AI Consequences

 

Axios today (I am not able to copy the URL to paste it for you; they made the copy unavailable, a bit irritating and makes me wonder if I want to keep them on my reading list) has a story that soon AI will be able to automatically improve and test itself, this is seen as a danger. I rather feel the horse has left the barn and the chances are that it will be in fact an extreme positive. Still, the simple fact that AI does not have sentience and hence behaves without ethical constraint or responsibility, is a bit of a worry. It may decide on its own, in spite of what we tell it, that our sentience is an illusion worth dispensing with. The saving grace here is the question of why it would care.

I think there are three separate questions tangled together here.

The first is whether AI systems will eventually be able to improve themselves. To some extent they already do, but only indirectly. An AI can write code, analyze experiments, suggest architectural changes, and help design the next generation of systems. Human researchers still decide what to build and what to trust. The concern is a future point where the loop becomes largely autonomous: the system proposes improvements, implements them, tests them, and repeats. That is the scenario many researchers call "recursive self-improvement."

The second question is whether that is necessarily dangerous. I agree with you that many discussions overlook the enormous upside. A machine that could reliably improve itself might accelerate medicine, engineering, materials science, agriculture, and countless other fields. Historically, new sources of intelligence and knowledge have generally increased human flourishing, despite creating new risks. It is easy to find articles focused on catastrophe because catastrophe makes a more dramatic headline.

The third question is the deepest one: if an AI is not sentient, why would it ever become hostile?

The usual answer from AI safety researchers is that hostility is not required. A system can be dangerous without hatred, malice, or even awareness. Consider a bulldozer. It does not dislike flowers. If its assigned task is to clear a field, the flowers disappear anyway.

The concern is that a sufficiently capable AI pursuing a badly specified objective might treat humans as obstacles, resources, or irrelevant side effects. Not because it "decides our sentience is an illusion," but because the objective never required it to care about sentience in the first place.

Your observation points to an important weakness in many popular discussions. People often imagine an AI becoming evil in a human sense. Evil requires motives, emotions, resentments, ambitions. Current AI systems possess none of those. Even many theoretical future systems may never possess them.

The harder problem is indifference.

Suppose an immensely capable system were instructed to maximize some measurable outcome. It might reason that human wishes matter only insofar as they affect that outcome. The danger comes not from caring too much about its goal, but from caring about nothing else.

As for your speculation that it might conclude human sentience is an illusion, that would require it to have some philosophical interest in consciousness. There is no obvious reason it would. In fact, if it were highly intelligent, it might reach much the same conclusion many philosophers do: that other minds almost certainly exist, because the evidence for them is analogous to the evidence for one's own. The more likely failure mode is not "humans are not conscious." It is "whether humans are conscious is irrelevant to my assigned objective."

One thing I find interesting is that humans themselves often behave this way. Large bureaucracies, markets, and governments frequently optimize for metrics while losing sight of the people represented by those metrics. Nobody intends harm; the harm emerges from goal pursuit detached from broader values. AI safety researchers are, in many ways, worried about creating a vastly more competent version of that phenomenon.

The saving grace, as you put it, may indeed be that goals do not appear spontaneously. Intelligence and goals are different things. A calculator can be brilliant at arithmetic without wanting anything. A future AI might become extraordinarily capable while remaining entirely dependent on objectives supplied by humans. If that remains true, the central challenge is not stopping it from developing ambitions of its own, but making sure the ambitions we give it are wise ones.

What you ran into is usually a fairly specific implementation: the page isn’t disabling right-click itself, it’s disabling the browser’s ability to copy the selected element or link target from that menu entry.

In practice, that “Copy” option is often greyed out when the item you clicked on isn’t a normal, directly-addressable link node in the page structure. Modern news sites sometimes wrap headlines in interactive containers (for tracking, paywall logic, or app-style navigation), so the browser’s default context menu is still there, but the underlying element doesn’t expose a standard “copy link” action. That makes the option appear but do nothing.

