OF RATS AND SUPEREMPOWERMENT

Here's a big think brief to get your mind working:

Artificial intelligence is not made from scratch in a lab, rather, it's a bootstrap that leverages what has already been developed in the biological world. Existing, biological intelligence is first modeled and then replicated in hardware/software combos (as we see in the picture to the right, where the Ecole Polytechnique Federale De Lausanne has produced a model of the upper levels of a rat's brain). This process of development gets really interesting when we exceed the computational equivalence of different strata of biological intelligence within the confines of an inexpensive, standard commercial chip. We recently passed the level of intelligence seen with insects, and the knock-on effects have been a plethora of UAV (unmanned aerial vehicles) drones, robots (check out this clip of a robot that uses insect intelligence to walk), and intelligent software agents. It will get even more interesting we pass to a mammal's level of intelligence, for example the rat, over the next few years.

Over the next decade, the growth rate of artificial intelligence will have the following impacts on warfare (there's LOTS more, but I'll leave that for my next book if I can find a way to make it financially worth while to write):

• Ubiquity. As the cost of an insects level of intelligence drops to a pittance (less than 5 years), it will make its way into almost every product imaginable. In a fashion similar to a queen bee, every person will own thousands of tiny workers that can do a myriad of tasks (many through the miracle of self-replication via software). For the Hamas guerrilla firing a low cost rocket into Israel, the cost of adding a guidance system that allows it to hit a specific window or electricity substation becomes trivial. For the RBN computer hacker, each "bot" in his swarm will be able to run rule sets that will make today's Storm network look trivially simple in comparison.
• Speed. The rate at which artificial intelligence will be made available to the average global consumer isn't a straight line. It's exponential (see Kurzweil's inset graph). This means that governments and corporations will quickly find that the delta between what is only possible in a lab and what is available in the mass market will shrink. In time, particularly for the government given its slow process of adoption, these organizations will find it's less about getting ahead of the technology curve, but rather keeping up. This lag will be exacerbated by the way in which tinkering networks of amateur inventors sprawl over the technology and apply these capabilities into new niches much faster than bigger organizations.
• Power. The improvement in the intelligence available to the individual will quickly reach a mammal's level (ie. rat) in a few years and human level equivalents by the end of the next decade. Shortly after that, as the cost of such capabilities reaches pennies, it's likely that vast majority (99.9999%) of intelligent users of the Internet will be some form artificial intelligence and not biological beings. The same level of saturation will be true for nearly every physical object we make, buy and sell. At that point, things will get really interesting.

Comments

[Duncan Kinder]

John, your argument rests upon the assumption that Moore's law will continue to operate.

But, it seems, for Moore's law to operate, the computer chip manufacturers and others in the computer industry would have to continue to enjoy the relatively tranquil environment it has for the past generation or so.

It is difficult to visualize how computer labs could continue to make progress when subjected to frequent and increasingly severe and sophisticated direct attacks.

In short, are you not supposing that global guerrillas will never turn their guns on the computer industry itself? Why should they remain secure while all else is being disrupted?

And it seems to me that a swarm of insects ( a plague of locusts? ) might very well impair Intel, Microsoft, and IBM. Which would abort their potential to evolve into the rat stage.

And then we'll have to revert to smashing one another over the head with tomahawks or something like that.

[Duncan Kinder]

To follow through on my "plague of locusts" point, are you not suggesting that future warfare will resemble Moses vs. the Pharaoh in Egypt?

[John Robb]

Duncan, self-assembly of circuits may be the next major step in the long term progression of fabs. That means very small, high error rate tolerant fabs. Intel turns into a producer of intellectual property (circuit design) rather than hardware.

[mittelwerk]

kurweil's curve is made of paradigm shifts. computers are still just turing machines: the per-unit cost of circuits can drop to zero and it won't mean a thing as long they depend on code, which is to say, qualified human labor. the program to run your smart condom would be massive and buggy as hell -- entailing the need for large organizations to develop, maintain, and market them, as well as steady investment streams, none of which is very futuristic; the program to run a weapons system that identified enemy combatants by pheromone would require a stadium full of programmers working around the clock....

true smart computers that think like a rat are probably going to smell like a rat -- they'll be the product of some sort of biocomputing, some way to give a machine the incentive information of a living thing.

[infopollen]

Mr. Robb, have you read Manuel DeLanda's "War In The Age of Intellegent Machines?" If so, I'd like to hear your thoughts. If not, you might want to pick up a copy. Cheers.

