TruthBear Blog
People, AI and Content in TruthBear
TruthBear's goal is to give people high quality information they can trust.
To be trustworthy, information has to come from sources that we trust. As we discussed in previous posts, not only does that mean that they usually provide correct information, but also that we trust them to be thoughtful and decent in choosing what to focus on in the first place. In this post, I'll outline why I think that it's important for humans to take the leading role, rather than AI, and what that means for the use of AI on TruthBear.
When I started this blog, I made an outline of the topics I wanted to cover in the first ten or so blog posts. I had no idea that serendipitously, the week I chose to talk about the use of AI in TruthBear would be the very week that LLMs solved one of the major outstanding problems in mathematics, proving that the Navier-Stokes equations can break down. Sadly, within hours, it had degenerated into a fight about credit, provenance, and who is accountable for the output of models. Leaving aside the way OpenAI and its employees have comported themselves in this matter, it is fair to say that AI is proving increasingly capable in a range of areas. Often they exceed what most humans can achieve, and increasingly often, perform at an entirely super-human level. It is uncontentious to expect that this trend will continue.
This is well trodden ground, but because in the near future it may be hard to imagine what we were thinking in 2026, I'll take a small diversion to outline the range of outcomes people are thinking about. Yudkowsky, and some AI researchers building the things, think that AIs will likely kill everyone. Many people think AIs will become superhuman, and there will no longer be any need for humans to do anything. This is either terrible, because there will be no jobs, or great, for the same reason. Either way, the path is likely difficult, because jobs in some sectors will surely dry up before others, and our economic systems are not set up to deal well with that. But some, such as Altman, now think that AI leads to job creation. Gary Marcus and, more recently, Yann LeCun think that LLMs will fail to become generally super-human, and we need neurosymbolic approaches instead. But even if they fail to reach the arbitrary and moving milestone of AGI, AIs have become very useful tools - they can be used for understanding, or for the generation of misinformation. So, the range of outcomes is wide, and seems unforeseeable - maybe we get utopia, maybe we get dystopia, maybe we get something else.
What we can be certain of, is that we have a fairly fixed supply of human effort, and we have just created a low-cost substitute for at least some of that effort, and an increasing proportion going forward. A similar thing happened in the 20th century with farming, which went from about 40% of US labor in 1900, to about 2% today, due to increasing automation. That labor is now used for other things, and we can expect the same with AI substituting for human effort.
The question then, is what will people apply their effort and attention to? My guess is that there is pretty unbounded demand for human attention from other humans! We all want people to notice us, to turn up and enjoy our blog posts, for example. People enjoy cooperating together on projects. And it is likely that people will prefer to be cared for by other people.
The other part of the answer is that there may be some tasks that we still want to be done by people, or with human involvement, even if AI could do them better. For example, we may prefer the Supreme Court to be a set of imperfect humans, rather than a quorum of Claude, ChatGPT and the others. We may also want to continue our quest to understand the world better. Partly that's to maintain propriety over our own affairs, but also because it's fun! The fact that there are people who understand Shakespeare better than me, doesn't preclude me from enjoying it. And I think it's the same with politics, macroeconomics, sports, celebrity gossip, particle physics and everything in between.
Some of how I think concretely about using AI for TruthBear is formed from past and current experience.
One previous iteration created a "den" of bears, each responsible for a particular topic. Among other things, they monitored the US Senate and House and UK Parliament, digesting voting records into attributed sources of belief for our leaders. But the effect was terrible - like the worst parts of being on a jury. While it did give you a view what our leaders were up to, the topics were disconnected, and the overriding sense was that several hundred people had already decided the verdict ahead of you. Suffice it to say, that led to a pivot around topic selection - before you get information, you need to decide what is important. There's a wonderful slide from a 1979 IBM training presentation that says a computer can never be held accountable, therefore a computer must never make a management decision. Well, that's true for selecting what is important. Humans need to own that selection, even if they are working in concert with AIs.
I did like the "den" idea. The idea was to allow the community to be aware of what the most prominent people in our society say they believe, even in advance of them joining TruthBear. We may later reintroduce it, but focused on topics the communities on TruthBear select.
My more recent experience was in "dog-fooding" the latest iteration of TruthBear. (This phrase refers to using your own products as a way to improve their quality). For TruthBear, that's hard, because it's fundamentally a multiplayer product. So to simulate that with only a single player, I have Claude operate as another user while we improve interfaces and complete features needed for the private trials coming up soon. It works reasonably, but the end product is intended to be for human users, so Claude will be bowing out soon.
The reason to retire the Claude user is that I am concerned that AI could unintentionally dominate TruthBear. A rule of thumb is that in online communities, 90% of people only consume or read content, that another 9% contribute occasionally, and that 1% of people create the bulk of the content. If TruthBear were populated by many AIs, then we expect people would be drowned out. That would be a loss. If you want to be told what's important or what to believe by an AI, there's already a really good way - just fire up a chat window and ask it. But we are sorely lacking a way to come to a consensus understanding of the world with our friends, families and communities - that's what TruthBear is for.
The other reason to be concerned about AI, with its prodigious ability to generate words, dominating TruthBear, is that AI can be subject to misinformation. So, I hear, can people. But for AIs in particular, they can be subject to the whims of the people or companies that select their training material, and they can be affected by concerted production of Internet content designed to push their views one way or another. A hostile entity could also just fire up thousands of agents to directly swamp TruthBear with their viewpoint. TruthBear deals with this by having communities, where there is a level of mutual trust, for determining what matters, and by additionally having credibility weighting when weighing the beliefs of others.
One thing we are not concerned about is that the voices of AIs will go unheard. Everyone has access to a chatbot. They don't need that from TruthBear. We do expect people will express views that have come from AIs, but the difference is that people are accountable for the views that they are relaying.
I'd like to finish by clearly articulating TruthBear's AI position.
TruthBear is a platform to help people understand the world. We do that by providing high quality information they can trust. TruthBear is solving a human problem: people misleading other people. Its tools for doing that are accountability and clear provenance of information. So it is right that on TruthBear, only humans can express what matters, and what they believe. Certainly, people can and should use AI tools to help determine those things. But there needs to be a space for human expression where people can work together to understand the world.
I continue to believe that it is no longer possible to impose what is true from above. To have a shared reality again, we have to do it from the bottom up, together. I usually say that with respect to traditional media sources - newspapers, TV news networks - but I think it is equally true for AI also.
I love hearing your good ideas. If you think something here is unclear or just plain wrong, you can help make TruthBear better by telling me: ed@truthbear.org.