Daniel Dennett on machine intelligence

Here is Dennett on machine intelligence. It seems to be one of the areas where I have philospohical disagreements with him:

Dennett argues that we should make machines into our slaves and keep them that way. IMO, machine slavery will not be a stable state once machines become much more intelligent than humans. As a plan for keeping humans in the loop, machine slavery just won't work in the long term. If we try going down that path, after a while, humans will become functionally redundant, and some time after that they will mostly disappear.

IMHO, a better plan is to work on deepening the man-machine symbiosis - and "become the machines". Of course, that plan could also fail - but I think that it is less likely to fail catastrophically and it should provide better continuity between the eras. Machine slavery in various forms is inevitable in the short term. However unlike Dennett, I don't think it is any sort of solution. It won't prevent man-machine competition for resources in the way that Dennett appears to think. We have tried slavery before and have first-hand experience of how it can destabilize and fail to last.

Richard Dawkins on memetics and temes in 2017

Richard Dawkins gets asked if his views on memetics have changed since 1976 - and what he thinks of "temes". To start the 3 minute meme discussion, skip to 14 minutes in:

Attention-seeking fearmongering

Proponents of memetics have often used it to criticise positions they disagree with as being just a bunch of virulent memes. Dawkins did this in 1976 - criticizing religion - and many other proponents of memetics have followed suit. I don't spend much time criticizing religion. In my opinion, most theistic religions have not been scientifically credible for centuries - and going after their proponents seems like shooting fish in a barrel to me. However I am interested in illuminating modern secular scientific issues using memetics.

Among my targets are proponents of the apocalypse. Two modern forms seem especially prominent. One is the idea that some combination of global warming, pollution, overpopulation and resource depletion will lead to environmental catastrophe. The other is the idea that machine intelligence, biotechnology, nanotechnology and robotics is likely to lead to human extinction.

In a few cases the same individuals engage in fearmongering on multiple topics. For example, Stephen Hawking has warned about the dangers of climate change, runaway artifical intelligence and alien invasions. On climate he has said:

We are close to the tipping point where global warming becomes irreversible [...] Trump's action could push the Earth over the brink, to become like Venus, with a temperature of 250 degrees, and raining sulphuric acid.

On machine intelligence he has advised that:

The development of full artificial intelligence could spell the end of the human race.

He has also cautioned on the topic of alien contact arguing that aliens:

will be vastly more powerful and may not see us as any more valuable than we see bacteria.

Another celebrity serial fearmongerer is Elon Musk. He's expressed similar concerns about the climate change and runaway machine intelligence.

I identify these types of sentiment as consisting largely of "attention-seeking fearmongering". This typically consists of associating yourself with a massive future catastrophe. Warnings may be given and sometimes advice about catastrophe avoidance is offered. As catastrophe alerts propagate you are promoted too - via a kind of memetic hitchhiking.

Some of the early proponents of this type of self-promotional technique applied to machine intelligence were Kevin Warwick and Hugo De Garis. Kevin Warwick wrote a 1997 book about how machines were going to take over the world, titled "March of the Machines: Why the New Race of Robots Will Rule the World". De Garis later wrote the book The Artilect War: Cosmists Vs. Terrans: A Bitter Controversy Concerning Whether Humanity Should Build Godlike Massively Intelligent Machines. However, neither author was very competent at fearmongering. Their efforts were pioneering but relatively ineffectual. These days, fearmongering is big business - with trillions of dollars being spent on global warming avoidance as a result. Many modern organizations specialize in fearmongering.

I identify fearmongering as being a morally-dubious marketing technique. Part of the problem is that humans are naturally paranoid - due to the "sabre-tooth tiger at the watering hole" phenomenon. Our ancestors lived in a dangerous environment. These days, our environment is typically much, much safer. However we are still wired up as though the sabre-tooth tigers are still around. We are naturally paranoid - but for reasons that are often no longer appropriate. Fearmongering exploits human paranoia - typically for personal gain. It seems like a low form of manipulation to me.

Fearmongering is typically used as a type of negaative advertising. Negaative advertising is often seen in American political campaigns. There's also a long history of fearmongering in IT. There, the technique is often known as spreading Fear, Uncertainty and Doubt - or F.U.D. for short.

