First, I don't advocate amygdalectomy. I just want to point out that it is technically possible to eliminate suffering without eliminating pain. It reduces reproductive fitness, but I think it is more practical than some suggestions within the Effective Altruism movement to genetically modify carnivores to become vegetarians.
Second, by "we" I mean humans. It is important to maintain a distinction between humans and robots by not programming them to express feelings or consciousness. Human civilization was made possible because we evolved empathy, meaning we tend to cooperate in the prisoners dilemma and not pull the lever in the trolley problem. Machines could exploit this to benefit their owners. Third, to deconstruct my maximum utility argument, I am using Legg and Hutter's definition of universal intelligence as the expected utility or reward gained by an agent in a universal (Solomonoff) distribution of environments in Hutter's AIXI model. In AIXI, the agent and environment are Turing machines that pass symbols back and forth at each step, and also the environment sends the agent a numeric reward or penalty signal. The optimal strategy for the agent is to guess that the environment is the shortest program consistent with all the past interactions and run it with each possible output to see which produces the greatest reward. This strategy is not computable because it requires computing the Kolmogorov complexity of the interaction history. Thus, agents have to adopt a less optimal strategy, such as the human strategy of repeating actions that were rewarded in the past. Real agents and environments are not Turing machines because Turing machines have infinite memory and we live in a finite universe where the largest possible memory is Mc^2/(kT ln 2) ≈ 10^92 bits, where M is the mass of the universe, c is the speed of light, k is Boltzmann's constant, and T is the temperature of the universe, taken to be 3 K for the cosmic microwave background. Seth Lloyd also estimated 10^90 bits could be encoded in the positions and velocities of all the particles in the universe within the resolution of the Heisenberg uncertainty principle. But a Turing machine is a close enough approximation for computers with more than a few hundred bits because there is no practical difference between a program that might run forever and one that must either repeat a state or halt after the heat death of the universe. Next, I model happiness as the rate of increase of utility because it seems to be a good model of human behavior. The more you have, the more you want. The US and EU have the richest economies in the world, but the number one political issue in both regions is the economy. It's never good enough. Next, since agents and environments in a finite universe are finite state machines, any possible utility function over a finite set of agent states have a maximum. There may be more than one such state, but no state is preferred over any other and serves no purpose from the viewpoint of the agent. Such an agent would be a zombie, performing a computation for no reason. Maybe I am using the wrong model of happiness. I am open for suggestions. -- Matt Mahoney, [email protected] On Sun, Sep 6, 2026, 11:14 AM James Bowery <[email protected]> wrote: > > > On Sun, Sep 6, 2026 at 8:38 AM Matt Mahoney <[email protected]> > wrote: > >> >> On Sat, Sep 5, 2026, 6:13 PM WriterOfMinds <[email protected]> >> wrote: >> >>> Deactivating your amygdala so you can't feel bad about anything is akin >>> to using mind-altering drugs to make yourself feel good about everything. >>> They're both forms of wireheading. Some of us are not naive reinforcement >>> learners and won't fall in that trap. >>> > I'd say technological civilization is what Matt refers to when he > obsessively uses the word "We" and that technological civilization, in > O'Neill's terms (hence Bezos's "Maui on its best day" terms regarding > O'Neill space habitats) is a manifestation of said "wireheading". > > E. O. Wilson pretty well summed it up in "The Social Conquest of Earth" > when he described eusociality as parasitic castration of children to > produce sterile workers. Amygdalectomy is akin to castration to produce > "angelic" sterile workers in a eusocial hive. This was the point of the > Soviet era novel "We" that the Soviets suppressed. > > But it also explains the amazing phenomena of mothers parading their moral > vanity by declaring their newborns as "trans" and the equally amazing > phenomena of many in the Democratic Party supporting government funded > "transitioning" of infants. > > In "The Genetic Omnidominance Hypothesis" I described this kind of thing > (including amygdala "castration" decades before it became a *ahem* popular > suggestion) as an extended phenotype of genetic dominant traits silencing > genes in other humans during global panmixia of populations. Matt is > describing humans as robots -- which is exactly what sterile workers are as > described by E. O. Wilson. A more evocative term is "slaves" of the > genetically omnidominant. > > Slaves freak out about this hypothesis not because of minor flaws in its > exposition that they point to, but because it violates the genetic > interests of their masters. > > > ... But all possible utility functions in a finite universe have a >> maximum state, which is death. >> > > This is the crux of Matt's argument and it would be good to deconstruct it > precisely before further palaver. > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + > delivery options <https://agi.topicbox.com/groups/agi/subscription> > Permalink > <https://agi.topicbox.com/groups/agi/T906c878e4ba411be-M91c3f558fe7b7cbbdaddbe05> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T906c878e4ba411be-M28b6395e617b1e0753020e1c Delivery options: https://agi.topicbox.com/groups/agi/subscription
