A Simple Privacy Concern with AI Data
Author: Niall McCarthy, CEO Aire AI
Published: Data Science Spotlight, 21 Jan 2025
The amount of information about ourselves and our businesses that we’re willing to give to AI, in particular Large Language Machines such as GPT and Claude, is disturbing.
Not so long ago, there was much concern about the giant sucking noise made by Google and other search engines. However, the practical reality is that traditional search engines were often based on getting to know you just through your search history and behavioral analysis. Of course, this concern hasn’t gone away, but the danger has increased by an order of magnitude.
Starting with search itself, it’s now well-known that users of AI voice-enabled search give far more away about themselves than with traditional written search. Their search question is longer and their recorded voice can be immediately analyzed for sentiment.
However, the real threat is the number of industry-specific applications being built directly upon Large Language Models. The potential market for this new generation of SaaS type solutions is immense. We see already how it’s revolutionizing the domain of marketing content generation. It goes even further in the domain of customer support, with AI Agents accessing entire knowledge bases of organizations and being rigorously “trained” with hundreds of examples of business processes to execute.
On the individual/consumer level, we see the mass adoption of voice assistants well beyond the domain of search (e.g., Alexa, Siri). We also see relatively few if any guardrails as to how users query AI. Large Language Models, for example, aren’t designed to say “no” to a child, but rather to generate never-ending output based on what has just been said. Commercially, it’s not in the interests of the software vendor to stop the conversation.
AI designs which incorporate “Human in the Loop” feedback are becoming more prevalent, but this is a double-edged sword. HITL improves the AI output, making it more accurate to the user’s requirement. However, it also fine-tunes what the owner of the AI knows about the user or their organization.
As the CEO of an AI software vendor, our approach to solution design has to take into account the digital sovereignty of our customers. As father to a young child, educating our boy not to trust a connected computer or mobile device, while simultaneously learning its potential has become a top parental challenge.
Leave a Reply
Want to join the discussion?Feel free to contribute!