explicit in stating that their AI model is not trained on users who located in the EU, EEA, UK, Switzerland, Hong Kong, or China – countries with strong data privacy laws. In the US, the responsibility of ensuring AI transparency and ethical data use lies with both companies and individuals. Without state or federal regulations, users will have to demand that companies like LinkedIn to strive for greater transparency, while taking an active role in managing their data and staying informed about how it is being used. Only through a collaborative effort can a balance be achieved between innovation and privacy, ensuring that AI technologies benefit us all without compromising our personal information.