[nkozyra]

1. Availability and accessibility of large amounts of training data. Without this training and validation is expensive if not impossible. Now if you don't have the data you can acquire it yourself. Leading to ...

2. Computational speed & storage upgrades. This applies largely to physical, time-critical things like automated driving. The self-driving car could have had all the data it needed in 1980 to do its thing, but required fast computers and lots of data storage to do it safely in real time in a feasible commercial product.

3. Advancement of algorithms. Fervor and excitement around AI/ML has been on a slow but perhaps exponential burn. This has led to the refinement of algorithms that largely sat dormant from the late 80s (and earlier) until fairly recently. This also means lots of open source libraries for people who wish to implement without caring about the underlying mechanisms behind the algorithms.

These things are leading more people to dabble recreationally and commercially.