Sunday, 3 September 2017

Max Tegmark's evolutionary classification scheme

I read Max Tegmark's article in Scientific American promoting his book, Life 3.0.

Tegmark proposes the following classification scheme:

In summary, we can divide the development of life into three stages, distinguished by life’s ability to design itself:

  • Life 1.0 (biological stage): evolves its hardware and software
  • Life 2.0 (cultural stage): evolves its hardware, designs much of its software
  • Life 3.0 (technological stage): designs its hardware and software
This isn't a classification scheme I have heard of before. Tegmark introduces the scheme by saying:

I find it helpful to classify life forms into three levels of sophistication: Life 1.0, 2.0 and 3.0.
My first reaction was that these categories were three of the floors in the Tower of Optimization classification scheme I proposed back in 2011.

My second reaction was that Tegmark's numbering scheme seems pseudoscientific. I named my tower floors, rather than numbering them to better allow for future insertions and deletions. However Tegmark only has three categories.

There's an existing literature on the major evolutionary transitions. To say that scientists don't agree with Tegmark's classification scheme seems like a big understatement to me. Numbering schemes seem rather premature.

In my essay, I at least cited some prior work in the field - while Tegmark doesn't seem to have any citations at all. Presumably Max Tegmark made this classification scheme up. It seems like an example of how not to perform scientific classification to me.


  1. Not read the article, but it seems to me a problem with this definition is that in the 2.0 stage, the great majority of the s/w over the great majority of the time evolved rather than was designed. This doesn't imply a lack of intelligent contributions, but these come as billions of small packets that are subject to selection. Only very recently have we tried to implement en-masse within a short period, theoretically complete systems that were mainly a product of conscious design. E.g. Marxism. Such systems typically don't go well, with reality and selection soon taking over again. This rather suggests that our designs are still very poor compared to the emergent versions. Or perhaps very unstable is a better way to say this, 'poor' having connotations of what system is 'better' or 'worse', which itself is in the eye of the beholder.

  2. The "Engines of Life" Andy West, I presume ;-) Tegmark acknowledges that his proposed categories can get a bit blurry around the edges, speaking of Life 1.1 and Life 2.1. My "Tower of Optimization" classification scheme features a broadly similar scale which runs from no intelligent design to mostly intelligent design - though it features a lot more categories. I think that the general idea of classifying progress in living systems based on the penetration of intelligent design is a reasonable one. However, I very much doubt that I will be talking about "Life 3.0" anytime soon. It's Tegmark's marketing gimmick. FWIW, I am not afraid to say 'higher' or 'lower' - despite Darwin's warning.

  3. Yup, that's one of mine. So much easier with fiction, you don't have to worry too much about being right 0: When I get around to expressing it properly (always too busy), I do have a serious question that relates to the story 'meme' from that book.

    That general idea for classification seems fine, and your tower seems like a way more scientific attempt than Tegmark's very blurry 1,2,3. However, I don't get your virtualization step, or at least not if it includes the word 'most'. It can be argued that the entire of politics is the clash of different virtualization models, of which none generally turn out to be right longer term (hence the constant need for reality corrections, which amounts back to ordinary selection). Hence, we can't *successfully* perform most virtualizations. This step should surely be later, in fact probably last. If either any naturally or artificially arising beings could successfully predict an optimum evolutionary pathway, this would be a mega feat indeed considering that life is essentially a wicked system. I doubt we even know what 'optimum' means in this context, and for sure that in itself would be a major contention. Also, you don't say with the cultural step as you do with the others, what benefits it provides from previous.

    1. Arguably, the stages I list as "Learning" and "Virtualization" could be clumped together into a "brains" category. I would defend my proposed split as a reasonable way to split agents with brains into simple RL creatures, and more sophisticed agents with the ability to run complex and detailed simulations of the surrounding world.

      As humans, we all live in world simulations, constructed by our brains. They don't have to be perfect or "optimal" to be favored by evolution. Statistical success is all that is needed. As the Numenta folk like to say, the brain is a prediction machine, constantly predicting what will happen next. To do that, it needs a world model. Such a world model is a simulated, or virtual world.

      When I say that "most" evaluations are performed under simulation, this is like saying that "most" of Deep Blue's moves are imagined moves - rather than actual moves. The claim is intended to be commonplace, rather than controversial.

      In case it helps, I have more on the evolutionary significance of virtualization in my "The virtualization of conflict" essay: