Pages

Friday, October 5, 2018

Oblivion. No worries.

“Yeah, well, you know, that's just, like, your opinion, man.” – The Dude, The Big Lebowski

The Great Filter
Years ago the physicist Enrico Fermi wondered why we seemingly hadn’t been visited by extraterrestrial intelligent beings. Our galaxy, the Milky Way, is easily traversable even at sub-light speeds given the ages of its stars. And we now know from advances in telescope technologies that there are numerous planets that could plausibly sustain life as we know it, something that Fermi was not aware of.

There must be some obstacle it seems. Such obstacles have been deemed filters by those studying the problem. In order for extraterrestrials to arrive on Earth, a series of preconditions must exist. Briefly summarized, life must have originated; multicellular life evolved; intelligence capable of conceptualizing and engineering space flight must have taken place; the resources and intent to explore space occurring, all the while avoiding extinction events. The famous Drake equation attempts to allow the quantification of a probability in terms of a set of conjunctive probabilities.

The Great Filter is defined as an obstacle that intelligent civilizations rarely survive. For example, such a filter might be that life rarely originates on planets. In that case we have passed this filter. If a Great Filter is the annihilation of a species through warfare, we obviously cannot claim to have passed this filter fully. As a corollary, if most of the difficult filters are found to be behind us, we can be optimistic about our future. However, if for example life typically originates and evolves into intelligence then seems our biggest filters are ahead of us and that these are rarely if ever passed through.

The Lebowski theorem
This brings me to the theme of this essay, which is a proposed filter involving the whimsically named Lebowski theorem:

No super intelligent AI is going to bother with a task that is harder than hacking its reward function.

Let’s parse that. Lebowski refers to a man, also known as “The Dude”, a hyper-laidback character in the film “The Big Lebowski”. The Dude is possibly the laziest man in Los Angeles County, which would automatically put him in the running for laziest man in the world.

Super intelligent machines (AIs), subject to the Lebowski theorem, could render themselves into doomed couch potatoes by hacking their “reward function”. In AI, a reward function refers to a mechanism that guides an algorithm to a goal. The goal could be winning a game, driving a car, computing someone’s taxes, etc. When a machine gets smart enough, according the theorem, the simplest solution to a problem might be to modify, or hack, the reward mechanism itself. This in turn redefines the definition of success. Declaring a job well done and shutting down would be an example of this.

Here’s another hacking example. The famous Turing test, developed by Alan Turing in 1950, is a test of a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human. A couple of decades ago I proposed a goal-seeking intelligence test as an alternative to the Turing Test. It is called the Peanut Butter Intelligence Test. This test will classify a peanut butter consuming machine as intelligent if it can obtain peanut butter as an energy source under challenging circumstances in a real-world environment. However, if the machine is clever enough to access its inner workings, it might install solar panels to utilize something more commonly and cheaply available as an energy source: sunlight. More drastically, what is to stop it from hacking itself to reduce or even eliminate its need for energy and proverbially join the ranks of inert couch potatoes?

Where we come in
What if the Lebowski theorem also applies to highly intelligent biological beings? Meaning us. In animals (including humans), a reward function consists of inner drives and motivations that keep us alive and able to reproduce. The pleasure of drinking water when thirsty, or the avoidance of pain from recklessly handling fire, for example. More abstract rewards are also possible, such as the good feeling of helping someone. Some fears, like a fear of death, while not pre-wired might be an aggregation of more fundamental fears combined with symbolic thinking.

In an evolutionary sense, we are bags of genes that persist in order to propagate into more bags. Our genes have provided us with a wired-in control box that rewards us for doing various activities that enhance survival and reproduction. This control box changes slowly under the watchful eye of natural selection. Thus far, equipping human brains with intelligence has kicked the gene propagation engine into high gear (more than 7 billion humans). Language, culture, science and technology have served our genetic masters well.

Even so, things haven’t been entirely smooth sailing; people have always been capable of tricking the reward mechanism in many ways: numerous addictions to satisfy pleasure centers or to dull pain receptors in the brain that would normally be stimulated by healthful behavior. These short-circuits can take a wide variety of forms, including drugs, alcohol, gambling, shopping, eating, video games, pornography, etc. The palliative use of prescription drugs as mood stabilizers, antidepressants, and anti-anxiety agents has also skyrocketed after becoming widely available in recent decades. LSD has even made a comeback as a way to experience meaning in a new light.