[Fabius Maximus]

Very thought-provoking!
.
Of course, exactly *when* these things happen is difficult to say. The history of AI is one of exciting "summers" followed by "winters" in which lack of progress results in collapse of funding. At the 1987 conference at Stamford of the Amer Assn for AI I saw prototype "expert systems" that would soon change the world. Or, as it happened, not.
.
Like biological evolution, AI seems to advance in punctuated equilibrium, where rapid progress follows periods of seeming stasis.
.
See this Wikipedia on AI for a nice summary:
http://en.wikipedia.org/wiki/History_of_artificial_intelligence#The_rise_of_expert_systems

[Fabius Maximus]

Just a follow-up, on how far machines must advance to equal us. This is really crude, and 8 years old, but FWIW (does anyone have more current data?):
.
"The human memory capacity can be estimated at 1011 neurons *104 synapses per neuron*4 bits per synapse = 4*1015 bits of memory. ... If this rate is extrapolated, then humanity will have a machine of human memory capacity by, roughly, the year 2010, that is, a single human generation from now. ...

"Waltz attempts to estimate how long it will be before computers have the same computing capacity as the human brain, meaning that these new machines will be able to process as many bits per second as the brain. To make this estimation, he needs an estimate of the processing capacity of the brain. He takes the following figures from the neurophysiologists.

"There are approximately a hundred billion neurons in the brain, each of which is linked to roughly ten thousand others; hence, there are ten to the power 15 neuronal connections, or synapses as they are called. Each neuron fires roughly 10 times per second.

"Let us assume that the information content of each synapse (that is, the strength of the connection) is 4 bits. Thus, the total bit-processing rate of the brain is roughly 4 times 10 to the power 16 bits per second.

"Waltz compares this figure with a similar figure for the performance of the Connection Machine (Hillis 1985). The Connection Machine is massively parallel, with 65,536 separate processors that can function simultaneously; and each processor is able to communicate with any other through an interprocessor communications network.

"If each processor is connected to 10,000 others and sends a 32-bit message in each message cycle, 170 such cycles can be executed per second. A Connection Machine costs about $4 million; so, for $20 million (the upper limit of what people are prepared to pay for a single upercomputer), five such machines can be bought.

"This results in a bit-processing rate of 6.5*104*104*32*170*5 = 2*1013 bits per second, which is a factor of about 2000 short of the human figure."

"What If AI Succeeds? The Rise of the Twenty-First Century Artilect", Hugo de Garis, AI Magazine (Summer 1989)
http://www.aaai.org/ojs/index.php/aimagazine/article/view/741/659

[Walter]

Kurzweil began publishing his pretty hockey-stick graphs and accompanying predictions of imminent artificial intelligence about 10 years ago. In the past 10 years there has not been one iota of progress toward AI. I realize that's a strong statement, and I will concede that there may have been *half* an iota of progress, but really we're no closer to having AI now than we were before these past 10 years of doubling CPU power.

Increasing CPU power does almost nothing to get us to AI -- if we suddenly had CPUs 10,000 times more powerful than our current dual-core Xeons, it would mean *nothing*, AI-wise. We don't have the slightest idea how to build an AI, and an AI isn't going to spontaneously emerge just because CPUs are faster, not soon, not ever.

I don't know why anyone takes Kurzweil's fantasies seriously.

[John Robb]

Walter, a couple of points. The entire AI from scratch in a lab is fantasy as you point out.

Instead its going to be a bootstrap that builds on existing biological capability (nothing gets thrown out). That wasn't possible until recently (due to CPU strength being under the required threshold to replicate that capability). I don't see anything in Kurzweil's writing or his graphics (i.e. the "hockey sticks" across a variety of fields) that don't fit that process of development perfectly.

Also, AI shouldn't be a synonym for human intelligence. Within the context of UAV development, for example, artificial insect or bird intelligence would be extremely useful. This tracks, since we are doing this today.

[Fabius Maximus]

As a follow-up and support to Robb's comments: efforts to build an "AI" (in the popular sense) effectively ceased during the AI Winter of 1987-93. Since then there has been progress on narrow aspects of "intelligence." Often physical "intelligence", like seeing and walking. Hence the rise of what Robb describes as "insect" intelligence. It is a different approach, with great promise.
.
Also, AI is often defined in terms of the big dream -- and useful results from AI research become just "technology" -- not AI (e.g., speech recognition). In this sense no no progress is AI is possible until HAL 9000 (or Skynet) is activated.

[stevenaustin6]

FM, would you be so kind to elaborate on the AI Winter? I was doing some background research on a specific AI topic, and noticed that the line of research abruptly ended in 1989. To your knowledge, what happened? And what heralded the end or at least mitigation of the winter in 1993? I couldn't find reference to that year in my wiki-oogling. (I have to imagine that the *onset* of the winter was somehow Cold War related, but it could very well be just coincidence.)

[Fabius Maximus]

The Wikipedia on AI History describes it better than I can. Here is my -- purely non-expert and subjective view about this...

My experience with AI was as one of the guys who do not understand but writes the checks. That gave a wonderful view of things at tech Conferences (like watching the Super Bowl from a luxury box), but (like a dog going hunting with his owner) it gives experience with little enlightenment.