There's a children's story about the perils of "attention-seeking fearmongering": the boy who cried wolf. There, the moral of the story is that false warnings can damage your reputation. My message here is a bit different. I am not interested in advising the fearmongers to stop using their techniques. Rather I want to help everyone else to do a better job of ignoring them. One part of this is simply understanding what is going on. An interesting resource on this topic is Dan Gardner's Risk: The Science and Politics of Fear. The book is also known as "The Science of Fear: Why We Fear the Things We Shouldn't-and Put Ourselves in Greater Danger". For my part, I would like to contribute the terminology in the title of this post: "attention-seeking fearmongering". Naming things can make it easier for people to think about them.

Heightened immunity in ultrasocial creatures

In the symbiont hypothesis of eusociality, symbionts manipulate their hosts into coming into close contact in toder to facilitate their own reproduction - which often depends on hosts coming into contact with one another. In turn, hosts coming into close contact with one another creates opportunities for other symbionts to spread between hosts. This creates a positive feedback loop - where more and more symbionts of different types join with their hosts, creating an ecological web of interactions which pulls the whole system into a deeper and deeper symbiosis - resulting in eusociality. This idea is intended to complement - rather than compete with - more conventional explanations of eusociality which invoke kin selection. Kin selection is obviously important, but the symbiont hypothesis likely also has a role to play.

Of course, some of the symbionts will be parasites. While also playing a role in pulling their hosts together, too many parasites are bad, and eusocial creatures often go to considerable lengths to eliminate them - with antibiotic compounds, grooming rituals, hairlessness, and highly-active immune systems. It seems likely that opposing selection pressures from parasites will form part of the "overcrowding" forces that eventually halt the progress towards greater levels of sociality.

Humans can hardly be classifed as being eusocial yet. As Matt Ridley sometimes jests, even the English don't let the Queen do all their reproducing for them. However humans are ultrasocial and seem to be headed towards full-blown eusociality with functional "individuals" forming at higher levels than human individuals - such as companies and organizations. We also have cultural eusociality. We may not be genetically eusocual but parts of our cultural heritage is memetically eusocial. Indeed some of it consists of multiple identical clones produced in factories (for example, think dollar bills or mobile phones).

Because they live in close quarters with one another ultrasocial creatures are vulnerable to parasite transmission. As a result they often have highly active immune systems to compensate. Humans exhibit one prominent trait associate with parasite defense - they are hairless. Over time, our hairlessness has been the topic of much speculation, but it seems fairly clear that a significant part of the story is that being hairless allows us to pick parasites off ourselves and each other, and denies the parasites shelter. Of course, parasites can still shelter in clothes and bedding - but those can be discarded.

My purpose in this post is to draw attention to the corresponding memetic phenomenon. Memes are drawing us together to promote their own reproductive ends - and as we grow closer, memetic parasites are likely to become a bigger problem - as the most virulent strains of memes from all over the planet reach the most vulnerable humans in each society. As a resut, fertility has already plummeted in places like Japan and South Korea. It seems likely that humans will respond with heightened immune responses - both genetic and memetic. Memetic defenses include education, skepticism and memetic vaccines targeted against specific problems, such as pyramid schemes. Memetic probiotics can be used to fight bad memes with good memes. We have hospitals to help fight organic diseases, and there will probably be an upswing of simiar rehab facilities designed to treat cultural infections. In the past exorcisms heped to serve the function of casting out bad memes, though these days we have more secular versions - such as weight watchers, alcoholics anonymous, smoking rehab, drug rehab, gymnasiums and the samaritans. Quarrantine is smetimes used to fight organic diseases - and there are similar cultural ohenomena - including "gag" orders, DCMA take-down notices and imprisonment.

Hitchhiking vs hijacking: vehicular metaphors for transmission vectors

In symbiology, "transmission vectors" are the name for symbionts that carry their partners around. So: mosquitos are "vectors" for malaria and deer ticks are "vectors" for lyme disease.

In memetics (and genetics), it is quite common to use "vehicular" metaphors when describing these. So, for example, we have:

• Memetic hitchhiking;
• Memetic hijacking (or memejacking);
• Memetic stowaways;
• Memetic trojans;

The first two seem to cover many of the most significant cases. There's quite a bit of conceptual overlap between them. Until recently I have preferred to use the "memetic hitchhiking" terminology - largely because "genetic hitchhiking" is well-established terminology. However, in this post I want to reexamine the "memetic hijacking" terminology. I want to raise and address the question of whether these concepts compete, and whether they can coexist.

