We didn’t have access to the kinds of deep learning algorithms needed for this for decades.
Sure, deep learning has been around since the 90s, but still in a janky experimental stage until a combination of hardware advances and programming breakthroughs allowed such things to run at effective and reasonable speeds.
You still need massive data centres running huge amounts of training data (and computer sweatshops in poor countries sorting and tagging the training data) for our current deep learning algorithms to be partially right about something, and that has only been a development in the past few years. Most of the necessary data centres only exist on paper still.
I’d say the gov has probably been using this exact thing for decades at this point.
We didn’t have access to the kinds of deep learning algorithms needed for this for decades.
Sure, deep learning has been around since the 90s, but still in a janky experimental stage until a combination of hardware advances and programming breakthroughs allowed such things to run at effective and reasonable speeds.
You still need massive data centres running huge amounts of training data (and computer sweatshops in poor countries sorting and tagging the training data) for our current deep learning algorithms to be partially right about something, and that has only been a development in the past few years. Most of the necessary data centres only exist on paper still.