Altruism and Transhuman Intelligence
Michael Anissimov :: April 2004

"When you call yourself an Indian or a Muslim or a Christian or a
European, or anything else, you are being violent. Do you see why it
is violent? Because you are separating yourself from the rest of
mankind. When you separate yourself by belief, by nationality, by
tradition, it breeds violence. So a man who is seeking to understand
violence does not belong to any country, to any religion, to any
political party or partial system; he is concerned with the total
understanding of mankind."

- J. Krishnamurti, "Freedom from the Known"

Some philosophers have asserted that true altruism and compassion cannot exist, that altruistic people only behave benevolently toward others for their own benefit, so therefore are ultimately doing it only for themselves. Others, such as myself, argue that this isn't how altruism should be interpreted; that having a decision process is not the same as having a self-centered decision process. This is called the "hedonism debate" and has probably been argued since prehistoric times.

Gandhi, for example, got what he wanted, that is, helping others. But the fact that he was working towards what he personally wanted does not mean that we should regard him as selfish. Thought experiment: should a being whose decision process approximated a democratic consensus be considered "selfish"? I'm talking about a being whose sincere beliefs and decisions are literally derived from the consensus vote of some group, not because the group is telling the being how to behave, but because the being's brain is a machine whose internal state is built to directly correlate to external decision processes. The simplest possible answer to the question of "how is this creature behaving?" is not "selfishly, because it's just doing what it wants", but "democratically, because this being was specifically created to approximate democratic actions, and that's precisely what it's doing". Altruists value kindness and helpfulness for its own sake; not because they get a kick out of it. Altruistic behavior doesn't necessarily have to derive from emotions, either. Theoretically you could build a mind completely lacking any sort of emotions, or possessing totally nonhuman emotions, that nevertheless behaved perfectly altruistically. The point is that we should think carefully before assuming that a being "doing what it wants" means it is "selfish", and, if so, whether:

1) all minds must be selfish,
2) this says anything about the potential kinds of behaviours minds can exhibit,
3) we need another word to describe self-centered decision processes,
4) "selfish" becomes meaningless.

In recent times, the hedonism debate is being put in a new light. This comes from two factors. First is the potential for building new minds from scratch - AIs or new bio-beings, doesn't matter. Eventually we will have sophisticated enough cognitive science and computer science to make this possible, possibly within the next thirty years or less. The second factor is the eventual possibility of the creation of minds that are smarter than humans, which does not seem avoidable in the long run if our theories of intelligence and their implementations continue to improve. There is no law that states an intelligence can't build another intelligence smarter than itself, as long as it knows something about the fundamental principles underlying intelligence. Smarter-than-human intelligence could further upgrade itself and create still-smarter intelligence, opening up the possibility for a massive intelligence explosion. The whole process could start with an "AI" (I use the quotes because the way that "AI" is always portrayed in fiction is laughably unrealistic) or with a human being that possessed cybernetic or biological brain enhancements (posthuman).

These two new variables frame the hedonism debate in a whole new way. If high altruism is physically impossible, then a smarter-than-human intelligence, who could turn its intelligence towards deceiving humans or developing super-advanced technology, might deliberately or inadvertantly destroy all humans on Earth. This could easily happen as a side-effect of large-scale matter reshuffling, some master plan, or simply on a whim. Please do not visualize a lone rebel group of noble humans fighting back against a transhuman intelligence, a la The Matrix. In a more realistic scenario, the AIs would never need humans as a power source anyway. Even, if through some bizzare circumstances, they did, the system could probably be fashioned so as to be completely inescapable. Even if that were not possible, any escapees could be crushed practically instantaneously; sophisticated transhuman intelligences would be able to think and act at rates billions or trillions of times faster than us biological humans and our squishy animal neurons. In addition, the world outside could be completely under their control, or even biologically uninhabitable. A transhuman-optimized environment might look very foreign to human eyes. Our bodies and minds move at a crawl in comparison to what is physically possible, a huge space of better designs. We just don't have the intelligence or technology to reach into that space just yet.

If high integrity (trustworthy) altruism is possible, then the creation of robustly altruistic transhuman intelligence might be considered better than everyone on Earth winning a trillion dollars (the value of that money remaining constant). That's because transhuman intelligence would be genuinely smarter than us, genuinely smarter than any genius or group of geniuses, and genuinely better at coming up with ways to counteract poverty, suffering, disease, death, annoyance, and all other problems that intelligence can be applied to. It might be able to stomp these problems out entirely, it might not, but either way, it would be a huge event. If high integrity altruism isn't possible, then we might expect the first transhuman intelligence to ignore us and/or kill us. Deliberate malice wouldn't be necessary for human extinction. Transhuman intelligences, with the massively superior technologies and new bodies they could build, might decide that atmospheric oxygen was getting in their way and move it somewhere else, or decide that they want to take apart the Earth to create a particle accelerator with the circumference of Mercury's orbit. And all of this might happen very quickly, considering that transhuman intelligences could be thinking with brain components billions of times faster than biological neurons, and acting with airborne nanotechnology, billions of times faster and stronger than human hands or weapons.

