Against The Singularity

Against The Singularity

Against The Singularity

Hi, I'm Tim Tyler - and today I will be discussing my take on "the singularity".

I have noticed that number of people have adopted the habit of referring to the imminent rise of machine intelligence as "the singularity".

The use of this terminology began with a science fiction writer - but has spread out into the wider community - and now there is a singularity institute, a singularity university, singularity conferences, a singularity movie, and a singularity book.

I see a problem with this. The problem is that I think the terminology is ridiculous. There is no prescident for using such a term to describe historical events. The term is overloaded with multiple contradictory meanings, and none of them make much sense.

Historians tend to use the term "revolution" to refer to radical changes in human history. So, we have the "agricultural revolution", the "industrial revolution", the "information revolution" - and so on. In the future, we may see an "intelligence revolution" and probably a "nanotechnology revolution" not long afterwards. That kind of terminology is simple, obvious and consistent with history. The chance of any credible future historian ever referring to the events in question as "the technological singularity" seems slim.

So, what is the case for the "singularity" terminology? As far as I can see there isn't any serious scientific case. This is more something that has arisen out of science fiction.

The term singularity comes from mathematics - where it refers to a discontinuity - a point where strange or discontinuous behaviour arises. The classical example is division by zero - which leads not to another value, but to infinity.

Some proponents have concluded that the rate of progress will actually do something like that. Here is an example of that, from one of the singularity advocates :

If computing speeds double every two years, what happens when computer-based AIs are doing the research?

  • Computing speed doubles every two years.
  • Computing speed doubles every two years of work.
  • Computing speed doubles every two subjective years of work.

    Two years after Artificial Intelligences reach human equivalence, their speed doubles.
    One year later, their speed doubles again.

    Six months - three months - 1.5 months ... Singularity.

  • While machines will construct smarter machines, the proposed timescale here seems rather over optimistic. One problem is with the third of the three steps. Progress does not just depend on thinking. It also depends on the ability to sense and modify the world - in order to perform experiments - and robot capabilities are not doubling every two years. Another problem lies with the assumption that the current growth does not depend on the contributions of computers to the design of computers in a significant way. In fact, designing and optimising integrated circuits is an incredibly compute-intensive task - and the involvement in computers in the process is already enormous. Machines don't yet do everything - but already they do a lot - and their capabilities matter.

    However fast progress goes, it won't ever become "infinitely" fast. We will run into physical limits first. So, the hypothetised "singularity" will never actually be reached.

    The "singularity" term has subsequently been liberally redefined and reused. That is not necessarily bad - but it does present critics with something of a moving target. It is hard to be critical when you don't know what someone is talking about. Here is Eliezer Yudkowsky describing some of the various interpretations of the idea that have been proposed:

    [Eliezer Yudkowsky footage]

    Yudkowsky writes:

    The thing about these three logically distinct schools of Singularity thought is that, while all three core claims support each other, all three strong claims tend to contradict each other."


    Prediction horizon

    One idea is that the singularity represents a kind of "prediction horizon". Here is Ray Kurzweil on that topic:

    [Ray Kurzweil footage]

    It is true that the far future is difficult to predict. The problems here are that different things are unpredictable on different time scales. Also, far future details have always been unpredictable - and probably will be challenging to predict for a long time to come. So, this "prediction horizon" seems to recede as we approach it - and therefore does not correspond to any particular future event.

    Another problem with the idea of a "prediction horizon" is that it seems more closely analogous to an event horizon than to a singularity - and so seems like poor justification for using the latter term.

    Future breakdown

    Another idea is that our "model of the future" will break down. Humans predict the future using many different kinds of models, and not very many of those models seem likely to be "broken" by the advent of highly intelligent machines. So: is such an event actually likely to happen?

    Here's Robin Hanson on the issue:

    [Robin Hanson footage]

    My perspective is that our ability to predict the future appears to be improving over time - as our ability to collect data improves, our models improve, and we develop bigger computers. This effect seems to be more significant than the fact that some of the things that are being modelled are themselves growing more complex. This effect will tend to push the point which we can't predict beyond farther out as time passes - and so it will never arrive. Our models of the future will not "break down". Rather, they will get better and better.

    It is the same with physics. In physics, the term singularity is used to describe the way that the laws of physics as they were understood in the 1950s failed with a kind of division by zero when applied to black holes. However, rather obviously this was a flaw in those theories - and not a real physical phenomenon. The modern theories that unite relativity and quantum physics do not have such singularities.

