Tuesday, February 28, 2006

Outwitting Rogue Superintelligence

"Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultraintelligent machine could design even better machines; there would then unquestionably be an 'intelligence explosion,' and the intelligence of man would be left far behind. Thus the first ultraintelligent machine is the last invention that man need ever make."
-I. J. Good, 1965

Superintelligent machines are expected to usher in an age in which the pace technological innovation outstrips our ability to comprehend it. This era is known as the Singularity.

You say this Singularity thing sounds dangerous? Give that man a cigar Mars bar!

The key issue is whether runaway intelligence and technological prowess is something against which we can defend ourselves.

The worst strategy is to try to ban AI technology. To do this would require a totalitarian regime that would be worse than any rogue super AI.

Another strategy is to race to be the first to build a superintelligence so that you can ensure its friendliness. This is tricky, but not impossible. This is the goal of the Singularity Institute for Artificial Intelligence.

Today, it occurred to me that there might be another strategy. There might be a way to tie together many independent computers into a giant distributed supercomputer. The idea being that no single computer, or subcluster could oppose the combined defensive computing might of this distributed system.

There are two dangers to this approach. The first is that, in acting to defend itself, the defensive network might supplant the threat. The hunted becoming the hunter, so to speak.

The second danger is that such a network might be pre-emptively hacked and controlled by a superintelligence.

So, it's not really a perfect plan, but I wonder whether there might be some workable variation on this "defensive swarm" theme.

Naturalism

One thing that surprises me is the amount of confusion surrounding the term naturalism. Most people rely on their gut to tell them where the boundary is between the natural and the supernatural. However, if we fail to properly define this boundary, it makes no sense to make the distinction in the first place.

Naturalism isn't an arbitrary choice. By definition, the natural world is the world we can know anything about.

In order for knowledge of the world to be possible, the world must be consistent and governed by relatively fixed rules (laws of causality, in the broad sense of the term). These two axioms, consistency and causality, are the axioms of science and logical positivism. From consistency we get the laws of logic and mathematics, and from causality we get the scientific principles of mathematical modeling.

Is it possible to prove that the world is natural? I don't think that's possible. I cannot prove that the universe is consistent, nor that every event has a fixed cause. However, the natural world is the only world we can know anything about. Anything beyond naturalism (supernaturalism) must must violate either the axiom of consistency or axiom of causality (or both). That is, the supernatural must either be inconsistent or be exempt from any causal rules whatsoever.

I'll explain why this makes knowledge of the supernatural impossible.

Suppose we have a true proposition, P, about a supernatural cause. If the supernatural cause is not bound by consistency, then it is possible that the contradictory proposition NOT P (~P) is simultaneously true. For example, there is no reason why "God is good" could not simultaneously be true and false of a supernatural God. One cannot describe knowledge of both contradictory propositions as knowledge at all.

If the supernatural cause violates the axiom of causality, then there is no discoverable rule that relates a supernatural cause to experience. Any experience could equally well signify P as ~P. P can cause an effect E and also cause effect ~E. Likewise, ~P can cause an effect E and also cause effect ~E.

Furthermore, supernatural events would remain unexplained, and we would never know whether they were supernaturally caused or whether we just hadn't yet figured out the naturalistic cause.

You may have your own definition of naturalism. However, the definition presented here is precise and executable.

Now, given my definition, you may prefer to believe that God is not supernatural. In this case, God would be subject to certain rules, or perhaps be indistinguishable from those rules.

Naturalistic scenarios would include:
  • God is an alien whose technology is so great that it simply appears like magic to us.

  • God is another word for all the laws of the universe, or God actually is identifiable with the universe.

  • We are living in a simulation created by a naturalistic being whom we call God.
In these cases, it is theoretically possible to gain knowledge of God.

However, few theologians subscribe to the view that God is a naturalistic phenomenon. The reason for this is that it's emotionally difficult to assign divinity to a naturalistic being. I think that few fans of Star Trek would say that Q was a God, despite his possession of God-like powers. He's just an evolved superbeing.

