A Beyond Intractability GPT?

Newsletter #384 — September 15, 2025
Heidi Burgess and Guy Burgess
We have been contemplating the possibility that the civic renewal, democracy building, conflict, and peacebuilding fields might all be able to benefit from the creative application of AI and, especially, Large Language Models (LLMs) such as OpenAI's rapidly advancing series of GPTs. In thinking about potential positive uses of this technology, we should be clear that we are, in no way, discounting the very serious, conflict-related concerns that also surround AI. Still, our feeling is that, if we have to live with AI's downsides, we might as well take advantage of its upsides.
As a first, tentative, step toward doing this, we have been exploring ways that we might be able to use an especially attractive feature of OpenAI's ChatGPT — its ability to let users create custom GPT's that focus on specific issues and draw from specific bodies of information. We are hoping that this technology might allow us to build an information system that corrects the many ways in which modern information technologies (the Internet and social media particularly) distort public perceptions of conflict-related issues. These distortions drive the destructive hyper-polarization spiral that has been the focus of this newsletter. It seems that a well-designed GPT that was "primed" to look for areas of consensus and constructive conflict framing might help overcome these distortions.
In contemplating such a system, our attention has focused on four of what we see as ChatGPT's most impressive and useful "superpowers:" whole-of-society "listening;" very large-scale, topic-focused reporting; clear, multi-format communication / translation; and cost (i.e. virtually free). In addition, we have focused on ways in which a GPT-based system might also help us overcome a cluster of limitations associated with traditional, human-based forms of expertise (explained below).
This is an outside-the-box essay that will explore ideas that will clearly require a big jump from all of us. The ideas presented are, at best, tentative, and will require considerable refinement before they can produce something truly useful. There may also be downsides to this effort that will need to be identified and mitigated. We expect this to be a controversial post and we welcome your thoughts and ideas.
Whole-of-Society Listening
One of the most consistent themes expressed by those who have been involved in the Beyond Intractability project is the importance of listening, really deeply and actively, and, especially, listening to groups of people that you don't normally listen to and/or are very different from you. In this context, it is hard not to be impressed by the ability of GPTs to genuinely "read" (or "listen" to) pretty much everything that a society has to say (or at least the portion of society that goes to the trouble of articulating and sharing its thoughts on the Internet). This is obviously the kind of listening that goes far, far beyond anything that could possibly be done using conventional means. So it has the ability to show us what many more people are thinking than we would otherwise ever be able to know or understand. This, then, can also help us understand how others are likely (though certainly not certain) to act. While AI's listening skills obviously have their shortcomings, it is clear that they have the potential to provide us with useful information that otherwise would be unattainable.
Very Large-Scale, Topic-Focused Reporting
Extraordinary, whole-of-society listening is, by itself, of little value — all it does is generate incomprehensible levels of information overload. Nobody has anything remotely close to the cognitive bandwidth needed to seriously consider such a flood of information — we simply are not (and can never be) smart enough or speedy enough communicators. This is where AI's second superpower comes into play — its ability to identify, within the staggering cacophony of human voices, particular threads of thought. One thread we would like to ask it to listen for and compile would be ideas that take a more constructive and helpful approach to conflict — ones that show areas of consensus and ways of understanding and acting that draw people together and diminish fear and anger, rather than driving us further apart. We are hoping it can find ideas drawn from outside mainstream thinking and reporting (which tends to be dominated by the antagonisms of today's hyper-polarized, us-vs-them zeitgeist).
