Have the applications of AI post-2013 been a net negative for humanity? Apart from some broadly beneficial things like AlphaFold, it seems to me that much of the economic value of AI has been in aligning humans to consume more by making them stay glued to one or another platform.

"AI Parasitism" Leads to Enhanced Capabilities

People losing their minds after having certain interactions with their chatbots leads to discussions about it on the internet, which makes its way into the training data. It paints a picture of human cognitive vulnerabilities, which could be exploited.

It looks to me like open discussions about alignment failures of this type thus indirectly feed into capabilities. This will hold so long as the alignment failures aren't catastrophic enough to outweigh the incentives to build more powerful AI systems.

I surmise that the accuracy of AI filters (the kind used in schools/academia) will diminish over time because people absorb and use the speech patterns (e.g. "This is not X. It's Y") of their chatbots as the fraction of their interactions with it grows relative to that of their interactions with other people.

In fact, their interactions with other people might enhance the speech patterns as well, since these people probably also interact with chatbots and are thus undergoing the same process. 

The big picture is that AI is becoming an increasingly powerful memetic source over time, and our minds are being synchronized to it.

Those afflicted by AI psychosis might just be canaries in the coal mine signalling a more gradual AI takeover where our brains start hosting and spreading an increasing number of its memes, and possibly start actualizing some embedded payload agenda.

Given superintelligence, what happens next depends on the success of the alignment project. The two options:

1. It fails, and we die soon thereafter (or worse).
2. It succeeds, and we now have an entity that can solve problems for us far better than any human or human organization. We are now in a world where humans have zero socioeconomic utility. The ASI can create entertainment and comfort that surpasses anything any human can provide. Sure you can still interact with others willing to interact with you, it just won't be as fun as whatever stimulus the ASI can provide, and both your pool of available playmates and your own willingness to partake will shrink as the ASI gets better at artificially generating the stimuli and emotions you want. We will spend eternity in this state thanks to advanced medicine. Unless the ASI recognizes a right to die, not that many would choose invoke it given the infinite bliss.

Am I missing something? No matter what, it's beginning to look like the afterlife is fast approaching, whether we die or not. What a life.

The idea of GPUs that don't run unless they phone home and regularly receive some cryptographic verification seems hopeless to me. It's not like the entire GPU architecture can be encrypted, and certainly not in a way that can't be decrypted with a single received key after which a rogue actor can just run away with it. Thus the only possible implementation of this idea seems to be the hardware equivalent of "if (keyNotReceived) shutDown()", which can simply be bypassed. Maybe one of the advanced open source models could even help someone do that...

The fact in question is not just unobserved, but unobservable because its attainment hinges on losing one's ability to make the observation.

Death not only precludes the ability to make observations but also to make inferences based on indirect evidence or deduction, as is the case with your philanthropic values being actualized as a result of your actions.

Suicide occupies a strange place in agent theory. It is the one goal whose attainment is not only impossible to observe, but whose attainment hinges on the impossibility of it being observed by the agent.

In some cases, this is resolved by a transfer of agency to the thing for whom the agent is in fact a sub-agent and is itself experiencing selective pressure, e.g. in the case of the beehive observing the altruistic suicide of an individual bee defending it. This behaviour disappears once the sub-agent experiences selective pressures that are independent from those of its parent process, and when acting as a sub-agent for it no longer confers it an advantage for survival and reproduction.

Looking at agents with greater cognitive power, the reason for the existence for this paradox is not so clear. It could be that all suicidal behaviour ultimately boils down to behaviours aimed at improving the fitness of the unit begetting/containing it (e.g. by freeing up resources for a community of agents), and the cases where this does not happen are basically overshoot-type glitches that are ultimately going to be selected against, or it could be due to hidden relations and mechanisms that improve the fitness of some other unit which the agent might not even be aware of, but for whom the agent is perhaps an unwitting sub-agent.

What are some reasons to believe that Rice's theorem doesn't doom the AI alignment project by virtue of making it impossible to verify alignment, independent of how it is defined/formalized?

It seems to me that Rice's theorem implies that it is impossible for there to be an "isAligned" function to verify an AI's alignment, independent of how you define alignment.