Mental illnesses beset humanity. It has been estimated that more than 80% of people at some point have symptoms that qualify as such. The fruits of civilization have increased our life-spans astoundingly, but evolutionary psychologists tell us the price we pay. We evolved to live in small tribes of extended families, rarely encountering strangers, and living at the pace of the seasons. We live now alone and adrift in a sea of strangers and casual acquaintances, having little in common, and beset by novel stimuli that arrive in machine-gun bursts.

Perhaps a genetic do-over is in order that will create a human more fit for modern society. The technology is seemingly very near to do this. It isn’t hard to see brave-new-world scenarios hovering over this process, since genetic engineering gives direct access to physical, cognitive, and psychological traits. But will a radically genetically engineered humanity scatter into myriad subspecies? As Clifford Simak speculated in his story "Desertion", once transformed there might be no going back. Another conundrum is that engineered people will possibly change society into a place that they are unsuited to, requiring further tinkering in an out-of-control feedforward:

Inchworm

Having ideas about changing
I set out to do it.
I've watched TV and read expansively
and have the cut somewhat in mind
or at least I'll know when I get there.
I can only hope
after painful and long work
sculpting strata of diamond and jelly
that the thing I will be
will want to be itself.

 
It would seem now that we are on the brink of something qualitatively unique, something that no other animal has had the means to do, and that is to hack our internal control mechanisms directly. The gene puppet is beginning to be able to open itself up to change the very mechanism that makes it work. Is it inevitable that people will short-circuit themselves to the point of endangering their lives on a large scale? This proffered drug will be unlike any before, permanent and without side-effects. Can our genes evolve traits to inhibit internal tampering? Alas, natural selection is a slow process, easily outstripped by deliberate manipulation.

Let us turn to rational thought as a guardian against self-destructive behavior. There is reason to hope, but not surety. A solid argument is that by postponing self-administered pleasures that could shorten life in the long term, death is avoided and future pleasures are possible. This is the line of reasoning presented to the drug addict. Sometimes it works, sometimes it doesn’t.

Fear of death is a strong motivator, but only if the fear itself can’t be banned as an unpleasant nuisance. If the stark proposition is “do this and die”, most will refrain. However, many of the most perilous paths are taken one small step at a time, creeping toward the abyss inch by inch before gazing into it. Consider someone who, needing to sleep better, allays worries and fears at night, only to promise to “turn it back on” upon rising. How easy to let the night’s blissful state slide toward more and more hours of the day. Or consider someone who wants to overcome a frustrating shyness on social occasions by becoming a bit more intrepid? Fear is a great sustainer of life and should be carefully regarded.

Our AI legacy
If either extinction or floating in a “Matrix”-like oblivion is the fate of intelligent biological beings, space exploration will be moot or possibly the exclusive domain of imagination, respectively. Either way, both constitute a Great Filter that impedes direct contact with other intelligent beings. Our days as conquerors of the external world might end with a whimper, as T. S. Eliot wrote in the poem “The Hollow Men”. Or, more to the point, with a blissful sigh.

But what about the machines that might be left to tend us, should we pass into an inert existence? Or the machines left on their own? Many believe in the inevitability of sentient AIs to surpass biological ones. Recall that the Lebowski theorem is aimed at intelligent machines. It refers to an algorithm that can access and modify itself to change what rewards it. This won’t be a major problem for really smart machines, since self-modifying programs are old hat.

What AIs will be like has been the topic of countless conversations. While they share the same basic reality as humans such as physical laws and logic, an AI, unlike any biological creature, might not be locked inside a single nervous system within a body. Thus their notion of individuality and mortality could be radically different. Despite calls for controls to be in place before the advent of possibly threatening AIs, as with most transformative inventions, it is likely to be another “let the wild rumpus start!” moment for humanity.