However, the supernaturalists are trapped. If they grant that their supernatural cause is knowable (logical and rule-based), then it automatically becomes a naturalistic automaton.

In order to give God the attributes that make him worthy of worship (e.g., perfect goodness), we end up making the concept incoherent or stretching language beyond all meaning.

In the end, I am an atheist for two reasons. First, we cannot have knowledge of the supernatural (propositions about God have no truth values). Second, I hold no Gods before me. That is, if there were a God-like alien (and so far there's no compelling evidence of this), then, while I might do what it forced me to do, I would not regard it as a God.

The two axioms I have described here are not exhaustive because there may be consistent, rule-based ystems that we still cannot know. That is, proposing that there is some consistent causal system that explains the world does you no good if you also posit that this consistent causal system is unknowable.

Wednesday, February 22, 2006

Potpourri

A few notes on various topics that I haven't got around to writing about in detail...

In Iraq, militants, presumably Sunnis, blew up the al-Askari shrine in Samarra. It is believed that no one was killed in the explosion. Sunni militants have been trying to start a war with the Shiites since the invasion. They have killed hundreds of Shiites in cruel and vicious attacks. But now they've really done it. They've gone and blown up a religious building. Now the Shiites are enraged. Only now?

In the last three weeks, the world has seen violent Muslim protests over cartoons that suggested that Islam was a violent religion.

George W. Bush approved the transfer of ownership of U.S. ports to a government-owned company of the United Arab Emirates (UAE). The UAE was home to two of the 9/11 terrorists and was the port of entry for half of the 9/11 attackers. UAE does not recognize the state of Israel, and has been criticized for a lack of cooperation with the 9/11 investigation. I don't know exactly what port ownership entails, but I expect it is similar to the role of airline and airport workers at a airport terminal. Though TSA mans the security checkpoints, airport workers can bypass security systems. Still, Bush is consistent: his friends stand to gain financially from the transaction. All of this Republican corruption and corruption gets in the way of actually securing our ports, something you may recall John Kerry proposing in 2004.

As the probability rises that the Democrats will retake the House of Representatives in the autumn, George Will wonders whether Democrats will live to regret it. I see no such silver lining for the Republicans, but the fact that Will is writing about this is a sign that the Republicans are worried.

Since the Dover trial, the pro-ID bloggers over at TelicThoughts have preferred to write about breaches of protocol. They ask whether supporters of evolutionary biology separating their science from their belief systems? It's not relevant. Intelligent Design wasn't cast out of the schools because its motivation was religious, but because that's all there ever was to it.

There's a very interesting post over at another pro-ID blog called iDesign@UCI. A chap named Wedge responded to my posts, and he's a much more friendly and thoughtful fellow than most of the ID bloggers I've run across. However, he says something quite interesting in response to my criticism of CSI:
CSI-as-design-predictor is not a hypothesis that design theorists are testing by applying it to biological systems. If it were, your criticism would be accurate. CSI is an attempt to make rigorous the process of design inference that humans use all the time, and apply it to biological systems. If you don't think that CSI is a good indicator of design, I think you at least must admit that humans distinguish design from chance and necessity all the time. There doesn't seem to be any compelling reason why these criteria couldn't be applied to biological systems.
This reveals quite clearly that CSI is intended to as a measure of what "looks designed." In other words, ID via CSI is precisely the ancient argument for God by design. There's nothing more to it.

There's an excellent article by PZ Meyers on gene transcription factors over at Pharyngula. I haven't finished reading the whole thing yet, but I highly recommend it. Meyers deserves a medal for his efforts at public education.

Saturday, February 11, 2006

Moral Persuasion (or when to be convinced by a moral argument)

Crazy Talk
The deeper I get into debates over metaphysics, the more abundant the arguments from universal morality. They go something like this: "Since science and logic are unable to establish a universal morality, we must hold that there is more than science and logic, otherwise we are doomed to live in an amoral/immoral world."