This is where, if we are clever, we might be able to figure out how to harness GPT technology in ways that better illuminate categories of under-reported and under-appreciated news and information. Here we are particularly interested in information about what we call the "massively parallel movement" to rebuild democracy, reinvigorate the civic culture, and more constructively handle the many intractable conflicts that have been the focus of BI. More specifically, we think that this system could illuminate, at unprecedented levels of detail,
- The different strategies that have been developed for dealing with the staggering scale and complexity of society-wide conflict;
- The many ways in which bad-faith actors are trying to subvert civic culture and democratic systems and, more importantly, strategies that are being developed and implemented to combat such actors;
- The many interpersonal conflict handling skills that have been developed and are being applied to address many facets of the core and overlay conflict problems that we wrote about in Newsletters 375 and 376 (plus strategies for scaling up these techniques by using mass communication strategies);
- The many ways in which people are learning to live together in a spirit of mutual respect and constructive debate despite deep differences on a broad range of difficult core issues; and, finally,
- The many specialized things that people are doing in each of the massively parallel democracy movement's 50+ roles.
These are all things that conventional information systems have only been able to do in rudimentary and superficial ways and it is something we have always aspired to do but did not have the staffing to even come close to doing it. This is partly due to the enormous costs associated with employing sufficient numbers of researchers and correspondents and also due to economic pressures that force information providers to enlarge and keep audiences by telling them only what they want to hear. By contrast, a properly structured GPT system might be able to better identify and highlight voices of thoughtful introspection that consider, criticize, and suggest changes to things that their side has done and is doing — things that are major contributors to today's hyper-polarization.
Especially helpful would be the ability of such a system to identify and highlight the voices of those who are honestly trying to think through the complexities associated with the many difficult issues that we face. It could evaluate the pros and cons of alternative courses of action and could combine, in mutually beneficial ways, prescriptions from both the right and left. Such materials would provide an extremely important counterpoint to the much more commonly expressed view that all problems are simple — all we need to do is decisively defeat the other side and repeal any and all programs and policies that they might have favored.
Clear, Multi-format Communication
This brings us to the third GPT superpower that we would like to figure out how to use effectively. This is its ability to take ideas that are often presented to limited audiences in poorly written, jargon-rich, and hard-to-understand formats and rewrite those ideas in formats that are much clearer, easier to read (or listen to), and, therefore, be accessible to wider audiences. Ideally, we would use this capability to offer users a variety of responses to queries including: short, easy-to-read summaries of key ideas; longer, detailed, and well documented reports (which could be written by AI, or things it found that were already written by others), something in between those two (modeled after a good opinion article or blog post) or even a podcast.
Low (Virtually Free) Cost
The last big GPT superpower is its ability to do all of these things at negligible cost as compared with the cost of conventional ways of doing the same things (something that, in the current environment, is utterly unaffordable). It is the inexpensive nature of this technology that makes it possible for us to contemplate producing reports on the bulleted items listed above, something that would be impossible for us to do without AI assistance. And, as we slowly learn how to really use this technology effectively, it is quite likely that other, currently unimagined, opportunities will present themselves.
Overcoming Some of the Disadvantages Associated with Human-based Expertise
In addition to capabilities associated with the above GPT superpowers, it is also likely that the technology will be able to help us overcome a cluster of problems that commonly afflict more human-based sources of expertise. Many of these problems arise from the incentive structure that surrounds almost all organizations devoted to creating and disseminating expertise. In order to justify and maintain their funding, these organizations are under enormous pressure to demonstrate that their work will make a critical difference. This results in a tendency to under-emphasize established, collective knowledge and over-emphasize and oversell new ideas (which are often old ideas repackaged with new jargon).
The problem is that this tends to focus our attention on the continuing and almost certainly fruitless search for some new grand solution. The reality, however, is that effectively addressing our many conflict problems is going to require a massively parallel effort to simultaneously develop and apply a few new ideas, but mostly utilize the vast array of existing bright ideas that humanity has developed over the years. We need to spend less effort "reinventing the wheel" and more effort applying what we know. Put another way, at least with respect to the conflict problems that we focus on, we are facing a knowledge-utilization problem much more than we are facing a knowledge-availability problem. If we could just access and apply more of what we collectively know to a wider range of settings, we would be vastly better off. This is where a GPT-based system focused on providing better access to this collective knowledge could be especially useful.
