It is based on a paper we are writing together that analyses how much harder the Fermi question (because it is not really a paradox, just a question with answers we tend to dislike/disagree on) becomes once you take modern ideas about self replication and exploratory engineering into account. The main finding is that intergalactic expansion is likely doable using local resources and a very high branching factor, and that makes the solar neighbourhood accessible to at least millions of times more potential alien civilizations. So either alien civilizations have to be even rarer than we think, they have to approach some non-visible behavioural attractor with very high fidelity, or they are here and hiding efficiently (in this case likely because the first expanding civilization used its probes to enforce some set of rules for everybody else).

My friends who happen to be members of the Enceladus Protection Society will be happy to know that no moons of Saturn were harmed in this analysis.

In today's Svenska Dagbladet one of their ideologists, Lennart Sacrédeus, argues their position: "allowing half a gender reassignment opens for a third gender" - his argument is that people desiring to change their gender to the other one is OK, but transexual males that can be pregnant are a third gender.

The whole argument is of course based on the assumption that there only can exist two *true* genders. A female wanting to become a male that can be pregnant is not truly desiring to be a real male, and hence the desire is for something inconsistent or wrong (and societal support for the transition should presumably not be given).

There are several problems here.

Even a cursory look at the psychological and anthropological literature shows that genders are much more complex out there in reality than simple gender binaries. From a religious perspective it might make sense to argue for a strict binary - it is an essential part of many religious interpretations after all. While the inside perspective from these interpretations is that deviations are wrong (morally or logically), the outside perspective is of course that the empirical evidence undermines the claims of the interpretations to represent reality (and hence undermine any moral force they have beyond encoding the local mores of different cultures at different times).

In a pluralistic society that doesn't buy a particular religious narrative about gender as anything more than (possibly) respectable point of view among others, decisions about shared rules cannot be based on just that narrative. Either it has to claim that the decisions affects people believing in it to a high degree and we should respect their rights (e.g. debates about circumcision and halal slaughter), or it has to propose general ethical or pragmatic principles that others can also largely agree with. In this case it seems unlikely that pregnant males will cause more distress among conservatively minded people than vanilla transexuals, and as far as I know nobody has mustered any convincing prudential or ethical argument why it would be a bad thing if there were more of them - in fact, there is a pretty broad consensus in Swedish society (at least the people who talk about it, and the political associations) that it is acceptable.

The growing awareness and acceptance of intersexuals show that a third gender might appear regardless of what transsexuals are allowed to do or not.

The assumption that transsexuals can only legitimately desire to become the opposite sex is problematic. What about somebody desiring to become intersexual or asexual? Leaving aside the inertia of the legal and medical system (where it will no doubt take a long time until the idea takes root), there doesn't seem to be any moral reason not to accept that desire if we accept the desires of some people to have another gender. The moral motivation for gender reassignment (besides autonomy and morphological freedom) is that it likely will increase well-being by providing a congruent body to the mental gender. If the same is true for desiring to becoming intersexual or a male with utreus, why not? This might be a very rare state of desire, but that doesn't automatically invalidate it.

By turning the question into a political question the Christian Democrats also inadvertently make gender even more of a socio-political matter and less of a metaphysical matter. If an argument against a change in the law is that third genders must be prevented, then that implies that the number of genders is a political question, open for whatever majorities and alliances that exist to decide upon.

Public opinion seems to be moving towards accepting more gender diversity and less paternalist control over reproduction: in the future we are likely to see far more blurring of gender binaries. I think that is fine: let people choose the bodies they want. And let us tolerate those choices, just like we tolerate religious choices (and for the same reasons). Toleration doesn't imply lack of critique: we should do our best to figure out how to make choices that actually improve well-being and comment on what we find - but that requires freedom to choose so we have experience to learn from. Armchair moralism is always trumped by real world experience.

I blog about the issue of whether there is a need for some mandatory international oversight of potentially dangerous biotechnology. It is a tricky issue, and my view is that while I do think it might be needed, we *really* need to figure out how such an oversight should function before we implement it.

As I mention in the post, I saw Contagion in the holidays. Typical cheerful Swedish holiday viewing. Very good and understated film, strongly recommended - although it will make you a bit more of a germophobe. It is so nice to know it was based on a real virus, merely given pandemic properties.

