Right now, it seems as if every tech vendor is innovating with generative AI.
Yet, as the trend accelerates, more experts have started to voice their concerns over the development of AI.
These concerns cover many spaces, and the enterprise is one of them.
There, possible risks include generative-AI-based applications sending fabricated information to customers and creating copyright breaches.
However, perhaps most worryingly, concerns remain over how they may endanger data privacy.
Microsoft recently moved to quell these by announcing plans to release a private mode for ChatGPT: PrivateGPT.
Hosted on dedicated servers, PrivateGPT will run “advanced privacy protection methods” on copies of the AI software to avoid data leaks.
So, problem solved, right? Perhaps, for now. Yet, additional risks may scupper future use cases of generative AI in the enterprise.
The Current Vision for Generative AI In the Enterprise and Its Flaws
The future of generative AI in the enterprise is not only for text generation. The next frontiers will be image and video models.
So far, these models have not digested all of the videos on YouTube or the podcast audio on Spotify. Yet, when they do – which is the direction of travel – enterprises will repurpose and feed them with massive corpora of audio and visual data.
Then, as more of the communication inside of businesses moves from email and messaging to video calls, recorded meetings, and audio notes, the conversation will move beyond companies wanting private versions of ChatGPT to upload their own text data.
Instead, businesses will want their own multimodal assistants running inside of these experiences.
Microsoft has painted that vision with Copilot. Yet, James Poulter, CEO of Vixen Labs, points out a rather large assumption within those plans that may limit its potential.
"[With Copilot] it seemed a given that we’ll all be happy to have this running in the background of our Teams calls," he said during CX Today’s latest BIG News Show. "But I think Microsoft is going to start hitting lots of significant hurdles with their big enterprise customers.
They will not be happy that all of those video calls are going to be auto-transcribed and uploaded, and that’s why significant orchestration around secure environments has got to happen.
For starters, end-to-end encryption of enterprise systems that leverage generative AI must happen, with encryption on either prompts or replies.
Then, there is on-cloud encryption, which ensures that generative AI applications stay away from some of the data stored within SaaS systems.
