The Wirehead Problem and Money as AI

Here is a potential problem facing engineers of artificial intelligence: A sufficiently intelligent agent will work to tap its reward source directly, without doing otherwise useful work.  In this video Tim Tyler invokes the heroin addict for comparison. He concludes that a superintelligent agent will calculate the long-term effects of becoming a junkie and likely will conclude, as most people do, that it's not worth it.

Tyler goes on to counter this argument by conjuring a monopolistic threat.  What if the superintelligent agent wiped out its competitors?  Then, under the dynamics of a monopoly, it would grow flabby and careless, because there would be no reward in making effort in any other direction. And this situation would become conducive to its yielding to the tempation to become a junkie.

It looks like the threat already has landed. The U.S. dollar/federal reserve note is the superintelligent agent Tyler warns against. The U.S. dollar has tapped it own pleasure center, and in an accelerating cycle of masturbations it has exhausted itself.

Each step in the deregulation of the U.S. banking system brought the dollar closer to its source. Finally, when the real-estate bubble burst, the fallout revealed that the originators of money, the private banks, had crossed a deregulatory threshold, and they had exploited the lifting of essentially all conditions on the granting of loans. Because banks originate dollars by crediting loans, unconditional lending became the U.S. dollar's unfettered tap into its own pleasure center, its source. Now the junkie is going through withdrawal. And if the U.S. dollar gets fired from its job as the world's reserve currency, losing its monopoly privileges? Cold turkey.

Money as superintelligent agent reveals a bias in AI theory, or at least in AI practice, insofar as AI focuses on circuits and programs and patterns of atoms. But money is pure abstraction. It can take any form: seashells, printed paper or electronic accounting entries. Higher intelligence doesn't get bogged down in precise material specifications. It is conceptual.

Consider that money can do many things that you cannot. Indeed, it uses you, and the rest of us, to do its bidding. It is not an enabler, as we're taught to think. We are its enablers.  Look at the wonders it has created with our help.

The terrors that it has instigated? Our myth systems teem with wrathful deities and gods of destruction. Maybe the religious sensibility knows something about the superintelligent abstraction that provokes and persuades and thereby gives form to the world. G-d is a junkie?    

Can Natural Selection explain cell differentiation?

Darwin used the phrase "Descent with modification" to summarize his theory of evolution. Despite many particulars that more recent science has amended to Darwin's theory, the basic idea of descent with modification remains. The descendant modifications are taken to be driven by varying degrees of "fitness" to environments. Even according to today's NeoDarwinian (or Modern Synthesis) model, the modifications observed in descendants are taken to be the result primarily of adaptations to environmental contingencies, operating under the mechanism of variation + selection, which runs without the benefit of any plan or program that might provide direction. And this "blind" process has produced all the phenotypes that were or are or will be. (An emerging Extended Synthesis model relies less on exogenous factors to explain phenotypes. Its influence on the discipline of evolutionary biology is yet to be known.)

However, another case of biological descent with modification apparently does benefit, or is assumed to benefit, from a guiding plan, or program. That is the descent of various tissue types from an undifferentiated zygote during ontogeny. This poses a paradox.

If natural selection is so powerful a causal agent that it can generate all the phenotypes that make up an ecosystem, then why is it necessary to suppose that there occurs in a zygote some sort of genetic plan or program that guides development of the organism? Why not just chalk it up to natural selection -- competition and cooperation among the cells in the organism? What evidence is there of a developmental program?

All the cell types that make up the body of a complex organism share the same genotype but differ as to which genes are active and which not. And that info must be heritable, hence a source of variation ("copying errors"). But any variation among cells in an embryo might provide an advantage to some cells and/or disadvantage to others. So, the stage is set for natural selection.

The tissues that make up a complex body and their symbiotic interdependencies are just the happenstance of competition among the cells -- is that a defensible proposition? The fit survive and go on to take their place in the somatic ecosystem of the body. The unfit go extinct. A clear case of unguided evolution. No need for a developmental program.

I am NOT proposing that this is what happens. I am only asking the question: What OBSERVATION could disprove this argument -- that the cells descend with modification from their common ancestor, a zygote, through a process of variation + selection?

  comments at

http://www.thescienceforum.com/Can-Natural-Selection-explain-cell-differentiation-30656t.php