On 2012-01-01 12:55, Stefano Vaj wrote:
> I do think that utilitarian ethical systems can be consistent, that is
> that they need not be intrinsically contradictory, but certainly most of
> them are dramatically at odd with actual ethical traditions not to
> mention everyday intuitions of most of us.

Of course, being at odds with tradition and everyday positions doesn't tell us much about the actual validity of a position. Which brings up the interesting question of just how weird a true objective morality might be if it exists - and how hard it would be for us to approximate it.

Our intuitions typically pertain to a small domain of everyday situations, where they have been set by evolution, culture and individual experience. When we go outside this domain our intuitions often fail spectacularly (mathematics, quantum mechanics, other cultures).

A moral system typically maps situations or actions to the set {"right", "wrong"} or some value scale. It can be viewed as a function F(X) -> Y. We can imagine looking for a function F that fits the "data" of our normal intuitions.

(I am ignoring the issue of computability here: there might very well be uncomputable moral problems of various kinds. Let's for the moment assume that F is an oracle that always provides an answer.)

This is a function fitting problem and the usual issues discussed in machine learning or numerics textbooks apply.

We could select F from a very large and flexible set, allowing it to perfectly fit all our intuitive data - but at the price of overfitting: it would very rapidly diverge from anything useful just outside our everyday domain. Even inside it would be making all sorts of weird contortions in between the cases we have given it ("So drinking tea is OK, drinking coffee is OK, but mixing them is as immoral as killing people?") since it would be fluctuating wildly in order to correctly categorize all cases. Any noise in our training data like a mislabelled case would be made part of this mess - it would require our intuitions to be exactly correct and entered exactly right in order to fit morality.

We can also select F from a more restricted set, in which case the fit to our intuitions would not be perfect (the moral system would tell us that some things we normally think are OK are wrong, and vice versa) but it could have various "nice" properties. For example, it might not change wildly from case to case, avoiding the coffee-mixing problem above. This would correspond to using a function with fewer free parameters, like a low degree polynomial. This embodies an intuition many ethicists seem to have: the true moral system cannot be enormously complex. We might also want to restrict some aspects of F, like adding reasonable constraints like the axioms of formal ethics (prescriptivity ("Practice what you preach"), consistency, ends-means rationality ("To achieve an end, do the necessary means") - in this case we get the universalizability axiom for free by using a deterministic F) - these would be constraints on the shape of F.

The problem is that F will behave strangely outside our everyday domain. The strangeness will partly be due to our lack of intuitions about what it should look like out there, but partly because it is indeed getting weird and extreme - it is extrapolating local intuitions towards infinity. Consider fitting a polynomial to the sequence 1,2,1,2,1,2 - unless it is constant it will go off towards positive infinity in at least one direction. So we might also want to prescribe limiting behaviours of F. But now we are prescribing things that are far outside our own domain of experience and our intuitions are not going to give us helpful information, just bias.

Attempts at extrapolating a moral system that can give answers for any case will hence either lead to

1. Fit with our moral intuitions but severe overfitting, complexity and lack of generalization to new domains.

2. Imperfect fit with our moral intuitions and strange behaviour outside our normal domain. (the typical utilitarian case)

3. Imperfect fit with our moral intuitions and apparently reasonable behaviour outside our normal domain, but this behaviour will very likely be due to our current biases and hence invalid.
   

Not extrapolating will mean that you cannot make judgements about new situations (what does the bible say about file sharing?)

Of course, Zen might have a point:

'Master Kyogen said, "It is like a man up a tree who hangs from a branch by his mouth. His hands cannot grasp a bough, his feet cannot touch the tree. Another man comes under the tree and asks him the meaning of Bodhidharma's coming from the West. If he does not answer, he does not meet the questioner's need. If he answers, he will lose his life. At such a time, how should he answer?"'

Sometimes questions have to be un-asked. Sometimes the point of a question is not the answer.

My own take on this is exercise is that useful work can be done by looking at what constitutes reasonable restrictions on F, restrictions that are not tied to our moral intuitions but rather linked to physical constraints (F actually has to be computable in the universe, if we agree with Kant's dictum "ought implies can" ought should also imply "can figure it out"), formal constraints (like the formal ethics axioms) and perhaps other kinds of desiderata - is it reasonable to argue that moral systems have to be infinitely differentiable, for example? Can the intuition that moral systems have to be simple be expressed as a Bayesian Jeffreys-Jaynes type argument that we should give a higher prior probability to few parameter moral systems?