    Particular thinkers may have found that their brain melts down when contemplating these far future events - but that is best seen as a property of their brains - and not the future events themselves. If your model breaks, then that's a problem with your model - and not some kind of an event in the future.

    Exponential growth

    What about the idea that exponential growth will lead to rapid technological change? Here is Terrence McKenna describing that:

    [Terrence McKenna footage]

    Another enthusiast for this idea is Ben Goertzel:

    [Ben Goertzel footage]

    It is likely to be true that the future will exhibit rapid change - but there appears to be nothing "singular" about exponential growth. Exponential growth is what you get when mice find a grain pile - there just isn't anything "singular" about it.

    Intelligence explosion

    What about the idea that there will be an intelligence explosion - and that self-improving systems will rapidly explode - and take over the world?

    I think that the "intelligence explosion" concept is a reasonable one - and note that it already has a perfectly good name. However, if you look at the effect, it is one which is already going on. Machines are already dramatically augmenting human intelligence - and have been doing so for some time. Most machine code in the world is now written by computers. Much of the world's refactoring is also done by computers. Companies are self-improving intelligent systems. The intelligence explosion is not really an event localised in the future, but is much more spread out - and we can clearly see the beginnings of it happening now. I've made another video about this, entitled: The intelligence explosion is happening now.

    The proposed "singular" aspect apparently involves the intelligence of computers suddenly reaching human-level, and then taking over from there. The problem here is that there is no such thing as "human-level" intelligence. Workers have varying levels of intelligence required for their jobs - and consequently automation takes their jobs gradually. Already we have automated bank tellers, checkout assistants, cleaners, factory workers, telephone operators - and so on. In the job market, machines will gradually catch up with humans - not overtake them suddenly. Functionally, human intelligence goes from Einstein - right down to heavily disabled individuals that are practically comatose. The distribution of human intelligence forms a bell-shaped curve - not a sharp spike - and in the job market - which is where humans and machines compete - the spread of intelligence required to perform tasks is even broader. Many humans are currently doing jobs that could be being done by much less intelligent machines.

    "Singular" just means "unique"

    What about the idea that the coming events are "singular" - because they are unique - and have never happened before? Many events fit the description of being unique. The second world war was unique. The millennium was unique. The moon landings were unique. The discovery of DNA was unique. Yet we do not describe these events as singularities - and nor should we.

    Robin Hanson has even tried rebranding the agricultural revolution and the industrial revolution as singularities in a recent article.

    Robin wrote: "whatever the Industrial Revolution was, clearly it was an event worthy of the name 'singularity'".

    I disagree - I thing calling the industrial revolution a singularity represents pandering to pseudoscience. I see the same problem as with all other usage of the term - the word is simply inaccurate and inappropriate. "Revolution" is a much better-established term. If the events are sufficiently dramatic, then we have precident for using the term "takeover" - and there are various other terms that would fall within existing scientific traditions.

    I think the best perspective on the "singularity" phenomenon, is that it is not science, but marketing. Usually the singularity folk have something to sell. Whether it is that the future is going to be bad, and we have to work to stop it - or that the future is going to be great, and we should unite to make it happen - or that the future is going to be weird, and that you need assistance understanding it - there is usually something that we need to contribute to, or invest in.

    I find it all rather nauseating. I would prefer it if people would use established, sensible, scientific terminology, instead of nonsensical, fantastical concepts derived from science fiction. Whenever I hear people talking about the singularity, I wonder if they know what they are talking about. Whenever I see institutions associating themselves with the singularity, my assessment of their credibility plummets. The term makes me wonder what people are selling.

    I don't mind marketing - but here we are mostly talking about ideas that are either confused, confusing, misleading, wrong - or just have nothing to do with the term being used. "The singularity" is pseudoscientific mystical-sounding mumbo-jumbo.

    My council for the singularity enthusiasts - is to reconsider your position. Scientifically speaking, your terminology sucks. It is not cool or sexy - it is mostly just bad marketing masquerading as science. For the sake of your credibility, my recommendation is to drop it.



    1. Tim Tyler - The Singularity is Nonsense;
    2. Michael Anissimov - The Word “Singularity” Has Lost All Meaning;
    3. Bob Mottram - Some thoughts on the Singularity;
    4. Kevin Kelly - The Singularity Is Always Near;
    5. Eliezer Yudkowsky - Introducing the "Singularity": Three Major Schools of Thought;
    6. Robin Hanson - Economics Of The Singularity;

    Tim Tyler | Contact |