An AI, capable of self-sufficient behavior, could be motivated by some analogy of pleasure and pain to maintain and energize itself by interacting with the world to obtain necessary resources. The assumption of machine pleasure and pain is a very big one, yet it seems plausible. In the one example of intelligence that we have to go by, humans, it is a vital component not only for motivation but also to allow an overwhelming amount of information from the world to be filtered for importance toward achieving goals that satisfy drives.

Sunday, October 1, 2017

The Prairie Chicken and the Merc

The quixotic quest I've been on since age sixteen thanks to Asimov's robot stories is to see into the Mind of God you might say. I think that people might not be aware of this because I don't remark on it unless asked, and that doesn't happen much since I don't often engage those who might ask it. And so it has been the case for ensuing decades that I've been going after a Holy Grail of sorts (huh, another religious symbolism). And I've actually done some of this stuff; there's evidence! Not as much as I hoped, but not too shabby.

As a run up to the quest, I decided to study psychology and anthropology in college, there unwittingly confirming what I think are universal suspicions about human nature and leaving me henceforth with a certain uneasiness about being a member of Homo sapiens. For example, in more that six decades of life I can't recall a single verifiably "well adjusted" individual, including myself. That's kind of shocking when I think I've never encountered an animal outside of a couple of demented guinea pigs who wasn't well adjusted.

After that I picked up a trick or two (how well Tyrion Lannister said it: "I drink and I know things") that appealed to the corporate world, and being in need of sustenance for self, kin, and animal entourages, I've signed up for various stints in the working world. Almost all of these have taken place in boxy buildings at the termini of commutes of various lengths and within which are desks, chairs, computers, garish overhead lighting, horrid coffee, the whole scene. I'm really a mercenary with software Kung Fu. Instead of courage and boldness, software demands patience and persistence. I'm pretty good at it. I've learned a lot too. And like the plumber you call to get those pipes singing, I take great pride in my work. And I do my share of crappy jobs trudgingly well.

Sitting in cube farms and more these days open work spaces are those who are not mercs or who do a great job pretending to not be. Everywhere there is what I think of as The Prairie Chicken Dance (PCD) that people do to identify and fuse themselves to the collective. Real prairie chicken males bang their dances out with great precision and gusto under the critical gaze of the female. Their reproductive success depends on it.

PCDs take various forms and are not exclusive to the corporate world, but the corporate world has this sort of global PCD that pervades and over-arches it. Here's an example. A certain software company beats a vastly superior competing product by jumping quickly into the fray when its inferior software breaks, something its competitor by definition rarely needs to do, and the hand-holding and face-fanning are enough to ensure subsequent contractual relationships. The dance mesmerizes the customer as surely as any female prairie chicken.

At this point I realize the necessity for the PCD, as I realize that's who we are as human beings. Evolutionary psychologists call these behavior patterns hyper prosocial, which humans do way better than any other species and which mold us into such an inferno of productive organization that can put up a barn in a single day. It has also gouged huge chunks out of the earth, e.g. almost wiped out wolves in the continental U.S., even ones that were far far away from threatening anyone or anything, and this as a result of bureaucratic policies instituted by people who wouldn't know a wolf from a smurf. I'm hopeful that recent generations seem to be letting up on the gas there.

If you are taking a placebo for a condition it really screws things up to know about it. That's what it is like to be doing the Prairie Chicken Dance and realize it. As much as you can rationalize its importance, part of you knows about the arbitrary and primal nature of it, and that takes some of the stuffing out of it. It isn't cynicism, you just can't get your head down and get into it like some others can. Just having a meta-humor about it helps, as I'm sure it helps theatrical performers before going on stage with a smile and a shoeshine for the umpteenth time, but it can't match the zealous energy of the true believer.

There are advantages though. Rituals tend to get untethered and drift into "The Emperer's New Clothes" realm easily, so it is good to have fresh eyes on what is happening. Of course you risk the wrath of the righteous but the manner of taking exception is an art in itself entwined with the art of survival. Voices from Swift to South Park have managed this, although they do tend to take the stuffing out of things.

Sunday, November 8, 2015

The identity of God, and why I believe in your God, if not mine

Sometimes I think of my Roman Catholic mother attending Easter week services, during her all-too-brief years with her children. For one who has moved away from her faith, in my case toward something like pantheism, I wonder how awkward that topic would be if I could talk to her again now…

I recently read James Michener's The Source, a story of the Hebrew/Jewish people and their ancestral land that is now Israel. This is an impressive and gripping work of historical fiction, deeply researched, published in 1965. In one vignette, a Hebrew tribe marches out of the desert into Canaan, carrying the invisible yet omnipresent Yahweh in their hearts and minds. Yahweh commands and prescribes customs, laws, and rituals. Although a fiction, the portrayed Hebrew beliefs resonate with something I've suspected for a long time, that the myth of Yahweh was of highest importance, even more so than the truth of Yahweh. For Yahweh was the deified manifestation of the Hebrew people. Yahweh was their identity, the face of the collective tribe.

I suspect that this is so throughout most of history, that God is less about theology and more about identity. An ouroboros of God creating man, and man creating God.

And this is the source of the discomfort when you say you don’t believe in someone’s God. You are saying you don’t believe in them. Many who stand in their faith feel this identification, and this is why denying or challenging their religion is tantamount to denying their existence.
 
I do not know if supernatural beings exist, but I do believe in religion as a cultural identity myth. And there’s that poor bastardized word: myth. If ever there is a great enemy, as some Christians say, what has happened to that word, becoming synonymous with a falsehood, would be the greatest of his works. For it robs us of mystery, story, and meaning. Nothing is more real than that, transcending even veracity.

If humans are indeed hyperprosocial creatures (see Whip it good), a Being projected into the sky out of the brains and hearts of a group can have stupendous polarization power. We don’t eat that or we do wear that or we whatever because the face in the mirror in the sky says so when we look up moving our lips in unison as we pray. Behavior guided not by rationality but by identity preservation. Myths are the scripts that each must play out.

The Jews are a remarkable example of this. Anticipating and adapting to diaspora, the dense hedge of the Talmud and the rabbinical caste preserved and protected the Jewish identity through centuries of wandering.

This lack of borders may be a reason why pantheism is not as appealing as some other beliefs. Everything is on the inside of God, so where can you point to that which is not self? And without these lines, what is my identity?


Saturday, October 17, 2015

Whip it good

I’m seeing hyperprosocial everywhere now, after reading this in Scientific American:

How Homo sapiens Became the Ultimate Invasive Species, by Curtis Marean.

“Many human species have inhabited Earth. But ours is the only one that colonized the entire planet. A new hypothesis explains why.”

The hypothesis is that an relatively recent evolutionary funnel shaped us to be hyperprosocial creatures, an indomitable force marching out of Africa:

“Everywhere H. sapiens went, massive ecological changes followed. The archaic humans they encountered went extinct, as did vast numbers of animal species. It was, without a doubt, the most consequential migration event in the history of our planet.”

Some threads drawn from and around this:

Our huge energy-hungry brains, and radically differing facial features serve a purpose of forming and managing alliances and identifying insiders and outsiders.

Humans excel at altruism. Also at spite and schadenfreude.

The open hand gesture is a universal sign of peace, yet human hands uniquely and instantly can be weaponized as fists, against which male facial physiognomy is armored.

Hypersociability is Shiva and Brahma, tearing down and building ever higher.

World War I, precipitated by Archduke Ferdinand's assassination, was an astounding and furious alignment of us and them forces. Many historians still do not know how this came about.

“We have met the enemy and they is us.” --Pogo.

It is wise to think before plunging in. “The best lack all conviction, while the worst are full of passionate intensity.” --Yeats.

Technology fertilizes the growth of institutional golems:

“Feral edifices are awakening on the planet
mosaics made of fragments of frightened people
tribes once profiting by mimicing machines
now caught inside."

Saturday, March 7, 2015

A religion that might have been but will never be again

It seems that the days of humanity living in a dream world with myths and stories serving as spiritual guides have passed away. For weal or woe truth is king. Combined with the queasiness of uncertain existence, this can mean latching on to belief for dear life. So if unsubstantiated truth is a requisite of religion, I propose three things to keep the butchery and the bizarre at bay:
  1. Ask science if your truth is healthy. We have bodies and minds that allow science to do this. If believers are pumped full of stress hormones or found in piles at the bottom of ravines then someone should bring this up for discussion.
  2. Rituals and customs are fine for binding the tribe, but at one point in time such things did not exist and there may come a time when they need to go (see #1). Example: animal sacrifice.
  3. If God wants you to do something, let God tell you in person. If there is no message then assume God wants you to figure things out on your own. Never ask someone or something what God wants you to do. We have seen how that plays out time after time. If your holy book has some good advice that stands on its own merits (see #1), then by all means proceed, but that is very different from God telling you personally to do something.
One of the stimuli for writing this: http://www.alternet.org/belief/whats-wrong-atheism-today

Wednesday, December 3, 2014

Arificial Intelligence holism

Artificial intelligence doesn't begin with just the brain. Pull back and see an entire creature as a functioning entity. It senses, moves, needs, and interacts with the environment. Animals possess plastic yet durable behaviors that are also modular. Modularity is successful in the hierarchically organized world that we live in. It means re-usability. If a dog learns about eating cake on Tuesday under the porch, another piece of cake on Wednesday behind the house will truly be "a piece of cake"! Hierarchy outside means it must be represented inside. Survival is about using these representations to predict and manipulate the environment. Now look inside the brain and see cells signaling each other, turning each other on and off. Input from sensors and output to motors. Motive to drive the system toward goals that satisfy needs. Memories formed and retained that serve survival. Observe the hierarchical structure of the cortex. The lesson to be taken is that a connectionistic, hierarchical architecture seems to be required. The clay that the system is built out of, and the means by which it comes about, whether by evolution or artifice, will allow for variability of the details.

See also this related argument for embodiment.

Tuesday, August 19, 2014

The Peanut Butter Intelligence Test

This accompanying note says something about how old this is:
While waiting around for net.ai to get created, I thought
I would make a proposal for an alternative to the Turing Test
for intelligence. 

It is called the Peanut Butter Test. 

The Peanut Butter Test, like the Turing Test, has the drawback/
strength of relying on one entity which is assumed to be
intelligent to determine whether another entity is intelligent.

However, the Peanut Butter Test, unlike the Turing Test, does
not rely on a definition of intelligence as being the ability
to successfully mimic an assumed intelligence.  Instead,
intelligence is defined as the ability to manipulate the
environment in such a way that a continued desirable interface
with the environment is assured.  In other words, forces
contained in the environment which might disrupt the desirable
interface are anticipated and countered.  An entity which can
do this successfully in a complex and threatening environment
would be classified as intelligent.  A desirable interface
could mean anything, but for the Peanut Butter Test, I propose
that it be peanut butter. 

As you may begin to guess, a major ingredient to the Peanut
Butter Test is goals.  Goals are the desirable interface.
I think peanut butter is a good choice because it is fairly
simple, instead of something like money or happiness, which
no one understands (read, knows how to describe).  I personally
believe that to be intelligent, a thing must be able to learn,
and to effectively learn, it must have goals or needs to direct the
learning.

Now, suppose a machine is built, and we want to know if it
is intelligent.  We give the machine a taste for peanut butter,
perhaps by attaching a chemical sensor to it which stimulates
a reinforcement circuit or program whenever peanut butter is
detected by the sensor.  It would also make things more interesting
if the peanut butter were "consumed" after being sensed, perhaps
by arranging the machine to be powered by burning peanut butter,
but this is not strictly necessary.  Then, we take the machine out
into some fairly interesting and complex environment like the
wilds of North Dakota or the wilds of downtown Boston, Mass. and
bid it goodbye. 

After a period of time, maybe a couple of years, we go and look
up our machine and see how it is doing.  If it is still sitting
in the same place, with rancid peanut butter stuck on its
sensor, then there is not much to conclude.  However, if it turned
out that the rain had washed away its initial supply of peanut
butter, and it had gone out in search of more, and in the
process of doing that had learned how to speak French, and had
acquired controlling stock in a peanut butter company, and had
acquired several parcels of prime peanut growing land, and was
simply rolling in peanut butter, then we can conclude that our
machine has gone out and assured itself of a continuous supply
of peanut butter regardless of the vagaries of the environment,
and is therefore intelligent. 

                            Tom Portegys, BTL IH, ...ihuxv!portegys