Robin Hanson occasionally asks AI researchers how far along they are on the road to human-like AI, and tends to get answers in the 1-20% range, after decades of effort: http://www.overcomingbias.com/2012/08/ai-progress-estimate.h...
Even going with the high estimates, going at this rate it's a century away.
For actual estimates of how far off human-level AI is, check out page 10 of http://intelligence.org/2014/04/30/new-paper-the-errors-insi... which has a scatter plot of predicted completion dates vs. when the predictions were made. "There is a strong tendency to predict the development of AI within 15– 25 years
from when the prediction is made (over a third of all predictions are in this
timeframe".
With exponentially-growing things, 1% done is halfway there. From a Ray Kurzweil article:
"Kurzweil noted that many people don’t understand the basic math of exponential trends. When the Human Genome project had sequenced only 1% of the genome after seven years of costly labor, many cited the lack of progress as evidence that the project was doomed. Yet the sequencing project was riding an exponential trend in the performance DNA sequencing methods. Instead of taking another seven years to sequence a second 1%, they reached it after only one year. Then they reached 4% in about another year, then 8%, 16%, and so on. It took about as much time to sequence the last 99% as it took to sequence the first 1%. That’s the nature of exponential trends — they seem to start glacially slowly but finish lightning fast."
Assuming its a constant linear progression, which it's probably not. We could see speed ups as we go, or breakthroughs. I'm not suggesting someone'll have a Eureka moment tonight and we'll have Skynet next week, but it's not like we sit on a computer for a predictable number of years and suddenly get more research progress.