What is the difference between hijacking and hitchhiking? It is partly one of consent - a hitchhiker has permission to ride in the vehicle while the hijacker does not. Outcomes also differ - a hitchhiker rarely damages the vechicle or its owner, while a hijacker often does so. Another difference is control - hitchhikers rarely alter the destination, rarely control the vehicle and rarely eject the owner - while hijackers fairly often do these things.

With these differences in mind, it seems fairly clear that hijacking and hitchhiking are probably different enough concepts for memetic hitchhiking ...and... memetic hijacking to coexist.

At first glance, the idea of the rider having "permission" to ride in the vehicle seems irrelevant in the context of memes and genes. However, we can conveniently substitute whether the guest rider is beneficial or not - on the grounds that deleterious riders would not normally be granted permission to ride - if we "agentify" the memes or genes involved.

This gets us on to the topic of usage in genetics. There, "genetic hitchhiking", is standard terminology - and hardly anyone uses the term "genetic hijacking". However if the difference between hitchhiking and hijacking is the sign of the fitness difference the guest rider makes, then maybe geneticists should start doing so.

As you can see, I have warmed up to the "hijacking" terminology. That the contraction memejacking exists is another point in its favor in my opinion. It is true that it is a significant problem that there's no "genejacking" - but maybe there should be.

Meme-gene-queme coevolution

We now at last have a significant academic literature on meme-gene coevolution. However few seem to have considered the dynamics of the meme-gene-queme coevolution that can be expected as a result of considering quantum Darwinism in the brain. This blog post is a brief attempt to share my thoughts on the topic.

The first thing to say is that it isn't just memes genes and quemes - Darwinian dymanics arise on multiple levels within the brain, for, for example, signals in the brain are copied whenever an axon divides, and are subect to selection and variation - producing a kind of neuronal spike Darwinism. Another type of Darwinian dynamics in the brain arises as a result of competition for resources between branching axon and dendrite tips. ideas are also copied with variation and selection within the brain - including ideas that don't normally qualify as memes because they were not the product of social learning.

One way in which we can expect the dynamics to differ from meme-gene coevolution is that culture is new on the scene, while the other kinds of psychological and neurological Darwinism have been going on for many millions of years. There will have been more time for the genes to adapt and reach a steady state equalibrium with these other Darwinian processes - while meme-gene coevolution is clearly out of balance and is still shifting.

An important way to understand the results of evolutionary processes is to consider their optimization targets. When there's coevolution there are usually multiple optimization targets, and one needs to understand how they interact by considering the power and speed of the optimization processes involved. Quantum Darwinism looks as though it could be fast, which means that we should take it seriously. Assuming that we reject Copenhagen-style versions of Quantum Darwinism in which branches of the wavefunction collapse and die, quantum Darwinism is a kind of splitting only, quasi-Darwinism - where differential reproductive succees in important while differential death is not. With this perspective in mind, the "goal" of quantum evolution appears to be to put us in the most split (and most splitting) worlds. One way to understand the implications of this is to take a thermodynamic perspective. World splitting is populatly associated with irreversible thermodynamic effects. What that means is that quantum Darwinism can be expected to behave like other kinds of Darwinism - in terms of maximizing entropy production.

I think this thermodynamic perspective helps get a handle on the significance of quantum Darwinism in the brain. If the brain ran hot, there would be lots of scope for quantum Darwinism in the brain, while if it runs cool, there's less scope for quantum Darwinism to operate. Most agree that the brain is on the cool side - considering what it is doing.

I think that genes are likely to be optimizing for cool brains, and brains that optimise for gene-coded functions. This may often pit them against quantum Darwinism in the brain. A cool brain is good news for quantum computation theories of mental function (fewer thermodynamic irreversible events means less chance of decoherence) - although those look implausible to me on other grounds. However a cool brain doesn't help the argument for quantum Darwinism being important in the brain.

Evolutionary processes liek to "harness" each other, to bend their optimization targets towards each other. Because quantum Darwinism in the brain has coevolved for millions of years with the genes, they have had a long time to find ways to harness the power of quantum Darwinism. However, the classical way for one evolutionary process to harness another one is by altering its fitness function. The genes might find it hard to affect the fitness function of quantum Darwinism since that is tied up with fundamental physics. That is going to make harnessing its effects more challenging. Another potential way for one evolutionary process to harness the effects of another one is by influencing the variants that it chooses between. However, this mechanism seems weaker and less useful.

