2016-10-19

More things in heaven and earth

Horatio : 
O day and night, but this is wondrous strange!

Hamlet : 
And therefore as a stranger give it welcome.
There are more things in heaven and earth, Horatio,
Than are dreamt of in your philosophy.

Horatio would certainly be as bewildered as we are today by the evergrowing number and diversity of things modern science investigation keeps discovering at a steady pace. A recurrent motto in science papers and articles I stumbled upon lately is more than expected, as the following short review illustrates, traveling outwards from earth to heaven. 

New living species, both living and fossil ones, are discovered almost on a daily basis in every corner of our planet, from the soil of our backyards to the most unlikely and remote places, and more and more studies suggest there are way more to discover than we already have. But the number of living things might be dangerously challenged by the growing number of artificial ones, products of our frantic industry cluttering our homes, backyards, cities and eventually landfills.

Even if a very populated one, our small planet is just itself a tiny thing in the universe, among a growing number of siblings. The number and variety of bodies in the Solar System, as well as the distance we can expect to find them, have been growing beyond expectations. Closer to us, a survey of impacts on the Moon over seven years has yielded more events than expected based on previous models of the distribution of small bodies in the inner Solar System. Images of the solar atmosphere by the SOHO coronograph has yielded an impressive number of spectacular sungrazing comets. And missions to planets have unveiled a wealth of amazing landscapes, comforting hopes to discover life in some of them.

Beyond the exploration of our home stellar system, the discovery of thousands of exoplanets did not come as a real surprise (our star being an exception would have been a big one), but there again we begin to discover more than expected, from an earth-sized planet around the star next door to improbable configurations such as planets orbiting binary stars. Moreover, free-floating, or so-called rogue planets, not tied to any specific star, are certainly cruising throughout our galaxy, and although very few of them have so far been actually detected, due to the extreme difficulty of such observations, some studies suggest they may outnumber the "regular" planets, those orbiting a star. Regarding stars themselves, the most recent catalog contains over one billion of them, which is less than 1% of the estimated total star population of our Milky Way galaxy, while new studies tend to indicate that the number of galaxies in the observable universe is at least one order of magnitude higher than previously thought. Even exotic thingies such as merging black holes, of which detection is now possible based on the transient ripples they create on space-time (aka gravitational waves) appear to be more frequent than expected. And the universe has certainly more in store, including the infamous missing mass, dark matter of which nature remains unknown.

The sheer number of objects unfolding in the depths of space and time is well beyond the grasp of human imagination and cataloguing power, not to mention philosophy. But fortunately the modern Horatio gets a little help from his friends, the machines. The overwhelming tasks of data acquisition, gathering and consolidation, identification, classification, cataloguing, are now more and more delegated to machines. Artificial intelligence, and singularly machine learning technology is beginning to be applied to tasks such as classifying galaxies or transient events. Using such black box systems for scientific tasks is stumbling again on issues linked to inscrutability, which we addressed in the previous post. Scientific enquiry is a very singular endeavour where whatever works is not easily accepted and the use of inscrutable information systems can be arguably considered as a non-starter. 

There are more and more things indeed in heaven and earth that we know of, and we are more and more eager to accept the unknown ones we discover every day. But the ones our poor imagination might be forever unable to fathom are those new ghosts haunting our intelligent machines. Are we ready to welcome those strangers?

[Edited, following +carey g. butler's comments to strikethrough above intelligent. Let me be agnostic about the fact that machine learning systems (or whatever systems to come) are intelligent or not, because I don't know what intelligent means exactly, be it natural or artificial. The "ghostly" point here is inscrutability.]

2016-10-13

I trust you because I don't know why

The ongoing quick and widespread development of neural networks and deep learning systems is triggering many debates and interrogations both practical and conceptual. Among various features of such systems, the most arguable ones are certainly inscrutability and fallibility. A deep learning system builds up knowledge and expertise, as natural intelligence does, by accumulation of experience of a great number of situations. It does better and better with time. But the drawback of this approach is that you can't open the box to understand how it achieves its expertise as you would do with a classical step-by-step algorithm (inscrutability), and the expertise is not 100% proof, it's bound to fail from time to time (fallibility). I've written on some philosophical aspects of those issues, and how they relate to ancient Chinese philosophy (in French here). 

A recent article in Nature entitled "Can we open the black box of AI" presents a very good review of those issues. And the bottom line of this article comforts me in the opinion that either all this debate is moot, or that it is not linked to this specific technology, and not even to any kind of technology. All the debate is to know if we can trust something we don't understand and which is, moreover, bound to fail at some point. This seems to fly in the face of centuries of science and technology development all based on understanding and control. 

Do we control and understand everything we trust? Or more exactly, do we need to understand and control before we trust? Most of the time, no. As children, we trust our parents and adult world to behave properly without understanding the why's and how's of this behavior. And if, growing up, we start trying to question those why's and how's, it might happen that for some reason we lose that trust. When I trust a friend to achieve what she promised, I won't, or a least I should not, try to control and check if she will do it or not, and how. Trust, in fact, if exactly the opposite of control. You trust because you can't afford to, or have not the technical or conceptual tools to, or simply believe it's useless, counter-productive or simply rude to understand and control.
That line of thought applies to more simple things that people. If I cross a bridge over a river, I don't check, and do not understand, most of the time, how it's built. I begin to check it if for some reason it seems poorly built, or rotten, looking like no one else has used it for ages. You trust food you eat because you trust your provider, you generally don't check the food chain again and again. You start to check when you suspect this chain to present some serious point of failure. It's not check before trusting, it's check because for some reason you don't trust anymore. The other way round is called paranoia.
Most of the time, you trust things to work safely as expected because so far they mostly did work safely. Based on experience, not logical analysis of how it works.This includes, and actually begins with, your own body and brain. Looking further at the world around you, you discover black boxes everywhere, and it's all right. Starting to check and control how they work is likely to lead you in some infinite recursion of effects and causes, and you will either reasonably stop at some point saying "well, it's gonna be all right", or pass the rest of your life lost in metaphysical and ontological mist, and fear of any action.
Let's face it. We trust before and without understanding and controlling. Every second of every day. And most of the time it's OK. Until it fails, at some point. We know that it will. We trust our body and brain in order to live, although we know they are bound to break down at some point. We are aware that things and people we trust are bound to fail once in a while. That's just how life goes. Parents have a second of distraction and a child dies crossing the street. Friends are stuck in a traffic jam, don't show up on time and miss their flight, bridges collapse in sudden earthquakes, hard drives break down, light bulbs explode, lovers betray each other ...
Despite of our awareness of such risk of failure, we keep trusting, and call this hope. Without trust we lose hope, and fall into depression and despair. This is a basic existential choice : trust and live, or try to control and understand everything, ask for total security, and despair because you can't find it. We trust each other although, and actually because, we don't know why. And knowing that each of us will eventually fail some day, if only once at this ultimate individual failure point which is called death, should make each of us more prone to forgiveness. 

Let me borrow those final words from the brand new and unexpected Nobel Prize in Literature

Trust yourself
Trust yourself to do the things that only you know best
...
Trust yourself
And look not for answers where no answers can be found
...