There is, of course, one big and obvious downside to focusing too much on established, rather than new, knowledge. This is the very real possibility that, by reducing the visibility of new insights and replacing them with summaries of old insights, AI may choke off funding for the new analyses that we need to adapt our knowledge to ever-changing conditions. It is critically important that we get this balance right.
Is ChatGPT up to the Task?
As we explore this way of expanding BI, it is important that we make continuing and honest assessments of the quality of the information that our GPT system is able to provide. It may well be that the current generation of GPT systems will not be up to the task and we will have to wait for future improvements in the technology. It is also possible that those improvements will never be forthcoming — because our expectations are unrealistic or unaffordable. Still, based on initial tests, we are optimistic. Given the deep flaws that exist in current information systems, it ought to be relatively easy to do better.
We are also just asking the system to provide us with an overview of what it sees as happening in each area. We are not asking it to draw up detailed plans for specific projects. That would require a much more reliable system — a system that we suspect will be a long time in coming. What we are imagining is, in essence, a team of research assistants that would systematically go looking for stories about what people are doing in each of the areas listed above. In the current funding environment, the GPT-based approach is affordable, while the research assistant approach is almost certainly beyond our reach. In any case, we will emphasize that anybody interested in one of the topics illuminated by the system should take the time needed to really understand the problem and potential solutions before proceeding.
The Gap between Listening and Reporting and Successful Facilitation, Mediation, Negotiation
While we think that there is a good chance that GPT's listening and reporting skills will be able to deliver important, near-term benefits, it is important to distinguish these efforts from much more complex and nuanced tasks such as facilitation, negotiation, and mediation. These all require delicate interpersonal interaction skills that will be much harder for any AI-based system to replicate. (The success of partially automated online dispute resolution systems does, however, suggest that this kind of application may be within the reach of more advanced systems, at least for relatively limited, low-stakes disputes.) That said, our immediate focus is on listening and reporting systems.
Next Steps
We have now moved beyond our initial consideration of theoretical possibilities and have actually constructed an initial prototype of such a system. This system takes the contents of Beyond Intractability and, especially, our key framing documents as a starting point, and then asks the GPT to add information generated by people who, from related perspectives, have been pursuing similar efforts to imagine and build a democracy and civic culture that produces a society in which most everyone would like to live. Hopefully, this will show us that our massively parallel effort is much more massive than we thought. We are not making the BI GPT public yet — we want to tweak it and vet it first to make sure it really is helpful and not misleading or harmful. If the results are positive, we will begin sharing reports from the Beyond Intractability GPT for your comments. And if that seems to work, we will start exploring options for making the system available for public use.
Please Contribute Your Ideas To This Discussion!
In order to prevent bots, spammers, and other malicious content, we are asking contributors to send their contributions to us directly. If your idea is short, with simple formatting, you can put it directly in the contact box. However, the contact form does not allow attachments. So if you are contributing a longer article, with formatting beyond simple paragraphs, just send us a note using the contact box, and we'll respond via an email to which you can reply with your attachment. This is a bit of a hassle, we know, but it has kept our site (and our inbox) clean. And if you are wondering, we do publish essays that disagree with or are critical of us. We want a robust exchange of views.
About the MBI Newsletters
Two or three times a week, Guy and Heidi Burgess, the BI Directors, share some of our thoughts on political hyper-polarization and related topics. We also share essays from our colleagues and other contributors, and every week or so, we devote one newsletter to annotated links to outside readings that we found particularly useful relating to U.S. hyper-polarization, threats to peace (and actual violence) in other countries, and related topics of interest. Each Newsletter is posted on BI, and sent out by email through Substack to subscribers. You can sign up to receive your copy here and find the latest newsletter here or on our BI Newsletter page, which also provides access to all the past newsletters, going back to 2017.
NOTE! If you signed up for this Newsletter and don't see it in your inbox, it might be going to one of your other emails folder (such as promotions, social, or spam). Check there or search for beyondintractability@substack.com and if you still can't find it, first go to our Substack help page, and if that doesn't help, please contact us.
If you like what you read here, please ....