Nassim Taleb has coined a new word, “antifragile”, in a recent book with that title. I was surprised that such a word did not previously exist, since, like most people, I had thought that the opposite of “fragile” was “robust,” but Taleb argues that something that is robust merely tolerates adverse or unexpected conditions, whereas something that is antifragile thrives – actually improves its performance – under such conditions. He uses the example of a mailed package labeled “fragile, do not shake.” The opposite would be a label that said “antifragile, please shake.”
Taleb’s book mostly considers the notions of fragility and antifragility in biological, medical, economic, and political systems, but I kept wondering: do we have any electronic systems that are antifragile? I’m not sure, but I’m still thinking about it.
Certainly, we engineers have done a superb job in designing robust systems. In spite of their burgeoning complexity, our systems have much better lifetimes than their predecessors of a few decades earlier. The Internet would never work without the robust flow control, alternative routing, and error control that have been incorporated. In spite of the periodic predictions of a collapse, the Internet has survived mostly intact for about three decades now.
We might argue that, just as evolution improves biological systems through survival of the fittest, there is a similar evolutionary improvement in the performance of electronic systems as we learn to anticipate failure mechanisms. However, I don’t think this is the same as the idea of designing a system that will actually work better than it does normally when it experiences unexpected or random conditions.
There are, of course, power supplies that harvest energy from random vibrations, but perhaps that is too trivial an example. The closest I can come to antifragility in complex systems lies in those that experience multipath phenomena.
For the first half of the last century multipath phenomena were harmful. In radio frequency transmission multipath caused signal fading as different paths became variously additive or destructive. In wireline transmission there were similar effects due to the non-uniform delay of signal frequency components, resulting in intersymbol interference and limiting digital transmission speeds.
In the second half of the last century these multipath impairments were alleviated through diversity and adaptive signal processing. Still, although the point is arguable, I think of these adaptive systems as robust, rather than truly antifragile. Perhaps where we crossed the line to antifragile was with the advent of MIMO – Multiple Input Multiple Output – systems where we deliberately send multiple copies of the signal from different antennas, hoping that there will be multipath phenomena that with processing can be used to enhance system performance. These systems are now commonly used in IEEE 802.11n WiFi, as well as elsewhere.
What I find fascinating now is the application of similar techniques in the new field of computational photography. In any scene, light arriving at our eyes or at a camera lens has come from multiple reflections, refractions, absorptions, and so forth. An ordinary camera captures an instantaneous superposition of all these arriving rays. Any information about the directionality of rays, relative time delays, and individual amplitudes has been lost.
The first camera to record more of this information, the Lytro, is now a commercial product. Using a multi-faceted lens, and taking advantage of the ever-growing sensor capacity, the Lytro captures directionality of arriving light rays. I am reminded of the early work in computer graphics creating life-like representations using ray-tracing, where light rays are traced from a source to a viewer – with all their reflections and refractions – and computed to produce the image. With the Lytro we have the inverse problem; given the rays, can we compute the scene? The Lytro software does exactly that, and permits the user to choose a focus or move the perspective subsequent to image capture.
Recent experimental work at MIT carries this field much further. Using a femtosecond laser to send extremely short pulses of light, and repeating many times to gather enough signal strength, they record times of arrival and relative amplitudes. The amazing claim is that it is possible to see around corners with this technology. Some of the arriving light pulses will have been reflected from surfaces that are not within the direct view of the detector. The more bounces, the better. Multipath is good; bring it on!
I invite the reader to think of existing or potential systems that might thrive under unexpected and/or adverse conditions. It’s a thought-provoking challenge.