An AI Winter followed the failed promises of expert systems. This boom-bust followed the general pattern of my experience with high-tech projects. I have green-lighted projects, with a terrible success rate (probably no worse than par), getting applause -- and no career consequences following their failure. I have cancelled projects, always a painful task and never with positive career results.

Wonder projects (how they get that status is a different discussion) continue until absolute and clear failure, which creates the boom-bust cycle.

For more on this, from better sources, I suggest skimming the archives of AI Magazine (which I consider one of the best of its type, often tought-provoking).

[Marcello]

"Also, AI shouldn't be a synonym for human intelligence. Within the context of UAV development, for example, artificial insect or bird intelligence would be extremely useful."

I strongly agree on this. A level of intelligence comparable with that of insects or a fraction of mammal level of intelligence would have substantial benefits. You do not need a HAL 9000 level of AI to have a self driving resupply truck (something which is actively and seemingly succesfully worked on as we speak) or an autonomous drone looking after the safety of an oil pipeline.
What I strongly disagree with is that this favors parties like Hezbollah. Because while it is all well and good that some guy in Baghdad or Beirut will be able to make some nasty bots in his workshop, the guys on the other side will have automated factories churning them out 24/7. Given that the US economy is approximatively 130
times larger than that of Iraq and the israeli one five times that of Lebanon that is not the sort of warfare you want to get into if you are a guerilla .
It seems to me quite intuitive that increasing the level of mechanization (or more properly in this case, roboticization) of warfare will favor the party with more resources.
One of the biggest weaknesses of the West in the sort of wars being fought now is the human element, in the form of the unwillingness to take casualties, the high cost of training and retaining soldiers and so on. If that is eliminated or reduced the fact that guerrillas will have a couple of additional toys to play with will be a small consolation compared to the strategic change against them.

[Duncan Kinder]

the guys on the other side will have automated factories churning them out 24/7.

Assuming:

1) These factories are not being sabotaged
2) Energy prices are low enough to make these factories financially reasonable.
3) These factories have not been outsourced to regions sympathetic to Global Guerrillas, where they can be co-opted or more easily disrupted.

The guys in the Ohio Rustbelt do seem quite willing to run factories such as you describe, but there aren't a whole lot of factories there anymore.

[Marcello]

"1) These factories are not being sabotaged"

How many sabotage campaigns have you heard about at Lockeed Martin plants or Ingalls or whatever?

"2) Energy prices are low enough to make these factories financially reasonable."

It is relative. If energy prices are high for the factory in the US so they are for the little guy in Baghdad. But the guys in the USA have lots more money to spend.

"3) These factories have not been outsourced to regions sympathetic to Global Guerrillas, where they can be co-opted or more easily disrupted."

The driving force behind outsourcing factories is reducing human labor costs. Guess what happens to that when you get rid of the pesky humans?

[Duncan Kinder]

John:

A basic problem with your thesis is that it is unduly complicated.

If I intended to attack you with Killer Bees or Rabid Rats, would it not make more sense for me to genetically modify the plain vanilla carbon-based bees or rats that we already have than to spend time and money on this exotic, silicon-based venture.

Also, it intuitively strikes me that it would be more feasible for me to breed my rats while hidden in some cave somewhere ( or perhaps in the dungeon of some castle in Bavaria ) than to engineer the gizmos you are describing.

[pm2075]

"A basic problem with your thesis is that it is unduly complicated."

Let me assist. Just think outside the box Duncan. Simplicity helps a great deal as well before you jump. Business is war or rather war is a business that "time" and "money." High reward/return for scalability and/or flexibility of operation with silicon based vs. carbon similars. To add reliability, maneuverability, you can say you are able take them anywhere at a moments notice time and not any sleep.

"Slow process of adoption," that's where the misconception lies. They aren't. U.S. military-DoD. The example of trickle down technologies are numerous. They've been the driving force and will be. AI-think cold war. Think 'War Games.'

[Duncan Kinder]

I'm sorry, but I still don't see why my genetically modified bumblebees can't do the job.

[Duncan Kinder]

BTW: you might want to consider renting a copy of Alfred Hitchcock's The Birds.

[USMC8652]

The power of AI in terms of application by GG's does not strike me as an effective tool used directly against the developed nations (this is more of a WMD approach). Rather, this "supercomputing swarm" seems more effective if used indirectly, or levered against us just as UBL used our own air traffic control, fear of direct confrontation, and willingness to clump together in large cities to work and play.

The GG's will wait until we become more dependent upon these super-complex, ubiquitous systems for our daily lives (vital even) before exploiting some undetermined weakness. Remember the Power Law!

[Kim McD]

There is a fine thinker who is articulating this in a careful and complete way. Wolfgang Giegerich's book is called Technology and the Soul: From the Nuclear Bomb to the World Wide Web. It is coldly clear in its understanding of our place in this particular mutational sequence. Surprisingly, neither the GGs nor the bureaucracies, corporate or state, are in charge of this. It is a long trajectory. We are all in the camp. Let our lights shine and notice another's, appropriately.