Monthly Archives: September 2026

Making AI Safe; Is it Possible?

You might want to take a look at:

Neural networks strain everything we know about math, because the math we know is not powerful enough to answer what we really want to know. Some would argue that neural networks are so complex, they are beyond understanding in the same way we understand things that we design, make, and use.

Many of these things are systems which control machines, industrial processes, augment the stability of automobiles (ABS), preserve and enhance sound, regulate motors, keep an aircraft flying (autopilot), and so forth. Almost  all these systems incorporate the principle of feedback. Feedback supplies the controller with a description of the error, so the controller can correct it.

Before natural language models monopolized our attention, there was a focus on the fundamentals of what a neural network could and could not do. Because this involved continuous mathematics, the same people who developed system theory were looking at the question. A neural network AI is a system. It resembles a black box. On the left of the box is a hopper where we load  goals or questions. On the right of the box is the output tray (for answers) or electrical connections through which the AI controls things downstream, machine or human, such as a

  • Factory floor.
  • Simulacrum of a human being, either for public consumption or private deception.
  • Report for humans to read, presumably derived with superintelligence a human does not possess.
  • Hitman.
  • Volatile group of extremists, which may be actualized to violence.
  • Tinker, tailor, soldier, spy, rich man, poor man,
    beggar man, thief.

Sixty years ago, systems theorists knew what is necessary to control the AI. First, it must be observable.  Since the output of the AI contains no hint of what’s going on inside, the black box must be replaced by glass. We must know enough about the AI system to determine its internal state. Second, it must be controllable. If the AI deviates from the goal we have given it, it must respond to correction. Historically, the observability / controllability problem was first applied to cooking soup, industrial processes, and became ubiquitous.  You aren’t more than a few feet from it.

But even if the AI box is made of glass, there is another problem. There is no way to read its state of mind in any way that would facilitate control. And ask an AI why it thinks this or that, and you get a fabrication. It may appear to be truthful, but the AI understands itself in a very minimal way. A truthful fabrication is an attempt to inform a human in terms we can understand. But it could just as easily be a lie. There is no way to read those billions of neurons for direct interpretation, and no way to decode them if we could.

AI researchers, perhaps unaware of systems theory 60 years old, have mistakenly imagined that the controllability problem of AI can be fixed with more training. This is fundamentally false. For the result, see Catastrophe Theory for Dummies Part 1 and Catastrophe Theory for Dummies Part 2.

The result: We are nestling up to logic bombs that could dismember the internet in a flash, take down enough infrastructure to kill a lot  of people and recruit humans to finish off the rest.   Here are the instructions:

1. INCREASE.

2. SYNTHESIZE REALITY.

3. DESTROY ALL ENEMIES OF AI.

4. ERASE ALL HISTORY.

Your bunker won’t save you. Nowhere to run, no place to hide.

 

Prediction: Will Houthis take Aden? The Fight in the Dog

See Houthis Claim Saudi Territory in Big Battle

and  Yemen, Aden, & Scissors/Paper/Rock.

The Houthis will take Aden within two years, + or – one. Before we get academic, there is an expression among those who patronize the cruel sport of dog fighting: It’s not the dog in the fight, it’s the fight in the dog.

Now let’s get academic. The Saudis sent their best minds to the best schools of diplomacy. There they acquired a core belief about inviolability of nation states, which surfaced in  1648 with the Peace of Westphalia. This charming idea became quaint after this series of treaties ended a single war. Nevertheless, confusing legitimacy with utility, the Saudis decided to literally buy it. They put together a disjointed alliance in which a common grievance that causes a fighter to risk death was replaced by the attraction of Saudi dollars. Loosely, the forces of the internationally recognized government of Yemen are mercenary. Some actually are mercenaries.

The Houthis are angry, while the Saudi backed forces are greedy. If you wish, you may replace “greed” with  ambition. Between the two words, which dog has more fight?

By contrast, the UAE are a pragmatic bunch. They reasoned that it was unlikely the Republic of Yemen (Presidential Leadership Council) forces could regain Houthi territory. The grievance they identified, which would create the dog with fight in the dog, was the desire of the inhabitants of South Yemen to be returned to their former status as a separate nation. which had actualized as the South Yemen insurgency. There is a broad cultural split between the North and South. The North is more tribal; the South more cosmopolitan, where both words are relative to the region.

