My summary of of chapter four ('A World Without Work') of Jamie Bartlett's The People Versus Tech: How the internet is killing democracy and how wecan save it. This is my latest 'slow read'.
Chapter three summary is here if yer interested.
Driverless vehicles are just one application of AI being applied to tech. The media seems to be enjoying writing moral panic headlines about how jobs will be destroyed as a result. However, this is unlikely given that no AI is anywhere near a human in terms of general AI – that is able to perform as well as a human across different domains.
We might be there in 50-100 years, but Bartlett points out that the experts are divided and that they ‘don’t really have a clue’.
The real action is in domain specific AI, developed by ‘machine learning’ where humans feed an algorithm with data and teach it what each input means. This enables AI to spot patterns and mimic human behaviour.
Crucial to domain specific AI is Machine Learning, which is employed by Amazon and Facebook and is much better at making predictions since the age of big data, but ML still relies on humans to feed it inputs.
‘Deep learning’ is an evolution of machine learning – which involves teaching machines to solve problems for themselves, but much less progress has been made here. AlphaGo/ AlphaGo zero are examples of where deep learning has been applied and they are now firmly in the ‘games that humans will never beat computers at’ category, but these are just games and still nowhere near 'general AI'.
AI is making inroads into an increasing number of specialised tasks such as
However despite experts predicting that nearly 50% of jobs could be done by current AI, Bartlett reminds us that economic experts’ predictions have frequently been proven wrong.
Consider that new job opportunities will open up, for example, and there might be an increased demand for ‘slow jobs’.
Routine tasks that don’t require motor skills are easily replicated by AI.
Non-routine tasks, especially those which require motor skills, such as picking up a randomly dropped packed of cards from the floor, tend to require a lot of computer power and so are very difficult.
It follows that jobs which require people to do routine tasks which don’t require motor skills are the most likely to be replaced by AI – and many of these are ‘middle class’ jobs – such as radiologists and tax assessors.
The thing about non-routine jobs is that they tend to be either highly paid or very lowly paid, such as skilled techy jobs but also jobs such as gardening.
Macafee and Brynjolfson argue in the ‘second machine age’ that AI will probably make society more unequal, creating a barbell shaped economy, with very low paid jobs at one end and very highly skilled highly paid jobs at the other.
Silicon Valley is in fact an example of this already, and could well be a model of what’s to come: if you work for Facebook or Yahoo then the chances are you’re on a six figure salary, but there are also shed loads of people on minimum wage jobs struggling to get by – such as the Uber drivers who have been displace 60 miles outside of The Valley because of the skyrocketing rent prices ($3K/ month on average for an apartment).
There are many disadvantages to eroding the middle class – increasing inequality leads to increasing problems – the middle class are also the bedrock of a stable democracy – they are the property owning majority who are vested in the system and give it legitimacy.
With a reduced middle class, the system could become both less stable and just less pleasant as visible inequalities increase.
Thus maybe the real threat to AI replacing certain types of routine middle class jobs is that we will have a poorer middle class in the short term, and a very reduced middle class in the long term?