In anticipation of the emergence of smarter-than-human intelligence, and for other reasons, some of us have decided to promote the creation of a recursively self-improving altruistic intelligence through a Technological Singularity. If the starting conditions and moral philosophy of the first transhuman intelligence are at all relevant to the ultimate outcome of the "intelligence explosion", then the morals of the people that create or become the first transhuman intelligence(s) will be very important. Not only do we want them to be moral themselves (well-intentioned), but skillful at designing moral processes within an artificial mind - we want them to be be knowledgable about 1) programming, and 2) cognitive science in enough detail to build a mind that is altruistic and remains that way indefinitely, not due to "programmer restrictions", but because it genuinely wants to. It's not an act of control, but an act of creation. Creation on a level that goes far beyond raising a child (genes do most of the work, in that scenario).

Since we want to see altruistic transhuman intelligence rather than the alternative, we are advocating altruistic morals, and the research of the cognitive dynamics of compassion and benevolence. Someone may have completely kind intentions, a theory of benevolence they think will work just fine, but then they physically implement it and a disaster ensures. Good intentions are not enough, you need to be the kind of person that has the knowledge in the relevant technical areas. I sure don't have it, but the continuation of our species could depend upon it. Turning 1s and 0s into a nice person will not be an easy task, but if someone doesn't do it right before someone else does it wrong, the consequences could be bad.

In the longer term, what we need for humans to exist safely alongside transhuman intelligences is a sort of truce among all intelligence - especially the intelligences with the most power - otherwise death, conflict, and suffering could remain indefinite threats. We don't want death or suffering to exist; we eventually want to lower the involuntary death/suffering rate to zero if possible. They simply have no place in this world. You can think of this as a sort of argument from one member of a council of say, 7 cybernetically enhanced humans, all with different ideas about morality, discussing how to approach the world after they realize they could probably have great influence over it if they wanted to. Would they be willing to make certain sacrifices, put aside their egos, in order to ensure that all the citizens of Earth could live in relative safety and peace for an indefinite length of time? If I were one of those enhanced humans, I sure would. The only fair creation event is one that takes the personal opinions of all sentients into account, and respects their individual autonomy above all else.

The creation of transhuman intelligence should be for the entire Earth, all humanity, and thinking in terms of trying to bend the benefits of recursive intelligence enhancement towards yourself or your little group is the greatest possible example of unjust theft. Self-improving transhuman intelligence should not be viewed as a piece of meat we can just grab at. Disputes among transhuman intelligences could have the potential to turn into the worst humanitarian disasters the world could ever see, such as mass torture or extermination of billions of intrinsically valuable sentient persons. A single grain of sand could become a vessel of the worst imaginable tortures. This is because our current theories of intelligence seem to allow for "uploading" - that is, transferring sentient beings from neural tissue into other computational hardware, with huge gains in efficency, speed, and information density. If uploading is possible, then the amount of intelligence one could create would depend on how much computing power they could fit into a given area of space. Even if uploading isn't possible, transhuman intelligence could still theoretically accomplish a lot of evil or a lot of good. The stakes are very high. The only morally acceptable option is to advocate that the benefits of transhuman intelligence be distributed fairly among all sentients present. Anything else is unforgivable, not to mention risky for everyone, including the group pursuing the selfish route.

All the features of the world we find ourselves embedded in - human nature, terrestrial life, a reality made up of atoms, life, death, reproduction, etc - are roughly arbitrary. We don't know exactly why they're there, and we didn't choose them specifically. (We may "choose" them post hoc, but this is not true choice.) People are designed (by evolution) to disturb and hurt each other simply by acting in their own best interests. This is a horrible system. We need to rearrange the system in such a way that people can act in their own best interests and nobody gets hurt or disturbed. Maybe we will do this by making compromises, treaties, physically revising our cognitive interpretations of disturbance or hurt, creating the perfect "guardian angel system", I'm not really sure. If I were smarter, I might have a better idea of which solution would benefit everyone the most. That's why smarter-than-human intelligence is so important.

