Parameters 241209

In building the Global Changelog we have to be very intentional about the seemingly “simplest” things. The primary question we need to answer is, “What Changed?” But to program our technical stack of automation, APIs, and AI/ML we need to tackle the simplest question: “How do we know when something has changed?”

For example: a President elect appoints someone to their cabinet. What has changed? Well, there’s a few posts on the Internet. There’s maybe a document sitting somewhere in Washington DC. But there are three semantic reasons why this doesn’t qualify as a change.

1. Announcing an appointment doesn’t (necessarily) make it so.
   In many cases, appointments need to be confirmed, people need to be sworn in, and (as we’ve seen) not everyone appointed ends up in the role.
2. You can’t appoint someone until you’re in control. So while there’s a lot going on right now (December, 2024) with cabinet and other positions, nothing can go into effect until the new administration is sworn in (in 2025).
3. People in power use disinformation and anchoring. Anchoring is widely used practice in negotiation. Anchoring occurs when one party proposes a decision that isn’t necessarily what they want, but serves to “center” the negotiation on a certain point or criteria. Sometimes you don’t want your “opponents” to know what you’re going to do. In this case, an administration could appoint someone they never intend to hold the position.
4. Journalistic bias can play a role in portraying something as having already happened or as a sure thing. We’ll defer distinguishing between “journalist” and “publication” biases until another time.[more later] But it’s obvious to say that four articles on a given topic may portray something as being in any stage from “idea”, to “in the making”, to “imminent”, to “already done”.

What does this mean for our algorithms and parameters? (Get thee to the comments, we want to know!) Evaluating all of the above criteria is something that we (okay, “most of us”) do innately and effectively in our subconscious. But an automated system needs to be able to adapt and account for these “nuances of state”.

This means that pur algorithm needs to disqualify (and ignore) a great deal of present-day news discussing things that MIGHT change. Even before dis/qualifying, our system needs to be able to figure out the reliable criteria for what qualifies a given subject matter as “change”.

Conclusions

New Parameters

Parameters need to be configured/updated as follows:

1. Diffusion of Journalistic Bias (1)
   Articles need to be blocked into topical, subject-driven categories.
2. Diffusion of Journalistic Bias (2)
   Multiple different sources need to be evaluated/cataloged for different potential biases.
3. Diffusion of Journalistic Bias (3)
   Changes must be federated across multiple articles with conflicting/correcting biases.
4. Is It Possible (1)
   Changes must be cross-checked with probability of the change happening. (Eg, “27 articles with ‘subject === sky’ and ‘change === turned green’, needs to be cross-checked with something like ‘has the sky ever turned green before?’ and then with…. and then with ‘what is the probability the sky has changed to green?’ Running an analysis like this will prevent Global Changelog from saying something like ‘241104 – Sky turned green’ when 27 news articles commented on the aurora borealis in Oregon that same day.”)
5. Is it Possible (2)
   A similar exercise needs to be followed for permissibility. The question of, “can this be done?” needs to be answered.
6. Did it Happen Yet (1)
   Finally, we need to determine when (or if) something has happened yet. So the algorithm needs to have a bias towards the past tense while also discerning whether the past tense is regarding the subject, the change, or other backstory conclusions that could be included in a given article.

Observations/Conjecture

These parameters seem awfully “stuffy” and “unsexy” in comparison to today’s news. Essentially the algorithm will bang on the “can it happen?” and “did it happen yet” questions more times than your moon landing conspiracy uncle on Thanksgiving day. It is unlikely (and undesirable) that these parameters will clear much of the daily news. As such, people will not get the same kind of “excitement” out of viewing Global Chagelog’s reports. But what they will get is reliability and consistency in the long run.

-AM