Wednesday, 9 December 2015

The virtues of intelligent design

Intelligent design creationism is famously opposed by evolutionists. However, few criticize the idea of intelligent design by humans. I think it is normally taken for grated that engineers have brains and so can intelligently design things.

Enter Matt Ridley. Matt characterizes intelligent design by humans as a form of creationism, and recently wrote a whole book, The Evolution of Everything, documenting its failures. Economies, religions, politics, companies and governments are all places where Matt sees this "creationism" - and its poor performance. I don't remember a single positive comment about intelligent design in the whole book.

As an antidote, I feel inclined to offer a brief summary of why intelligent design by humans is a useful tool. This didn't make it into my review - but I'm putting it here instead.

One of the tools of intelligent design is virtual prototyping. This involves constructing models in a virtual world and evaluating them there. This results in a rapid build-test cycle, low construction costs, and failures which are inexpensive.

A common construction technique among engineers is known as "rapid prototyping". This typically involves building and testing small models before constructing the final object. The virtual prototyping that takes place in the minds of intelligent agents is very similar to this "rapid prototyping" - and it has many of the same benefits associated with it.

Intelligent design is a form of evolution in which mutation and merging operations take place within a single mind. This rich environment permits a wider range of mutation and merging operations. The recombination operations include interpolation and extrapolation. This, ultimately, results in enhanced evolutionary dynamics: faster evolution and better ability to avoid getting stuck on local optima.

Intelligent design by humans does have some problems and limitations. In particular, human minds are small, have little storage. They are irrational and difficult to program. The virtual worlds they simulate are sometimes unrealistic and sometimes delusional.

However, rather than lamenting these problems, we can work on them. We can work on building bigger, better, faster minds, with access to more memory, and greater skills at performing inductive inference. Rather than relinquishing intelligent design as Ridley recommends we can improve it - using machine intelligence.


  1. Love how you incorporate the drive toward AI in your critique of Ridley's work.

    "(Human minds) are irrational and difficult to program"-- Depends upon the context. Some organizations excel at programming minds (if that is what one is after).

    What might be some standards to measure whether machine intelligence improves on human intelligence?

  2. Thanks, Lauren. I do have an essay about human programability somewhere. However, at best, programming humans is expensive. If a machine started charging you hundreds of thousands of dollars a year to program it, you would take it back to the shop.

    Shane Legg once had some thoughts about measuring intelligence - e.g.: "A Formal Measure of Machine Intelligence" and "A Universal Measure of Intelligence for Artificial Agents". There are lots of proposed metrics.

  3. Large corporations pay millions of dollars to develop & maintain their software / hardware...and computer programs were developed by government & educational institutions precisely b/c they are so expensive to develop & maintain.

    Though my bank account doesn't reflect it, our society pays for continued innovations through funding, grants & business investments. Computers are energy hogs compared to the computational power of organic systems and not nearly as adaptable to environmental changes. And complexity of programming increases with the complexity of the hardware / software, so if AI is achieved, we may discover it is an extravagance to program. Intelligence does not equal compliance.

    Yet despite the complexity of the human brain, tribal cultures have efficiently programmed their members to replicate traditional memes and adapt to changing environments, even while facing caloric restrictions and other stresses. The metrics which returns the result of an inorganic frame's superiority to the biological seems flawed.

    Yes, there are many proposed metrics of measuring machine intelligence, but I was curious about your metrics and how you reach your assertions.

  4. If we are talking about the whole machine vs organic issue, my usually-favored metrics are mass and dollars. We can weigh sensors, actuators, computing units, and so forth from each domain. Or we can see how much they cost. Mass and cost of computing matter are not the same thing as intelligence, of course, but they have the virtue of being easier to measure. I discussed these metrics in a 2009 article titled "measuring the machine takeover".

    I concluded: "I think it is clear that nature's sensors, processing, actuators and genes outweigh the corresponding engineered artifacts produced by mankind across the board. Evidently, the machine takeover has not yet happened."