This might not tell us enough to determine what kind of function F to use, but it can still rule out a lot of behaviour outside our everyday domain. And it can help us figure out where our everyday intuitions have the most variance against less biased approaches: those are the sore spots where we need to investigate our moral thinking the most.

Another important aspect is for investigating AI safety. If an AI were to extrapolate away from human intuitions, what would we end up with? Are there ways of ensuring that this extrapolation hits what is right - or survivable?

We are not dealing with a very far future (just the next 15 years or so), so the challenges to identity are pretty "normal. Just the growth of online identity as a key part of our social identities, which means that we get embedded in technological identity metasystems, social spaces spanning time, space and legal borders, and that identity technology begins to automatically tie together different aspects of our identities whether we want it or not. At the same time we are developing individualized conceptions of illness and health, with the body at least partially an expression of our values and aspirations. Further on medical advances in biotechnology, life extension, and neurotechnology are going to make our lives increasingly science fictional... from our current, equally science fictional perspective.

Another talk I enjoyed today was Meredith Patterson's talk about linguistics, Turing completeness and software security. A good explanation why protocols should not be too complex.

A new study shows that the picture is more complex: James White, G David Batty, Intelligence across childhood in relation to illegal drug use in adulthood: 1970 British Cohort Study, J Epidemiol Community Health (2011). doi:10.1136/jech-2011-200252

They looked at use of illcit drugs among a cohort of 8,000 born in 1970.

When intelligence was factored in, the analysis showed that men with high IQ scores at the age of 5 were around 50% more likely to have used amphetamines, ecstasy, and several illicit drugs than those with low scores, 25 years later.

The link was even stronger among women, who were more than twice as likely to have used cannabis and cocaine as those with low IQ scores.

...

The findings held true, irrespective of anxiety/depression during adolescence, parental social class, and lifetime household income.

Why? A likely guess is that smart people are novelty seekers - they figure out things more quickly and hence get bored at school and jobs not tuned to their abilities, but they might also become smarter just because they try new things and hence have educational experiences. The paper also suggests that some might feel stigmatized by their peers. Yesterday I saw the new Sherlock Holmes film, and one can certainly see how these explanations fit into our culture.

Another interesting thing the study itself is silent on, but is worth investigating, is whether higher intelligence protects from the health effects of drug use. Smarter people generally have better health behaviours and better health (including both lower incidence of smoking and easier smoking cessation). I would conjecture that cognitive ability helps people manage their drug use, making it less likely to turn into abuse. We'll see.

It all started when I did an entry on Wikipedia for crises in dynamical systems, and then I realized I needed an illustration. The Ikeda map is pretty good for this, since it has plenty of crises.

See their SIGGRAPH 2010 slides for more detail.

This is a beautiful example of what WBE attempts to do to the brain, as well as a nice warning of just how tricky even well defined systems can be to handle. After all, this is a human-designed 2D system where we know perfectly what the components are supposed to do, everything has clean interfaces and there are just about 3,510 transistors to deal with. Still, it shows that we can reconstruct complex artifacts in a semiautomated manner. This ability will only get better over time.

Now I want to make myself a wall-poster of the real and uploaded processor.

See my previous post for a description and some comments on the talk.

Heliotropes, a poem by Brian Christian and film by Michael Langan.

There is a lovely similarity between nature and human systems. Today I explored the arteries of Berlin, the broad boulevards enabled by loose soil and occasional destruction. Cities are efficient organisms, making use of the closeness of people and businesses to avoid waste.

Take home message: tools that limit the information flow in society are probably more dangerous than tools that add more information flows. This is why if you think it is a bad thing that Oxford taxis are supposed to record conversations, you should be even more worried about bills aiming to block IP infringing websites. Mission creep, entrenching of power differentials and lack of accountability make both bad, but blocking information also prevents an open society from correcting itself.

Extraordinary extensions of state power require extraordinary safeguards. Unfortunately we seem to have become used to extensions of state power.

Eric's talk covered an important issue: how can we figure out things about technologies we have not built yet?

His answer is exploratory engineering. It has similarities both with science and engineering, but is distinct from both (just as they are distinct from each other).