This is the aforementioned crazy talk.

This argument doesn't work because it betrays a lack of understanding about how morality works in practice, and because it confuses a moral codes with the imperative to follow a moral code.

To expose the bankruptcy of the universal morality argument, I'll explain why subjective morality doesn't lead to immoral doom. Then I'll describe the impossibility and amorality of universal moral law. Finally, I'll say a few words about humanity's apparent moral progress, and how we might continue to make moral progress.

Subjective Morality
Suppose there is no universal morality, and morality is just a matter of personal taste. In that case, when Mary does what she thinks is right, Paul may see her actions as wrong or evil. The question is, what does this say about Paul's ability to hold Mary accountable for her actions? If Paul is a physicalist, then he knows that his moral preference has no more universal significance that Mary's moral preference. Does that imply that he should allow Mary to do whatever she wants? Plainly it does not. This is a contest of wills, not a contest of oughts. Paul must decide whether he is willing to act to stop Mary from taking her actions or not. Likewise, Mary must decide whether to forego her actions in light of Paul's disapproval and potential retribution.

Now imagine the same scenario with many more people in the mix. What emerges is a social contract. The greater the commonality of the moral preference, the greater its reflection in the contract and the greater its enforcement. The social contract may contain clauses that Mary disagrees with, but Mary benefits overall from being a signatory. For example, Mary is protected from car theft by the social contract, and that may be more important to her than the loss of her right to smoke on an airplane.

Then again, Mary might not be inclined to obey certain clauses in the contract, nor submit to the penalties for bypassing those clauses. If the overall contract is good, but Mary finds clause 1.2.56(a) objectionable, she may opt to act surreptitiously in violation of that clause. If caught, Mary may be more likely to flee social justice because it lacks personal justice in her eyes. However, Mary must weigh the consequences that Paul may similarly violate clauses that Mary holds dear. Mary may have to compromise.

This is how subjective morality works to create a social morality that appears to have some form of universality.

History teaches us that social morality changes. Slavery was once good. Racism was once good. Homophobia was once good. These things are now evils. What happened?

Two things. First, humans became less fearful and more empathic. I think that the more we saw blacks as human, the less tolerable it was for us to accept that they should be enslaved. Discrimination against homosexuals is fading as men become better educated and more secure in their sexuality, and as homosexuals are viewed as fully human.

But ratchet up the fear another level, and our morality goes out the window. Spying on Americans without safeguards? “Gee, I’m terrified by the terrorists, let’s do that.” Racial profiling? “They scare me!” Torturing people? “They might be terrorists, so I’m sure our boys and girls were justified.” Invasion of another nation without cause? “There might be terrorists there! Better safe than sorry.” Holding the families of suspects in an effort to bring in wanted men? “Well, I won’t kick up a fuss about that, because the wanted men fill me with terror.” And on it goes.

Osama is winning this thing as long as our morality crumbles in the face of his terror.

The second reason that bad "-isms" are fading is that democracy has displaced dictatorship, making the common man's choices more politically important.

I conclude that, given only subjective morality, the world would look much as it actually does, and that moral progress is possible.

Universal Morality
Keen positivists in the audience (if I have an audience that has read this far through my post) will recognize this discussion of universal morality is moot. The concept of universal morality is literally meaningless. How would you determine whether an action was universally moral?

If there is no way to know an absolute morality when you see one, then the term must be meaningless. The age-old principle applies: if you cannot devise a recipe for assigning widgets to category X, then category X is meaningless (as in, undefined).

Suppose there were some way to know that a moral code was universal. Such a recipe would still rely upon subjective moral appeal!