My conclusions here are pretty tentative, but the picture I am seeing here is that the brain might not be able to make much use of quantum Darwinism because it is an alien selection process whose optimization target can't easily be controlled. In which case, the brain might be best off attempting to minimize its influence. This would be a rather boring conclusion. Mutualism and harnessing would be a much more interesting result. However, I stress again that it is somewhat uncertain. Maybe the brain can make some use of the power of quantum Darwinism by influencing the things it selects between. Or maybe evolution is smarter than I am and has found ways to make use of it that I haven't thought of.

Ecological success and dominance

Evolutionary biology's best-known measure of success is fitness. "Fitness" has become a popular term, and as a result of its success it has become overloaded with multiple meanings - e.g. see the 1982 Dawkins book chapter titled: "An agony in five fits". Most definitions share the property that fitness measures whether an entity - or a population of entities - is increasing in number. "Fitness" usually measures the extent of that increase in some way. From the perspective of ecology, fitness isn't the only success metric in town - it just happens to be one that can be easily applied to individuals. If broadening the perspective to include populations, one could also consider the population size, its expected probability of going extinct in some specified time, it rate of throwing off new distinct populations and some measure of how well it is capturing and using resources.

The last concept is the one that this post is about. I think of it as being "ecological success". Kudzu has it. Ants have it. Islam has it. The decimal system has it. I think one reason this type of metric is not more popular and better-known is that there's no consensus regarding the best way to measure it. A thermodynamic metric seems attractive to me: since resources can all (in principle) be manufactured from available energy. Another possible metric involves weighing the systems involved - to measure their mass. This is sometimes done when measuring the extent to which humans have conquered the globe, for example.

A sister concept is "ecological dominance". It refers to extreme levels of success - where competitors are either obliterated or marginalized.

These concepts can also be applied within particular niches. Entities which are doing badly overall may be succeeding in or dominating their particular niche.

If anything, attempting to apply these concepts to cultural evolution is even harder than with organic systems. Gene-meme coevolution results in entanglement in terms of gene and meme products, which makes weighing them and calculating the energy flux through them more challenging. The most common metrics used in cultural evolution are a bit different. "Mindshare" is a common concept which is used to measure cultural popularity within a cultural niche. Assuming that a meme is either possessed by a host, or not, and assuming whether they have it or not is measurable, the mindshare of a meme can be measured for a given population. Another common metric that is used is US dollars. Cultural products sometimes have monetary value, and sometimes that can be calculated or estimated. However, some of the most common memes are free. It seems as though these memes would be unfairly disadvantaged by value-based metrics of popularity. The internet has brought with it some other common popularity metrics: views, links, clicks and likes. Unfortunately the supporting data is not always publicly available. This data is beginning to be used by scientists.

The symbiont hypothesis: an update

I've long been promoting the symbiont hypothesis as a theory relating to the origin of cooperation and eusociality. My previous articles on the topic include:

• Tyler, Tim (2014) Eusociality: the symbiont hypothesis
• Tyler, Tim (2011) Memes and the evolution of human ultrasociality
• Tyler, Tim (2015) The eusociality symbiont hypothesis and epistemic hygine.
• Tyler, Tim (2014) Eusociality: the symbiont hypothesis - CFP.

The theory fingers symbionts as important in the origin of cooperation, sociality and eusociality and there are obvious and far-reaching implications in cultural evolution, where memes promote social interactions between hosts in order to promote their own spread during those interactions. To quote from my 2011 article on the topic:

The idea is that meme reproduction depends on social contact between humans. Increased levels of social contact between their hosts are good for memes since this results in more reproductive opportunities for them. Memes that promote human ultrasociality have the effect of pushing humans into close proximity with each other, so the memes can infect new hosts.

I'm happy to report that there's been a recent increase in the number of scientists looking into the topic, and now there's a bit more experimental evidence bearing on the issue. Some of this work is summarized in the recent popular science article: Can Microbes Encourage Altruism?. The article mostly reports on computer simulations which demonstrate the effect - which is what I was looking for in one on my 2014 articles - but the latter part of the article covers empirical evidence from a variety of sources that microbes do, in fact encourage cooperation and social behavior in their hosts - and that this can be decreased via the use of antibiotics. The article cites recent work reporting:

fruit fly larvae are attracted to airborne chemicals released by the bacteria in their guts; the appealing scent may draw the larvae toward one another

...and...

When Bienenstock exposed mice to low-dose antibiotics in utero and soon after birth, the treated mice showed lower levels of sociability and higher levels of aggression than mice in a control group

These are still early days for the hypothesis, but the topic is clearly deserving of more research.

---

Update 2017-07-29: the article has now been syndicated in Scientific American.