ACM Interactions

In the AI universe, we find ourselves balancing on the edge of a tightrope between groundbreaking promise and perilous pitfalls, especially with those LLM-based platforms we broadly label as “AI platform.” The allure is magnetic, but the risks? They’re impossible to ignore.

AI Platform’s Information Interaction and Service Features: An Unsustainable Model
Central to every AI platform is its adeptness in accumulating, deciphering, and deploying colossal data streams. Such data, often intimate and sensitive, is either freely given by users or, at times, unwittingly surrendered. The dynamic appears simple: You nourish the AI with data, and in reciprocation, it bestows upon you tailored suggestions, insights, or solutions. It’s like having a conversation with a very knowledgeable friend who remembers everything you’ve ever told them. But here’s the catch: Unlike your friend, AI platforms store this information, often indefinitely, and use it to refine their algorithms, sell to advertisers, or even share with third parties.

The current mechanisms of information exchange between AI platforms and users can be likened to a two-way street with no traffic lights. On one flank, users persistently pour their data, aspiring for superior amenities and encounters. Conversely, AI platforms, with their insatiable quest for excellence, feast upon this data, often devoid of explicit confines or oversight. Such unbridled data interchange has culminated in palpable apprehensions, predominantly surrounding user privacy and potential data malfeasance [1].

Following this unchecked two-way data street, the unsustainable model now forces a dicey trade-off between “personalized experiences” and “personal data privacy.” This has led to a staggering concentration of user data on major AI platforms, at levels and depths previously unimaginable. What’s more alarming? This data pile just keeps growing with time. And let’s not kid ourselves: These AI platforms, hoarding mountains of critical user data, are far from impenetrable fortresses. A single breach could spell disaster.

One of the most recent and notable incidents involves ChatGPT. During its early deployment, there was an inadvertent leak of sensitive commercial information. Specifically, Samsung employees reportedly leaked sensitive confidential company information to OpenAI’s ChatGPT on multiple occasions [2]. This incident not only caused a stir in the tech community but also ignited a broader debate about the safety and reliability of AI platforms. The inadvertent leak raised concerns about the potential misuse of AI in business espionage, the risk of exposing proprietary business strategies, and the potential financial implications for companies whose sensitive data might be inadvertently shared.

Whether we like it or not, we need to see and face this fact. The current rapidly growing information interaction model based on traditional user data storage methods is unsustainable. It’s a ticking time bomb, waiting for the right moment to explode. And unless we address these issues head-on, we are bound for a digital disaster.

User Experience Design Based on This Mechanism: Subsequent Problems and Challenges
Beyond the lurking privacy threats and the looming digital apocalypse, this unsustainable info-exchange model is already a thorn in the user experience side. Let’s dive deeper into these annoyances to grasp the gravity of the situation.

The cross-platform invocation dilemma. Major AI platforms operate in silos, creating a fragmented ecosystem where user data lacks interoperability. With the advent of new models and platforms, this fragmentation is only intensifying [3]. Imagine having to introduce yourself every time you meet someone, even if you’ve met them before. That’s the predicament users find themselves in. Every time they switch to a new AI platform, they’re forced to retrain the system with their personal data to receive customized results. This not only is tedious but also amplifies the risk of data breaches. It’s like giving out your home address to every stranger you meet, hoping they won’t misuse it.

Inefficiencies in historical interaction records. The current AI models have a flawed approach to storing and managing historical interaction records [4]. Take ChatGPT, for instance. Even within the platform, one session’s history can’t give a nod to another’s. It’s like they’re strangers at a party. Users struggle to retrieve past interactions, making the entire process of data retrieval cumbersome and inefficient. This inefficiency not only frustrates users but also diminishes the value proposition of these platforms.

Token overload in single channels. Information overload is a real concern. When a single channel is bombarded with excessive information, the AI platform’s performance takes a hit [5]. It’s like trying to listen to multiple radio stations at once; the result is just noise. The current model’s technical limitations become evident as it struggles to scale with increased user interaction, leading to slower response times and a degraded user experience.

A Call for Change
As we draw the curtains on our discussion, it’s evident that the current AI ecosystem, while revolutionary, is far from perfect. The model’s unsustainability is not just a theoretical concern but rather a tangible reality that users grapple with daily.

The complexity of data misuse is a significant concern. Both active and passive data misuse are like icebergs—what’s visible is just the tip, and the real danger lurks beneath the surface. These misuses are not only concealed but also highly unpredictable. It’s akin to navigating a minefield blindfolded; one never knows when or where the next explosion will occur.

Relying solely on corporate responsibility and legal regulations is akin to putting a band-aid on a gunshot wound. While these measures might offer temporary relief, they don’t address the root cause of the problem. The need of the hour is a fundamental change. We must advocate for a deeper, more profound, root-level redesign of the information interaction mechanisms between users and AI platforms. It’s not just about patching up the existing system but envisioning a new one that prioritizes user experience, privacy, and security.

The AI ecosystem is at a crossroads. We can either continue down the current path, ignoring the glaring issues, or we can take the bold step of overhauling the system. The choice is clear: For a more sustainable, ethical, and user-friendly AI future, change is not just necessary; it’s imperative.

Endnotes
1. Harari, Y.N. 21 Lessons for the 21st Century. Spiegel & Grau, 2018.
2. Greenberg, A. Oops: Samsung employees leaked confidential data to ChatGPT. Gizmodo. Apr. 6, 2023; https://gizmodo.com/chatgpt-ai...
3. Forbes Tech Council. AI and large language models: The future of healthcare data interoperability. Forbes. Jun. 20, 2023; https://www.forbes.com/sites/f...
4. Broussard, M. The challenges of AI preservation. The American Historical Review 128, 3 (Sep. 2023), 1378–1381; https://doi.org/10.1093/ahr/rh...
5. Vontobel, M. AI could repair the damage done by data overload. VentureBeat. Jan. 4, 2022; https://venturebeat.com/datade...

Shuhan Sheng