When Will AI Actually Outdo Us?

September 13, 2018

author:

When Will AI Actually Outdo Us?

Do you ever get the feeling that all these self-proclaimed “experts in AI” don’t actually know what they’re talking about? Considering that we don’t even know how intelligence works, it can seem like just a lot of hot air. But before you ignore all the techy soothsaying, think back to the day when an IBM computer won Jeopardy. Or if that doesn’t ring a bell, maybe just ask Siri about it. You can’t deny that AI is going somewhere. And there’s nowhere to go but up.

 

A Few Cool Predictions

After consulting the crystal ball (actually, 352 specialists from Ivy-league research institutions) we’ve compiled a list of impressive future milestones for our beeping, electricity-eating friends:

 

  • In 7.2 years
    AI will be able to answer easily-Googleable “factoid” questions than experts (using the internet, of course)
  • In 7.8 years
    AI will be able to transcribe human speech (in a variety of accents!) in a noisy environment as well as you can
  • In 8 years
    AI will be able to translate most kinds of text as good as an amateur translator (in all the popular languages, including the tough ones like Chinese and Arabic!)
  • In 8.2 years
    AI will be able to provide phone banking services as well as human telephone operators (without annoying customers!)
  • In 8.4 years
    AI will be able to assemble any LEGO (Don’t go thinking that this is just child’s play)
  • In 9 years
    AI will be able to do convincing text-to-speech that can’t be distinguished from a voice actor by expert listeners
  • In 9.6 years
    AI will be able to write an essay for a high school history class (that would get you an “A” and pass plagiarism detectors, too)
  • In 9.8 years
    AI will be able to answer any “easily Googleable” open-ended question better than experts (again, with the help of the internet!)
  • In 10 years
    AI will be able to give good, natural-sounding answers to questions without definite answers (potentially handy for use with inquisitive children)
  • In 10.8 years
    AI will be able to produce a song that is indistinguishable from a new song by a particular artist (example: a song that sounds so much like Taylor Swift that even the biggest superfans would be fooled)
  • In 11.4 years
    AI will be able to skip imitating singers entirely and generate a “US Top 40” pop song in  a ready-to-go audio file
  • In 17.6 years
    AI will be able to defeat the best Go players (training by playing only as many games as human players: For reference, DeepMind’s AlphaGo has probably played a 100,000,000+ games, while Lee Sedol has probably played 50,000 games in his life!)
  • In 33 years
    AI will be able to write a “New York Times bestseller” novel or short story
  • In 33.8 years
    AI will be able to outperform humans in the Putnam mathematics competition
  • In 43.4 years
    AI will be able to prove mathematical theorems for publication in top mathematics journals (and generate new ones!)

 

The predictions above don’t take into account the stuff we haven’t even invented yet. And there’s that quantum computing thing, too. It’s a lot to take in. But for now, sit back, relax, and wait for the next robo-Taylor-Swift smash hit single.