Science aims at acquiring knowledge while normal engineering aims at providing products. Exploratory engineering aims at knowledge production. Normal engineering aims at manufacturing while exploratory engineering aims at valid modelling. However, this modelling is more like the modelling in engineering: reliable bounds are preferred over science's aim of exact beautiful descriptions, predictable behaviour is preferred over surprising discoveries, and having alternatives and parameters in the design is preferred over science's typical aim to exclude all alternatives to a theory.

While normal engineering produces detailed specifications for a design (since it has to be built) exploratory engineering contents itself with having a parametric model for the design (since what matters is the overall properties of the kind of design). Rather than aiming for robust products the goal is to get robust analyses. In particular, this means that in exploratory engineering one can use enormous margins in designs that would be uneconomical in a conventional engineering design since they both make the analysis easier and lowers the design/analysis costs.

The end result is a description of something that could exist and would have certain properties, not something that will exist. It is an existence proof showing what kind of things are possible. This way we can investigate what exists out there in the space of possibility, getting both an inkling about the potential of as-yet uninvented technologies. This can then be followed up with more careful modelling and strategizing, for example looking at pathways from our current capabilities towards desirable technologies. Or one can take the models as inputs to analysis of impacts - if a technology with *at least* these capabilities were around, what could be done?

An interesting problem is that while exploratory engineering can tell us plenty about the space of technological possibility, we do not have anything similar for science. We cannot know potential future sciences in the same way as we can know a potential technology, since we do not have any proper "physical theories" for the space of sciences. We know a bit about their sociology and history, but there is no way to rule out or bound an imagined science. This is bad news for investigations of issues that actually depend on the state of future science. The only way of finding out what is possible in science is to do it: unlike engineering it is opaque.

Changes in scientific knowledge is not an enormous problem for exploratory engineering - in this regard the landscape is indeed timeless. A new scientific discovery might change some of the shape of the boundary set by physical law, but since exploratory engineering anyway tends to use well-tested phenomena in a robust way this rarely impinges on them. A revolution in quantum gravity will not in the least change the performance metrics of macroscopic harmonic oscillators and hence designs relying on them will be unaffected. A surprise like superluminal communications might enable new technologies (a previously off-limits area in the diagram is now potentially accessible), but unless we have reliable examples of what can be done with it exploratory engineering cannot give us very deep insights.

Another issue is how to be "conservative". A conservative exploratory engineering design is something that has a very high likelihood of actually behaving like predicted if it was actually built in the real world. In analysing the future we are however interested in various loss functions: some impacts are good, others are bad. A useful heuristic is to assume fast technological change when looking for potential trouble, slow technological change when looking for potential benefits.

Atomically precise manufacturing (APM) is of course dear to Eric's heart. He sketched out some insights we have gained from exploratory engineering, normal engineering and current nanoscience. Overall the problem is that the different communities have little understanding of each other's needs: scientists are rewarded for discovering neat stuff but not for making it applicable, leading to a weakness of the transitional research that would lead to engineering. A stronger focus on discovering and inventing useful building-blocks would probably produce very useful results.

The problem is that the measure of being 'ethical' in the study seems more like giving strong lip service to strict deontological rules, not actually thinking of behaving in an ethical manner. A much simpler explanation might be that the study measures conscientiousness, a trait that is known to correlate with subjective well-being.

I think a proper moral system probably has to produce subjective well-being, since otherwise people could become happier by switching to something else and the system would be unstable. Some thinkers like Kant think happiness has little to do with morality (if we have complete virtue, then he thinks we deserve complete happiness, but we are unlikely to reach either of these) - in fact, being motivated to do good because it makes you happy ruins the virtue for a Kantian. But most thinkers would argue the aim of moral behaviour is to achieve happiness, either because the striving for morality makes us happy or because the good is human happiness.

But again, some (e.g. Peter Singer?) might think that we should aim for happiness for everybody, which might leave us lacking in enjoyment as we struggle to help the worst off. I would argue that if that was the true morality, then the best feasible morality would be the closest approximation that makes us happy - that would be the only stable state.

Of course, things are a bit more complicated in that we do not simply optimize happiness. In fact, we are lousy at it, and most people do not seem to make a deliberate effort towards enjoying life better despite the growing knowledge that it can be done. Happiness is perhaps primarily a means rather than a goal - for evolution, and perhaps for morality. But we better learn how to use this tool better.