I’m reading tweets from space. More specifically, the twitter feed of @VanguardAdrift, part of an interactive art experience I read about in an article titled ‘Tweet me, I’m About to Die in Space’ (https://www.newscientist.com/article/2120497-tweet-me-im-about-to-die-in-space/). Artists Cath Le Couteur and Nick Ryan’s Project Adrift incorporates a combination of film, soundscape, and twitter feeds to explore the ‘lives’ of space junk, of which there are now more than 170 million pieces in orbit around the Earth. The oldest surviving man-made space object still in orbit, Vanguard 1, tweets updates about its journey around Earth at 21,630km/h. As I’m writing this, it tweets ‘I was there when the ISS was first launched; when astronauts began to live in space. I will outlive them all’.

The anthropomorphising of these pieces of metal whirling around the Earth is designed to raise awareness about the growing problem of space junk. However, built in 1958, Vanguard 1 does not possess the sentience to write these missives, though I think the project is awesomely nerdy and the existential tweets are cool. But surely it stops being anthropomorphising once the object possesses its own intelligence? When a machine can think for itself? In this week’s New Yorker, I’m reading an article (http://www.newyorker.com/magazine/2017/04/03/ai-versus-md) by Siddhartha Mukherjee investigating machine-learning diagnostic algorithms, specifically those that can detect and diagnose skin cancer, which utilise a computing strategy known as a ‘neural network’ modeled on how the human brain functions. In a process called ‘deep learning’, the machine recognises various types of skin lesions and gives a diagnosis on whether or not it is cancerous, with the trials the article discusses finding that the machine outperforms human dermatologists in accuracy. The fascinating thing is that the scientists running these programs don’t really understand how the computer is learning.  They call this the ‘black box problem’ because ‘all the internal adjustments and processes that allow the network to learn happen away from our scrutiny’. This to me is staggering; the computer is learning on its own and we don’t know how. Currently, while the algorithm can accurately diagnose cancerous lesions it cannot investigate the cause; it cannot understand the ‘why’. But, at the speed at which this technology is developing, the ability to understand meaning and cause can’t be that far away.