Preface

This is the first joint blog entry by both Glenn and Chris. Glenn and I often discuss topics outside of a group setting, we can be more honest and blunt with each other as I think we are both well beyond the ability to feel or inflict any level of uncomfortableness with each other – a perfect platform and background for discussions focused on growth of the mind.

Background

Glenn first introduced me to the Ray Kurzweil definition of singularity, click here for a link to his site for more information.

For those with any sort of physics or cosmology background, we are not talking about what is theoretically at the center of a black hole here (not that that would not make an interesting topic of discussion as well), but instead the definition that once some artificial means becomes able to recreate and improve upon itself, whether through hardware or software, we will have reached the singularity and things will snowball from there. It follows in the discussion that this may be the next natural step in human evolution.

I read a book a couple years ago written as a discussion (fact, not fiction) between a Monk who decided to get a PHD in Physics, and a guy with a PHD in physics who decided to become a Monk! Anyhow, I found the discussion format was very productive, and so I will attempt it here.

Thoughts on the Singularity

Glenn:

I’m wondering if I have an interesting enough idea here for another blog entry on happypsychotic.com. Let me know what you think.

Chris:

Given my reply – maybe we can be co-authors?

I have some thoughts, maybe better discussed in person but for now, email will have to do.

Glenn:

If you think about human evolution over the years (whether or not you believe we evolved all the way from single celled organisms or not), I think we would all agree that humans have become more complex and “smarter”. I.e. We can master more difficult and complicated tasks as time progresses.

Chris:

This is most certainly true, in fact I think there is documented evidence (I read recently) that the average IQ of humans is in creasing about 3 points per decade. This may seem slow, but in fact is quite significant. On a greater time scale, say in the 6000 years that the earth has been in existence (ha-ha, obvious joke :) ) this becomes even more obvious that this evolution or change is taking place and real.

Glenn:

Some humans are now playing with Artificial Intelligence (A.I.) and some (most notably Ray Kurzweil) believe that in the near future (within 30 years) we will have created an AI with sufficient intelligence to improve on its own programming. This moment of time has been labelled “the singularity”. The thought being that once an AI can improve itself, it will become smarter and more complicated at exponentially shorter time intervals. It should very quickly be able to meet and surpass the complexity of humans.

Chris:

I have considered this from a software perspective recently (my specialty). I should also note I have done both AI and AL programming in the past, even 10 years ago? so I have some interest (also many books read, yada yada yada). Having said all that, I smell a certain stink here, and although it is just my opinion, you asked for it :)

I am concerned that Ray is treating the AI growth scenario as a linear problem that is just a numbers game or inevitable somehow given enough time/energy/complexity/connections/etc. Not all problems are like this however in mathematics, the classic and well known example is the NP class of problems (travelling salesman and distribution of primes are common well known examples of NP type problems). When I recently revisited this problem (basically I wanted to try and create a software version of the singularity) I very quickly faced a couple show-stopper level problems:

  1. Software (like life I guess) needs to have a purpose – without a purpose, there is no code to create really – this was immediately apparent to me.
  2. Even if you could find a purpose, then it is very difficult (if not impossible) to determine evolutionary and incremental improvement steps, and hence your software (in my attempt) can not evolve

Although creating software that evolves is extremely common, it is always narrowly defined and solves a simple problem in the sense that it has a definate, predefined answer. This is not what I am discussing here.

If you think the purpose in point 1 might be to become conscious – that sounds great and well defined, but I assure you it is not :) That is kind of the whole point: since we can not define intelligence or even really consciousness, we can not develop hardware or software to achieve it, let alone measure how close we are to it.

So I guess I would argue that although it makes sense and may even seem obvious at a high level, and after a single sentence discussion that this singularity theory of Ray’s holds water, I am not sure after an actual implementation attempt and closer theoretical analysis that is actually does.

Glenn:

My question is this…Assuming that this can happen, should we look at this AI as the next step of human evolution? If we do look at it this way, we should embrace and support this movement even if it meant the extinction of the human race as we know it – because we would say that the human race lives on through our created AI. Or should we fear this new intelligence and try to prevent it from happening and look at it as a threat to the human race (see the Terminator movies).

Chris:

I think we should embrace it – if it was possible, but as earlier stated, if “it” does happen, then I am pretty sure “it” will not take any form that we as humans have considered at this point in time.

Glenn:

Currently I can’t see how an AI could encompass the breadth and diversity of human beings but I wouldn’t say it’s impossible. If we are the creators, can we not make the created our equals?

Chris:

Sounds like you smell a little odour as well :)