A useful (if not super novel) Australian summary of potential impacts of automation, and potential policy responses. 2017.

Impacts of automation

• Jobs most vulnerable are routine. Most physical jobs actually aren’t routine. Moravec’s paradox: high-level reasoning is easier than physical movement and sensing, so it is easier to replace some ‘high-skilled’ jobs such as translating or accounting than many ‘low-skilled’ jobs such as gardening or cooking. “Musical showdown” example: Bach vs academic vs computer; audience picked computer’s piece as the Bach.
• New technology companies create fewer jobs. Examples of WhatsApp ($19B, 55 staff) and Instagram ($1B, 13 staff). Note some counter-examples like Amazon, ~600K staff, but probably generally true.
• Reference to Sapiens discussing the Agricultural Revolution, in context of technology might advance even though most people are worse off.
• Hollowing-out of mid-skill jobs. For Aus 1991-2011, middle-pay occupations -8.5%, split among high-pay and low-pay. Similar in most OECD countries.
• Estimates vary on how many jobs are susceptible to automation. Analysis either at job level or task level.
  • Frey and Osborne: ~ half of US jobs in high risk category, >70% chance of atuomation in next two decades. Low-wage jobs most likely to disappear.
  • Durrant-Whyte: 40% of Aus jobs at high risk, using same method as Frey and Osborne.
  • OECD: task-level analysis, only 9% of jobs at high risk since even if some tasks are automatable, many aren’t.
  • McKinsey: at least one-third of tasks are automatable in ~60% of jobs.
  • Interesting to think here about whether ‘automatable tasks’ actually implies job loss, or just higher productivity. Salesperson data entry vs making more sales calls.
• Potential that AI shakes up underperforming industries, for the better. Richard Susskind on the professions: “By and large, our professions are unaffordable, under-exploiting technology, disempowering, ethically challengeable, underperforming, and inscrutable.”
• Mention of Tyler Cowen’s workers in two camps: “one capable of working with machines, the other replaced by them”. Above and below the API, Venkatesh Rao.

Inequality & current economy

• Stats on inequality:
  • US: Median real household income fell 10% from 1999 to 2011.
  • US: Real wealth of bottom 80% in US decreased from 1983 to 2009.
  • Aus: In decade to 2017, bottom 20% had +25% income and +4% wealth, but top 20% had +42% income and +38% wealth.
  • Aus: Top 1% wealthier than bottom 70%; wealth share of top 1% has doubled since 1970s.
• Reasons for inequality include globalisation, technology job displacement, cuts to top tax rates, more tax exemptions. And real estate?
• Intergenerational mobility is very low in the US: likelihood that earnings are predicted by parents’ earnings is almost as high as aristocratic UK. Australia middle of the pack; Scandinavia has highest mobility. Mobility correlated with income equality.
• Lack of wage growth as wages have decoupled from productivity and profits. In Aus, labour compensation as % of GDP hit lowest ever level in 2017.
• Migrants hold almost all of the full-time jobs created in Australia since 2007! “being hired for work that the locals are not qualified for” - George Megalogenis.
• Miles Corak: Investment in education decreases inequality only if spent on early education; if spent on university, it may actually increase inequality.

Education

• Aus: Less students are taking maths in high school. In 2001, ~10% took no maths; in 2014, 23% took no maths. And basic maths now a larger fraction of those who do take maths. Cultural aspects: it is fine to say “I’m hopeless at maths” in Australia.
• Maths/Science education must evolve to include more “computational thinking” Computational thinking around formulating a problem systematically so that a computer can perform calculations. Implication that you must write a program as an output. Seems slightly narrow - broader issues around understanding what’s possible with data/logic; expressing the flow without writing code would be sufficient.
• Have short specialised courses for adults, e.g. ‘computational thinking for accountants’, ‘data analysis for logistics’. In a retraining context for displaced workers, needs to look more like an apprenticeship than a degree, to avoid demotivation at having to fully start over. Singapore SkillsFuture model where people do ongoing training as a regular part of their jobs. How do you make MOOC-like options routine for people who aren’t already highly educated?