Imagine that there is a Federal Institute for Moral Research, and that this agency discovered that all pleasurable food consumption was universally wrong. Not that there were any universal penalties for consuming tasty food, but simply that it was wrong. All tasty food would have to be made bitter and distasteful for its consumption to be moral. Would you choose to follow such a moral code? In other words, what use are moralities if they have nothing to do with outcomes?

We have no reason to expect that universal moral laws would be appealing to humans - or so the universal moralists will gladly inform us when we object to their claimed universal code!

If moral considerations had to be blind to outcomes then few would choose to be moral for morality's sake. A moral code is nothing without an imperative to follow it.

Thus, it would seem that a universal moral law is nothing without a universal regime for compliance. Of course, moral universalists are willing to posit the idea that we are each individually punished (infinitely and disproportionately) for our transgressions - the ultimate compliance scheme. Unfortunately, the cost of inventing Heaven and Hell is an acknowledgement that morality is a function of nothing more than subjective taste and a sense of consequence. That's why Hell is exactly what we don't want, no matter what our personal tastes.

And, this brings us back full circle. Advocates of universal moral law like to argue that we must accept the existence of universal moral truths, or else the things we regard as subjectively immoral would be acceptable. This claim is contradictory because the claim acknowledges the value of subjective morality, and wrong because subjective morality leads to social contract.

Ironically, it is authoritarianism that poses the greatest threat to our subjective moral good (unless you're the authority, that is).

When to be convinced by a moral argument
Having dispelled the concept of universal morality (until such time as the dissenters voice comment), I want to say something about moral progress.

We think of the elimination of slavery, racism, sexism, torture, illiteracy, etc., as forms of moral progress. Yet, if morality is subjective, how was this progress made, and how can we continue to make moral progress?

Our social contract changed for the better when we became less fearful and more empathic. Once we became secure enough that we weren't going to lose our prosperity by losing our whites-only, aristocratic, patriarchal way of life, we let go of our bad habits.

History has a few lessons for us about moral progress. When considering whether to grant or revoke a right:
  • Purge yourself of irrational fears,

  • Empathize with those who favor and oppose the new legislation,

  • Maintain perspective by studying the scope and scale of the consequences.

Sunday, February 05, 2006

In Defense of God

I just found an eBook called In Defense of God that catalogs some of the moral inconsistency of the Bible.

Reading the book preview it becomes clear that the Bible is consistent with the writings of a primitive Bronze Age culture for whom racism, violence and sexual abuse would have been normal behavior. The authors of the Old Testament simply created God in their own image.

The author hasn't given up on theism (yet), but at least he has the sense to know that the God described in the Bible is less moral than most modern humans.

Clark in 2008

I read a recent Q&A session with former WMD inspector Scott Ritter. Scott Ritter strikes me as a bit of a loon, but he did say something interesting:
A lot of Democrats voted for Ronald Reagan. You need a candidate who motivates Republicans. You are weak on foreign policy, national security and homeland security. All you have to do is tell Americans they should be afraid. Republicans exploit that fear. Democrats need someone who can explain how to deal with their fear.
Sad, but true. American voters are fearful and irrational.

I used to think that our national institutions ensured the great success of our democracy. Now, I'm beginning to see that that success was a blend of good leadership, when we had it, and good fortune.

A large proportion of American voters are ignorant about what is going on in their country and around the world. They need to be led by someone who knows what's going on, and cares about America. Today, we lack that leadership. The Republicans have exploited American cowardice to implement policies that have devastated America and its reputation.

Wesley Clark is the strongest leader I have seen on the Democratic side. He's also a four-star General, an American hero, and a Rhodes Scholar. That should be enough to win even the American chickenry vote.

Check out the Q&A session with Wesley Clark at the end of his Real State of the Union address. He speaks like a president should.

If these reasons aren't enough to win votes, then the Democrats must start playing by the rules of the Republicans: treat Americans like cowardly simpletons, and exploit their fears. If we're going to elect a leader using an irrational process, we should at least end up with someone who'll move the country in the right direction. The stakes are too high for us to lose again.