Tag Archives: ChatGPT

Privacy: Exposing AI Chat Plugin Access to User’s Conversation History.

There are many benefits for allowing third-party plugin access to a user’s Chat history. For example, an OpenAPI, ChatGPT Plugin could periodically troll through a user’s chat history and proactively follow up on a conversation thread that appears to still be open-ended. Or, periodically, the Chat Plugin could aggregate the chat history into subjects by “smart tagging” conversations and then ask the user if they want to talk about the Manchester United, football game last night. Note, in the case of OpenAPI ChatGPT, it has “Limited knowledge of world and events after 2021.” Also, note presently OpenAPI ChatGPT API or the ChatGPT plugin has not exposed the user’s chat history.

3rd Party, Security Permissions for OpenAI, ChatGPT API, and Plugins

Just like authenticating 3rd party apps with your Google credentials, allowing the app to access Google user’s data, this level of authentication should be presented to the user, i.e., “Would you like to allow XYZ ChatGPT Plugin access to your Chat History?” I’m sure there are many other security questions that could be presented to the user BEFORE they authenticate the ChatGPT plugin, such as access to personal data. For example, if the AI Chat application has access to the user’s Google Calendar and “recognizes” the user is taking a business trip next week, the Chat app can proactively ping the user a reminder to pack for warm weather, in contrast to the user’s local weather.

Grass Roots, Industry Standards Body: Defining All Aspects of AI Chat Implementations

We don’t need another big tech mogul marching up to Washington to try and scare a committee of lawmakers into the benefits of defining and enforcing legal standardization, whatever that might be for some and not for others. One of the items that was suggested is capping the sizes of AI models with oversight for exceptions. This could cripple the AI Chat evolution.

Just like we’ve had an industry standards body on the OAuth definition for implementation, another cross-industry standards body can be formed to help define all aspects of an AI Chat Implementation, technology agnostic, to help put aside the proprietary nature.

In terms of industry standards for artificial intelligence, Chat standards, permissions for the chat app, and 3rd party plugins should be high on the list of items to invoke standards.

Extensions to AI Chat – Tools in Their Hands

Far more important than the size of the AI Chat Model may be the tools or integrations to the AI Chat that should be regulated/reviewed for implementation. The knowledge base of the Chat Model may be far less impactful than what you can do with that knowledge. Just like we see in many software products, they have an ecosystem of plugins that can be integrated into the main software product, such as within JIRA or Azure DevOps marketplaces. With relatively simple implementation, some plugins may be restricted for implementation. Many AI Chat applications’ extensibility requires manual coding to integrate APIs/Tools; however, assigned API keys can solve the same issue to limit the distribution of some AI Chat tools.

AI Chat “Plugins/Extensions” can vary from access to repositories and tools like SalesForce, DropBox, and many, many more. That’s on the private sector side. On the government sector side, AI Chat plugins can range, some of which may require classified access, but all stem from a marketplace of extensibility for the AI Chatbots. That’s the real power of these chatbots. It’s not necessarily the knowledge of cheating on a university term paper. Educators are already adapting to OpenAPI, ChatGPT. A recent article in the MIT Technology Review, explains how teachers who think generative AI could actually make learning better.

Grassroots, Industry Standards Bodies should be driving the technology standards, and not lawmakers, at least until these standards bodies could expose all facets of AI Chat. Standards may also spawn from other areas of AI such as image/object recognition, and not all items brought about during the discovery phase should necessarily be restrictive. Some standards may positively grow the capabilities of AI solutions.

Chat Reactive versus Proactive Dialogs

We are still predominantly in a phase of reactive chat, answering our questions regarding the infinite. Proactive dialogs will help us by interjecting at the right moments and assist us in our time of need, whether we recognize it or not. I believe this is the scary bit for many folks who are engaging in this technology. Mix proactive dialog capabilities with Chat Plugins/Extensions with N capabilities/tools, creating a recipe for challenges that can be put beyond our control.