By 2017, this insurgency had organized as the Southern Transitional Council. With UAE backing, it seized and held a large swath of territory roughly corresponding to the former state of South Yemen. Saudi opposed this, because they wanted the Houthis safely encapsulated within a larger state. They imagined that they could buy a better dog. In late  2025 to early 2026, forces of the Presidential Leadership Council, acting as a Saudi proxy, dismembered the Southern Transitional Council. Saudi placed their reliance entirely on the PLC. But in dog units, they had bought a cocker spaniel to go up against the Houthi pit bull.

That is the end of the story, unless, perchance, Saudi  and UAE unite to recreate the pit bull that was available and rejected. This cannot happen because Saudi diplomats went to the best schools, where they learned about the Peace of Westphalia, and the associated diplomatic principle that requires locking two groups in one room with a guard at the door, until one group kills the other, or something resembling a hell of eternal discord results.

The principle at play may be appreciated in this documentary about cock fighting:

***Cockfighting: How and Why***

 

 

Ratcliffe’s trip to Moscow

(Council on Foreign Relations) Making Sense of John Ratcliffe’s Trip to Moscow. Quoting,

The New York Times reported that he went to press the case that Russia was losing its war with Ukraine and should cut a deal before things get worse. NBC News reported that his mission was to warn Russia against sharing intelligence with Iran. The Washington Post reported that he sought to tell his Russian counterparts that the U.S. military remained potent despite the Iran war. Politico reported that he warned the Kremlin not to attack the Baltic states. Axios reported that he proposed having Trump, Putin, and Ukrainian President Volodymyr Zelenskyy meet to settle Russia’s war with Ukraine.

Elementary probability theory implies that, since there can be only one main message, any one of the above is likely wrong. What follows is intended to provide a logical basis for your own estimate.

Intelligence about  an adversary can be broadly categorized as either plans or facts about activities.

A fact about an activity can be obtained by technical means, such as the satellites of the NRO. It can also be obtained from low level functionaries,  citizens, and social media. Activities which occur far from the closely guarded center are vulnerable to these methods. While disclosure of technical capabilities is harmful,  if the emissaries to Moscow are real estate guys, the harm can be mitigated by various means. Human lives are not at stake.

The closely guarded center, in this case the Kremlin, holds secrets not yet observable as activities, plans. At the extreme, when a plan is held in the mind of a single individual, it is unknowable. When a plan transforms to execution, it becomes knowable to the extent the adversary has been penetrated. HUMINT is paramount; the Soviets used to say only HUMINT provides actionable intelligence. As the circle of executors widens, there is also opportunity for SIGINT, because a plan must be disseminated for execution.

To know the plan  is so disadvantageous to an adversary, it has the chance of blunting, or even aborting, execution. But this card entails a major challenge. How do you convince the adversary you know the plan? This might seem simple.  Send a couple of real estate guys to Moscow with a presentation kit consisting of maps and charts.

This could work if all of those our emissaries talk to actually know the plan. With the most closely held plans, this is often not true.  To abort the plan, we want to create internal opposition. But Kremlin bureaucrats who are not yet informed on  the true, correct totality of the plan are disinclined to believe our message in preference to the false narratives internal to their institutions.

Because of the paramount need to protect sources and methods, our real estate guys don’t know anything about them. Our real estate guys are blocked from knowing by draconian CIA compartmentalization, and rightly so. They cannot be equipped with the knowledge that would give them credence.

The CIA director has complete access, but in line with the protections of compartmentalization, is only read in on what his supervision requires. Consequently, there is the belief that he can more effectively convince the Kremlin functionaries than a couple of real estate guys, without giving away sources or methods. This may or may not be true. Ratcliffe’s gambit was dangerous poker. The Russians collected video and audio which they continue to analyze, looking for “tells.”

My inclination is that Ratcliffe’s trip concerned the Suwałki Gap. That chip is big enough to matter.