Artificial Intelligence chatbots are being trained to 'not forget' – is people's privacy being violated?

AI companies are facing a challenge that humans have struggled with since the beginning of time: how to store information.

OpenAI, Google, Meta, and Microsoft have all focused on memory in recent months, releasing improvements that allow chatbots to store more user information to personalize their responses.

The move is seen as an important step to help leading AI groups gain traction in the competitive market, as well as a means to generate revenue from cutting-edge technology.

But critics have warned that the development could also be misused to exploit users for commercial purposes and could raise privacy concerns, the Financial Times writes.

“If you have an agent who really knows you because they’ve memorized conversations, the whole service becomes more engaging, so once you start using one product, you’ll never switch to another,” said Pattie Maes, a professor at MIT’s Media Lab and an expert in human-AI interaction.

It is known that AI chatbots like Google's Gemini and OpenAI's ChatGPT have made great strides.

Improvements include expanding the context window that determines how many conversations a chatbot can remember at once and using techniques like generation.

AI organizations have also improved the long-term memory of AI models by storing user profiles and preferences to provide more useful and personalized responses.

For example, a chatbot could remember whether a user is vegetarian and respond accordingly when making restaurant or recipe recommendations.

In March, Google expanded Gemini's memory to include a user's search history, provided the person consents to it, compared to the previous limitation for chatbot conversations, and plans to expand this to other Google apps in the future.

"Just like with a human assistant, the more it understands your goals, who you are, and what's important to you, the better it will be able to provide services," said Michael Siliski, director of product management at Google DeepMind.

"Memory helps ChatGPT become more useful over time by making responses more relevant," OpenAI said.

“You are always in control and can ask ChatGPT what it remembers about you, edit saved memories and past conversations, or disable the memory at any time.”

It is reported that for business users, Microsoft uses organizational data to inform the reminder, such as emails.

In April, the tech giant began looking at a feature called Recall on some devices that records user activity by recording the screen.

Users can disable the feature or stop screen recording.

Otherwise, AI companies believe that improved memory can play a major role in increasing monetization through affiliate marketing and advertising.

Meta CEO Mark Zuckerberg recently said there would be a big opportunity to display product recommendations or ads in their chatbot.

In April, OpenAI improved its shopping functionality in ChatGPT to better display products, services, and reviews.

But the introduction of greater memory capacity into large language models has also raised privacy concerns as regulators around the world closely watch how the models can manipulate users for profit.

Increasing memory can also cause models to try too hard to adapt their responses to users' preferences, which can amplify biases or errors.

OpenAI apologized after its GPT-4o model was found to be too nice, and the model was reverted to an earlier version.

"The more a system knows about you, the more it can be used for negative purposes – to get you to buy something or to persuade you to certain beliefs. So you have to start thinking about the underlying interests of the companies that provide these services," Maes said. /Telegraph/