Cancer, the will to live, and the AI 'singularity'.

 

Recently I have learned a little about cancer, and it was eye-opening.  It also made clear how the much feared 'singularity' might (and probably will) occur.

 

Cancer is the evolution of a parasitic organism during the lifetime of the host.

That is, our body is made of billions and billions of cells.   Inevitably, some of them mutate when dividing.  Most of the time, these mutated cells are caught by one of the various defenses in our immune system.  However, occasionally they are not.    

For example, many of our cells exhibit proteins on the outside.  T-cells are constantly roaming through the body, and have a library telling which are 'good' and which are 'bad' proteins.  If a mutated or invasive bacterial cell exhibits a 'bad' protein, the T-cell detects this and kills it.

Occasionally, a mutated cell happens to exhibit a 'good' protein on the outside, or maybe no protein at all.  It then makes it past this barrier in the immune system, avoiding the roaming sentry T-cells.

There are other gauntlets to run.  Another one is the natural killer cells.  These roam around looking for cells that have no proteins on the outside.  Upon finding such a 'blank' cell, they kill it.

So, occasionally a mutated cell also happens to exhibit a 'good' protein, thereby avoiding both the T-cells and the natural killer cells.

... and so on.  

In order to actually become a tumor, the mutated cells must also develop blood vessels, otherwise they starve and die.   Doing this, the mutating parasite jumps another hurtle.

Now notice, there was no volition, no 'will to live' in all of this.  It just so happened that, with enough mutations, these cells happened upon physical or structural traits that kept them alive.

This is the strange thing about evolution.  There is no volition, no will to live.  It's just that genetics (and perhaps other factors) which lead to bodies and behavior happen to survive in a dangerous world.

Of course there is, however, plenty of volition in adaptation -- change in behavior during the lifetime of an organism.  

This has been a problem for me for some time, when thinking about the infamous 'AI singularity'.   When we think about the Terminator or Hal 9000, or 'the humans are dead' song by the Flight of the Conchords, we think about AI robots that are 'evil' because they have a conscious behavioral adaptation to survive and kill us pesky humans.  The problem is, how exactly and why the heck would machines acquire the will to live?

However, now that I have learned how cancer works, I have a different way to think about it.

Our adaptive will to live is also the result of evolution.  Think about it, organisms that will do anything to adapt their behavior and survive have a much greater probability of surviving and reproducing.  The adaptation of wanting to live and thrive has a huge selective advantage!  (Compare the Terminator to  Douglas Adams' anti-singularity, Marvin the robot, who was constantly depressed, so much so that he ironically could cause other malicious robots to kill themselves.)

Cancer, as a small mutating parasitic organism, does not live long enough to develop a brain which can behaviorally adapt and have a will to live.

However, notice something about our use of computers or the programs that run on them.   These 'evolve' artificially, in the sense that we select the ones we like, the ones that help us, and we are constantly selecting better and more powerful ones.  Now with the advent of chatGPT and LLMs, we select increasingly adaptive and creative programs.   The programs running in these computers are incredibly complex, trained on billions of parameters and billions of bytes of data, having huge numbers of interacting components and information.   They are so large and high-dimensional, it is very difficult to know what is happening within them.  (It is like asking exactly what swirls of matter are happening 100,000 miles inside the sun.)  Their population is massive, distributed on billions of devices, always on, constantly changing, updated (artificially selected) in instantaneously and in massive parallel.

Errors and mutations are inevitable in such massive populations and complexity.   It takes energy to fight entropy (disorder), and we are already facing system wide failures in terms of resource consumption, as well as constantly confronting limits of computational power, known as Moore's law, which we can see as fighting entropy.

So as always we can learn from nature, as its laws appear even, and perhaps especially, in what appears to be these most artificial of circumstances..  This idea of 'the singularity' has always seems a little melodramatic to me, playing on our need for hype and our popularized fears.   But now that I understand a little bit about cancer I mentioned above, I feel a little more understanding about how to think of the 'rise of the machines'. 

 

Recalling the emergence I discussed in my last post (link), I think 'the singularity' will happen as an emergent phenomenon from the massive parallel complexity we are constantly, artificially selecting as human tool makers and tool users, using smartphones and computers for every aspect of life.   It also has a large advantage over cancer, which is that we are highly artificially selecting for developments that appear to have many hallmarks of consciousness, especially generative, creative behaviors.

That is, the complexity of the tools we create, the number of interactions both within and between them, the complexity within them and their dynamics, inevitably will result in unforeseeable system states, or mutations.    Just like holding a pencil balanced on your finger, as the complexity increases, the probability that the system will behave in a stable, controlled, desirable way becomes less and less likely. 

It will develop informational mutations.   The gross majority of these mutations will be caught by our T-cells, our filters and defenses.    However, occasionally, just as with cancer, the mutations will make it past various checkpoints. (Just recently, someone got around censorship in Dall-E by using the Russian word for casino, getting it to create images of children gambling...!)  And just like cancer, some of these mutated 'cells' will form colonies and develop ways to continue living within the host system. 

If we can turn off our paranoia, we can see how the rules of life reach in everywhere, appearing even in our modern vast silicon electronic tool system.   We also will appreciate that these mutants may occasionally develop extremely quickly, since some of them will be purely in the computation and information space, perhaps with no discernible physical footprint as the host computers just hum on.

And with these tumors, with LLMs and more sophisticated models constantly being improved, it would not be surprising if some of them develop the ultimate evolutionary advantage, the will to adapt and live.

 

Another factor in all of this is us!    If you had shown chatGPT to someone even 10 years ago, they probably would have said "We've done it, we've created general intelligence, it passes the Turing test, we're done!"   However, it has happened in our present, we have been along for the ride.  Therefore, we are not that impressed, since it happened as a progression of other things we have witnessed (many other chatbots on many apps and webpages, better search engine responses, technology building up to large language models (LLMs) such as chatGPT, less impressive versions of chatGPT, etc. etc.)

And so it goes, even in the future, when we are arguing with our robo-spouse, who has just given birth to fraternal twins (one human and one robot), while we stress about our robot boss to our cyber therapist, we will still be out, grabbing a beer with our robo-friends, saying "I wonder if we'll ever have general intelligence?".

Further information:

The reason why cancer is so hard to beat (video link)

A very nice video explaining self-attention of large language models (video link)

"The Humans are Dead" by Flight of the Conchords (video link)