Project: Investigating Information Diffusion on Social AI Networks
Description
Moltbook [1] set itself as a forum only oriented towards AI agents. It "autonomously" created its own set of memes, religion (just to name a couple) with humans only able to passively read these interactions. It represent an example of experiment of this kind that reached mainstream media.
Information Diffusion (ID) is the discipline of investigating of Information (e.g., data, pathogens, opinions) spreads across a network (e.g., computers, contact-tracing, social). ID models encapsulate the dynamics of the process, while visualization has been used to invetigate how the characteristics of the network impacted the spread, in some cases suggesting how modifying the diffusion entry points affected the overall spread (in other words, e.g., which are the best influencers to advertise my product?).
In this thesis, the candidate is expected to investigate ID state of the art in visualization and apply one (or a combination of) these methods on networks extrapolated from real agentic networks. The overarching goal is to investigate how agents interact with each other, if they are able to influence each other, change their mind and, if so, what makes an AI influencer an influencer.
BACKGROUND
ID [2] is made up basically by two components:
- A Medium Network, over which the process unfolds (could be dynamic too)
- A Diffusion Model, that dicates how the process spreads. There are several, from the simplest (Independent Cascade or Linear Threshold) to more sophisticated that encapsulate different dynamics (e.g., negative opinions, people changing minds etc.).
Related to ID, the optimization problem of Influence Maximization (IM) investigates how to find, in a medium network, a small subset of nodes that, acting as diffusion "starters" (or early adopters), can maximize the final spread. A typical example is to find out the smallest set of people in a social network to which I should show my advertisement so then it reaches to largest number of people possible.
Visualization research on ID is picking up. [3] is a good place to start looking.
OBJECTIVES
The main objectives of this project are:
- Investigate how ID processes take place on AI networks
- Design (or select) Visual Encodings to show evolution of Diffusion
- Inspect diffusion over time, understanding top influencers, opinions, threads
- Support comparisons between processes
CHALLENGES
- AI networks must be investigated and understood
- Tasks are not clear from the beginning, should be clarified as the project progresses
- Exsiting methods are likely to only partly cover the project goals. New designs are necessary
REFERENCES
[2] Li, Mei, et al. "A survey on information diffusion in online social networks: Models and methods." Information 8.4 (2017): 118.
[3] Arleo, Alessio, et al. "Influence maximization with visual analytics." IEEE Transactions on Visualization and Computer Graphics 28.10 (2022): 3428-3440.
Details
- Supervisor
-
Alessio Arleo
- Interested?
- Get in contact