In a previous post the potential disruption to employment opportunities and work routines was discussed, with reference to the increasing utilisation of Machine Learning in the work place. Now, for those who fear this development, there is a handy tool to determine the likelihood of this impacting upon you. This follows from a 2015 study by Oxford University and Deloitte which concluded that 35% of current jobs would be under threat in the next 20 years. The tool can be tried here:
Thankfully writers are assessed as at low risk (33%) of automation.
That said, however, OpenAI have created a fake text generator (GPT2), capable of writing both news stories and fiction. In their defence, they have made the decision to withhold the technology and research behind it over concerns that it could be used for ‘malicious’ purposes. The data is fed into GPT2 in the form of a passage of writing, either pages or simply a few lines. The system then generates what it predicts would be the logical following lines. That figure above of 33% may need to be increased.
Rather more alarming is the ‘reproducibility crisis’ in scientific research. This is the increasing phenomenon whereby a set of results or findings are not able to be reproduced by subsequent tests or research. This has been largely attributed to the expanding use of machine learning to analyse data in scientific research. The machines can only work on the data they are given, so if a different or significantly larger data-set is employed, often the results vary hugely. It has been estimated that this is negatively impacting on up to 85% of research in bio-medicine globally and has been described as a ‘crisis in science’
Looking on the Bright Side
In case you are now considering buying a hammer and joining the Luddites or creating a playlist composed entirely of Rage Against the Machine, there are some uplifting and encouraging stories too.
The Association for Computing Machinery has awarded the Turing Award to 3 of the pioneers and early proponents of deep learning using neural networks. Geoffrey Hinton, Yoshua Bengio and Yann LeCun were jointly awarded the 2019 award for their computer engineering breakthroughs. They join illustrious former recipients such as Sir Tim Berners-Lee (2017: for the World Wide Web) and Ken Thompson (1983: for developing UNIX).
There is further cause for optimism for lovers of pesto and indeed lovers in general. Firstly MIT researchers have been using AI to improve the flavour of basil. Using data collated by mass spectrometry and gas chromatography, machine learning algorithms generated some surprising results, not least of which being that plants were adjudged to be more ‘tasty’ when exposed to light 24 hours a day.
Finally, Janelle Shane, a research scientist in optics has been, in her spare time, training neural network bots to perform a variety of more light hearted tasks, from coming up with names for paint colours to generating text for ‘love heart’ sweets. The results of the latter experiment were endearing, alarming and in some cases downright worrying. The best of them can be found on her blog here: