Photo credit: Serhii Tyaglovsky on Unsplash

Adoption of AI can’t increase productivity *and* create a cornucopia of new well-paid jobs. That’s a fantasy world.

Based on the behaviour of corporations over the last decades, do you expect they will voluntarily pass on those increased profits to their workers? Or will it accelerate the movement of wealth to the company owners, as has been the case since the 1970s?

Forget AI for a moment, and imagine we wave a magic wand and immediately increase the profit margins for all those companies through some other means — e.g. eliminate corporate taxation.

In either sense, it’s a reduction of money spent on personnel costs (i.e. wages) per dollar of GDP — that’s what increased productivity means.

Let’s do a thought experiment.

If we take their predictions at face value, what AI will deliver is either increased revenue for companies adopting AI, with no change in personnel costs; a massive reduction in personnel costs for the same amount of revenue; or some combination of the two.

But to win in the market, #neuromorphic companies will need to clearly demonstrate the benefits over commodity competitors.

In a new paper @natcomms.nature.com, Sadique Sheik and I show how #neuromorphic compute can succeed, with inspiration from the success of tensor processors.

Dance-off for tenure 💯

What an incredible failure of #privatised #healthcare. Healthcare and medicine shouldn’t make a profit, and shouldn’t be privately funded or delivered. Shut down Australia’s private healthcare industry, and fully fund #Medicare. #Auspol

A toddler in cardiac distress is ignored for hours on a Saturday morning, and tragically dies. Months later, a woman goes into labour on a Saturday, needs an emergency caesarean, but no theatre staff are present to operate. 50 minutes later she gives birth vaginally, and the baby doesn’t survive.

Consequently, the hospital cuts staffing numbers to save money, especially on weekends.

A private healthcare company, given a contract by the #LNP to run a public hospital, is placed under financial pressure by the awful Australian private health insurance industry, which doesn’t pay for health services that people need.

Inform them clearly about the statistics of the PhD-to-Professor pipeline. Advise them about the political wrangling and support needed to succeed in academia.

But beyond this superficial similarity, *mechanistically* they are all fundamentally different and incomparable.

The early network layers produce things which look like receptive fields — because any optimisation method attempting to maximise information about visual scenes will extract the statistics of visual scenes. PCA, ICA, CNNs… it’s not surprising.

Good point for same constraints producing rhyming solutions to a common problem. We see that superficially with networks trained on vision tasks.