The Unjournal Vision: Making AI Research Truly Accessible

Informal Discussion December 2025
Captured across Jeju locations
Jiliang Tang Knut Neumann Chris Quintana Janice Gobert + more
Back to Blog

Between formal sessions, some of the most valuable ideas emerge. These informal conversations across Jeju - at cafes, during walks, over meals - crystallized key insights about making AI research accessible and how the Unjournal should actually work.

Beyond the Conference Room

These discussions were captured informally as participants explored Jeju together, thinking out loud about practical implementation challenges that formal sessions don't always address.

The GitHub Problem: Why Researchers Can't Use Our Tools

The Hard Truth

In AI research, every paper requires open-sourcing code and data. But putting code on GitHub doesn't mean education researchers can actually use it. "Even if you put your code on GitHub, many people from the educational domain don't know how to get it."

The Technical Barrier

In the AI field, papers without open-source code aren't trusted. But education researchers often can't use command lines, set up environments, or run analysis scripts - even with detailed instructions.

View Details

The conversation revealed a significant gap between AI research norms and education research capabilities:

  • AI field requires open-source code for credibility
  • Education researchers lack technical skills to run code
  • "GitHub is for developers" - the interface itself is a barrier
  • Even clear instructions require command line knowledge
View original discussion

Jiliang Tang: "In our field, every paper - you need to open source. Your system, your implementation, your sheets. Because of AI, it works too fast, there are so many algorithms. If your paper is not open source, people will not trust you."

Participant: "The problem is, even if you put your code in GitHub, many people from the educational domain - even if you put it there - they don't know how to get it."

Jiliang Tang: "GitHub is for developers. It's a great thing, but yeah... you still need to run a few command lines, set up the environment, and so on. Many people cannot."

The Solution: "Just Make It Like ChatGPT"

The Breakthrough Idea

Instead of GitHub repositories with command-line instructions, create natural language interfaces. "The interface is just natural language. Just say 'hey, I want to get it' and you get it."

Current Approach

Clone repo → Install dependencies → Set up environment → Run command lines → Debug errors

ChatGPT-Style

"I want to analyze student responses using your rubric" → Get results

Natural Language as the Universal Interface

English is "the most popular programming language" now. With natural language interfaces, there's no technical barrier - anyone can access and use research tools.

View Details

The concept of "WIFO coding" - Without Code - emerged from this discussion:

  • Users describe what they want in plain language
  • AI handles all technical implementation
  • No environment setup, no command lines, no debugging
  • Researchers can copy their rubrics and data directly
  • Analysis runs automatically and returns results
View original discussion

Jiliang Tang: "The best way is still just like ChatGPT. You provide the language interface. You just speak. You speak, and you get what you need. In English - because English is the most popular programming language. We call it WIFO coding - you don't even write code, you get it."

Participant: "In this way, there's no technical barrier. So all in English. It doesn't even speak, you get it."

AI-Powered Unjournal: Using What We're Studying

The Irony

Current journals prohibit AI use. But we're creating an Unjournal about AI in education. If we ban AI from our own publication, "that's a little weird, right?"

AI-Powered Recording and Summary

Record discussions, AI transcribes, prompts generate summaries. Participants review for accuracy. No one needs to watch hours of video - they get verified summaries instantly.

View Details

A participant proposed a workflow that got immediate enthusiasm:

  • Create prompts for what to extract from discussions
  • Record working sessions and conversations
  • AI transcribes and generates summaries based on prompts
  • Discussants review and verify accuracy
  • Publish verified summaries with original recordings
View original discussion

Participant: "Maybe creating a set of prompts - we record it, AI will transcribe, and based on the prompts, come up with a good summary of what they discussed, what they agreed."

Another participant: "That's brilliant! Then we don't need to worry about it. You don't need to watch - if you record it, you can watch everything but it's really hard to summarize. And then the discussing person can go through and make sure the AI summary is accurate."

AI "Lenses" for Exploration

As the repository grows, create AI-powered exploration tools. Users can search across all content, find connections between papers, and explore through different conceptual "lenses."

View Details

The vision connects to Knowledge Forum from University of Toronto - but enhanced with modern AI:

  • Unified repository: instruments, articles, and related research in one place
  • AI can synthesize across the whole collection
  • Create epistemic networks between elements
  • Search features that understand conceptual connections
  • "AI lenses" for different ways to explore the content
View original discussion

Chris Quintana: "Over time, if we have this repository that's growing with more stuff, you could almost create AI lenses for the community to use to explore the repository."

Participant: "Usually you have the instrument here, the article you want to read here, and a body of research that uses those things - all in separate places. If this is all in this unified way, then your summary could be about that whole collection."

What the Unjournal Could Be

Video Clips

Not just text - actual demonstrations and discussions

Data Repository

Annotated datasets ready for others to use

Commentary

Like Reddit - ongoing discussion and feedback

Speed

Not waiting a year - "the field changed since we showed up"

A New Type of Dissemination

"It's a site. The Unjournal - we're going to have video clips, repositories, commentary. It's a new type of dissemination that's quick. Because things are happening too fast."

View Details

The vision differentiates from traditional journals in several ways:

  • Multiple content types: video, code, data, text, discussions
  • Continuous rather than issue-based publishing
  • Community commentary built in
  • AI-powered exploration and synthesis
  • Accessible to non-technical audiences
View original discussion

Chris Quintana: "It's a site - the Unjournal. We're going to have video clips, we're going to have repositories, we're going to have commentary. You know how Reddit has commentary? It's a new type of dissemination that's quick. It's not waiting a year to get something out. Because things are happening too fast. The field changed since we showed up here a couple days ago."

The Deeper Challenge: Building Capacity, Not Just Completing Tasks

Knut's Closing Insight

"If you take the AI away, the capacity is lost at the same time. That gives whoever owns the AI a huge lever over those who are using it."

The Global South Challenge

Papers from Africa now have excellent English - likely due to LLM access. But if the AI disappears, so does the capability. AI in education must build human capacity, not replace it.

View Details

Knut Neumann's reflection connected the discussions to a fundamental principle:

  • Sending money and resources hasn't always built capacity
  • AI tools are similar - they can complete tasks without building skills
  • The dependency created gives AI owners enormous power
  • Education's role is to ensure AI builds capacity, not replaces it
  • Future generations need the ability to do tasks, not just access to tools
View original discussion

Knut Neumann: "We're talking specifically about the Global South and the challenges that come with using AI to increase equity. What we've been doing in the past is similar - we've been sending money out and doing all kinds of things because we thought it would help. But what we failed to do is to build capacity in a lot of ways."

"I see a lot of papers submitted to JRST or IJSE now from Africa that have really good English. I think it's because they have access to large language models. Still, the problem is that if you take the AI away, the capacity is lost at the same time. That kind of gives whoever owns the AI a huge lever over those who are using it."

"In education, AI must be used not just to do a task, but to build the capacity in future generations to do that task."

Final Words

"It was rich. It was honestly rich. And I feel really energetic and full of ideas. I sincerely hope we can continue this in this group and push things forward." — Knut Neumann