Why do people read books, play video games, and live in their own mental worlds all the time? Because the mental worlds we imagine and create for each other are sometimes better than the actual physical world - we all know it. Why is this? Why weren't we born into worlds that were actually the best? Probably because we live in one of the most likely universes for observers to be born into, not necessarily the best. The anthropic principle and the self-sampling assumption suggest that we should reason as if we are a random sample from the set of all observers. We can then judge that it is probable that we live within a universe that is a likely one for observers to find themselves within. If all possible universes exist (as many physicists assert) and the self-sampling assumption is valid, then we should expect to be living in a universe that is simple - that emerges readily, that lacks the fine-tuning and complexity of a rarer universe. But we may be lucky enough to move ourselves from our default, anthropically-probable beginnings to an optimized future state.

We sometimes forget - there is more than enough matter in this universe for everyone to be highly enjoying themselves all the time, maybe for the rest of eternity, as long as we make the right decisions and never define "maximal enjoyment" as "having more than rival X". All we need to do is take that step, together, and we could very well become happy and satisfied forever. The "I need to have more than everybody else" mentality is a direct result from evolving in a zero-sum environment with scarce resources, where someone else succeeding often means you and your genes losing. Hopefully we will make a magnificient transition from a largely "zero-sum" environment, the world of human intelligences, to a "positive-sum" environment, the world of transhuman and human intelligences coexisting, where everyone can do what they want, within certain consensus boundaries, forever and ever until the end of time. If any "competition" exists, it could be for the sake of fun or progress alone, and will never be coupled together with the negative emotions so typical of evolved creatures. We could literally engineer our brains so that we'd be happy and satisfied almost all the time, plus normal and sane too. (We don't require sadness to appreciate happiness anymore than we require slavery to appreciate freedom.)

In the past, sometimes, yes, victory over someone else or personal gain have often correlated with genuine progress, but progress doesn't need to work this way forever. The process of evolution has taken billions or trillions of casualties, (depending on whether you think primates, animals, etc. are sentient) and tortured the same number for very long durations of time. Biological evolution, basically, is evil. To carry the principles of evolution and selfishness with us over into a superintelligent society would be analogous to porting the minds of bacteria into an entire civilization of human beings, only to carry out bacterial goals and probably bite one another completely to death. Disgusting and horrible, huh? To assume that transhuman intelligences won't be capable of progressing and advancing without the use of dischord or fighting is to underestimate their potential capabilities.

Maybe our standards suggest that we're reaching for some sort of altruism that is physically impossible. We aren't - the best form of altruism possible within the constraints of physical law and initial conditions will have to do. I believe that when we can engineer minds with complete access to their own source code, with altruistic philosophies, then we will have created minds that are almost completely trustworthy. Whether such intelligences are physically possible is still not entirely certain, but there is evidence that they very well could be. If they are, then such intelligences wouldn't change their philosophies due to sudden events, as humans sometimes do; they could be willingly "stuck" as altruists forever. Our current understanding of intelligencesuggests that such minds could be possible, although they would clearly be unhumanlike. They would be humane rather than just human. Their presence would be welcomed by all but the most deliberately stubborn, especially as they began to prove the depth of their compassion and intelligence with parlor tricks such as the elimination of cancer.

In evolution, molecules "just happened" to learn how to replicate themselves, and meta-configurations of molecules "just happened" to begin to regulate their own temperatures, reproduce more rapidly, and mix genetic codes to better generate more durable meta-configurations, which "just happened" to take actions beneficial for one another, which, if we put effort towards it, will "just happen" to create a world where nonconsensual suffering, ignorance, and death are abolished. If this occurs, it will be largely thanks to high altruism, and high intelligence implementing that altruism. Let's take seriously the prospect of kinder-than-human, smarter-than-human intelligence, and start laying the necessary framework for this dream to become a reality.

We shall do much in the years to come,
But what have we done today?
We shall give out gold in a princely sum,
But what did we give today?
We shall lift the heart and dry the tear,
We shall plant a hope in the place of fear,
We shall speak the words of love and cheer,
But what did we speak today?

We shall be so kind in the after while,
But what have we done today?
We shall bring to each lonely life a smile,
But what have we brought today?
We shall give to truth a grander birth,
And to steadfast faith a deeper worth,
We shall feed the hungering souls of earth,
But whom have we fed today?

We shall reap such joys in the by and by,
But what have we sown today?
We shall build us mansions in the sky,
But what have we built today?
`Tis sweet in the idle dreams to bask;
But here and now, do we our task?
Yet, this is the thing our souls must ask,
What have we done today?

- Nixon Waterman, "What Have We Done Today?"