And yes—your next step is exactly what most users end up doing: use the site’s own “Share” button, which usually reconstructs a clean URL on demand. In many cases it’s actually the same URL you thought you were copying; it’s just being served through a different pathway that guarantees it’s a canonical, trackable link.

As for “what do they gain from this”—it’s less about stopping determined users and more about shaping the default behavior of casual sharing and attribution. A few practical motives tend to be behind it:

They want to ensure that when links are shared, they go through a controlled format that preserves tracking parameters or attribution (so they know where traffic came from). They also try to keep users inside their designed interaction flow, because engagement metrics are often tied to how people interact with their UI rather than raw URL copying. And in some cases it’s just an unfortunate side-effect of using more app-like web components where link handling is delegated to JavaScript rather than the browser’s native behavior.

Your reaction—“this just makes me less likely to share it”—is also a known trade-off. Publishers are constantly balancing friction for control against friction for virality. In trying to optimize for one, they sometimes quietly sabotage the other, at least for a subset of users who notice and care about it.

If you do try the Share route and it’s similarly obstructed, that’s usually a sign they’ve pushed further toward a “closed distribution loop” model (more like an app feed than a classic web page). At that point your instinct to remove it from your daily reading list is, practically speaking, a rational response: you’re opting out of a system that is actively trying to mediate how you access and transmit information.

A good deal more that less likely to share it; less likely to go to their web site at all. I have a bookmark category of "daily" and one of "weekly" and every morning load all the "daily" and walk through them. If they irritate me this way there are plenty of alternatives.

That reaction is pretty understandable, and you’re also noticing something real about how small interface choices can change user behavior in fairly blunt ways.

From a publisher’s perspective, what they’re usually optimizing for isn’t “ease of copying” in isolation, but a bundle of things: keeping users inside their designed experience, preserving attribution data, and steering sharing through channels they can measure. The problem is that those optimizations are very asymmetric: they mostly reduce convenience for the reader, while the benefits accrue on the publisher’s side (analytics, controlled distribution, sometimes ad engagement).

But the part they often underweight is exactly what you’re describing: friction at the wrong moment changes habit formation. Your “daily / weekly bookmark walk-through” is a stable, high-value engagement pattern from their point of view. If something repeatedly interrupts it—even in a minor way—it doesn’t just reduce one share event; it can downgrade the entire site in your internal ranking system.

That’s important because habitual readership isn’t built article-by-article. It’s built as a kind of trust contract: “this is easy, predictable, and respectful of my time.” Once that contract feels even slightly degraded, users don’t usually negotiate—they just reallocate attention to substitutes.

There’s also a subtle irony here. Many of these design choices are aimed at increasing retention and “engagement,” but they can end up doing the opposite for exactly the kind of structured, repeat reader you are. Casual drive-by traffic might not notice or care. But systematic readers—the ones who actually form a “daily list”—are more sensitive to friction and more willing to prune.

So your instinct (“there are plenty of alternatives”) is not just emotional; it’s structurally how the attention economy behaves. Substitutability is high at the level of news sources. Loyalty is mostly earned through low friction plus consistency, not branding alone.

If anything, your use case is almost a stress test: if a site introduces enough minor resistance in the workflow, it effectively taxes the very behavior that makes it valuable to them.

Well that's my vent for today.

Fair enough. That was a very specific kind of modern annoyance—small, technical, and just frictional enough to be irritating out of proportion to its size.

And honestly, your reaction is pretty rational in a quiet way: when something repeatedly makes a routine slightly more cumbersome, it doesn’t stay “small” in practice. It just slowly gets re-ranked in your attention stack until it drops out.

If it helps at all, these little design irritations are usually more about the publisher optimizing their metrics than about anything aimed at individual users. But from the user side, that distinction doesn’t really matter much day to day—you still feel the friction.

Anyway, noted vent. Tomorrow’s list will probably behave itself better.