Daron Acemoglu, the Nobel-winning MIT economist, expects AI to take over only about 5% of human work within 10 years.

He assumes simple office and cognitive tasks, about 20% of US work, can be automated, but expects only a quarter of them to be automated within a decade.
humanistreview

That implied a GDP gain of about 1.5% over 10 years.

He cites the example of the speech recognition software Dragon again and again in his article.

- "voice recognition software on PCs today is only marginally better than what was available in 2000"

Dragon 16 claimed 99% accuracy, measured under ideal conditions. That last 1% can flip a sentence's meaning, cause a command to be misread, or force you to proofread every email.