By James Wallace Harris, Wednesday, December 17, 2014
People commonly accept that robots are replacing humans at manual labor, but think they will never replace us at mental labor, believing that our brain power and creativity are exclusive to biological beings. Think again. Watch this video from Jeremy Howard, it will be worth the twenty minutes it will cost you. It’s one of the most impactful TED Talks I’ve seen.
What Howard is reporting on is machine learning, especially Deep Learning. Humans could never program machines to think, but what if machines learn to think through interaction with reality – like we do?
But just before I watched that TED Talk, I came across this article, “It’s Happening: Robots May Be The Creative Artists of the Future” over at MakeUseOf. Brad Merrill reviews robots that write essays, compose music, paints pictures and learning to see. Here’s the thing, up till now, we think of robots as doing physical tasks that are programmed by humans. We picture humans minds analyzing all the possible steps in the task, and then creating algorithms in a computer language to get the computers to do jobs we don’t want to do. But could we ever tell a computer to, “Compose me a melody!” without defining all the steps?
The example Jeremy Howard gives of machine learning, is Arthur Samuel teaching a computer to play checkers. Instead of programming all the possible moves and game strategy, Samuel programmed the computer to play checkers against itself and to learn the game through experience – he programmed a learning method. That was a long time ago. We’re now teaching computers to see, by giving them millions of photographs to analyze, and then helping them to learn the common names for distinctive objects they detect. Sort of like what we do with kids when they point to a dog.
What has kept robots in factories doing grunt work is they can’t see and hear like we do, or understand language and talk like people. What’s happening in computer science right now is they can get computers to do each of these things separately, and are close to getting machines that can combine all these human like abilities into one system. How many humans will McDonalds hire to take orders when they have a machine that listens and talks to customers and works 24x7x365 with no breaks? As Howard points out, 80% of the workforce in most industrialized countries are service workers. What happens when machines can do service work cheaper than humans?
Corporations are out to make money. If they can find any way to do something cheaper, they will, and one of the biggest way to eliminate overhead is to get rid of humans. Greed is the driving force of our economy and politics. We will not stop or outlaw automation. Over at io9, they offer, “12 Reasons Robots Will Always Have An Advantage Over Humans.”
Now, I’m not even saying we should stop all of this. I doubt we could anyway. I’m saying we need to learn to adapt to living with machines. A good example is playing chess. Machines can already beat humans, so why keep playing chess? But what if you combined humans and chess machines, to play as teams against other teams, who will win? Read “The Chess Master and the Computer” by Garry Kasparov over at The New York Review of Books. In a 2005 free for all match, it wasn’t Grand Masters with supercomputers that won, but two so-so human amateur players using three regular computers. As Howard points out, humans without medical experience are using Deep Learning programs to analyze medical scans and diagnose cancers as well or better than experienced doctors.
When Jeremy Howard talks about Deep Learning algorithms, I wished I had a machine that could read the internet for me and process thousands of articles to help me write essays. So I could say to my computer, “Find me 12 computer programs that paint artistically and links to their artwork.” That way I wouldn’t have to do all the grunt work with Google myself. For example, it should find Harold Cohen’s AI artist, AARON. I found that with a little effort, but who else is working in this area around the world? Finding that out would take a good bit of work which I’d like to offload.
Imagine the science fiction novel I could write with the aid of an intelligent machine. I think we’re getting close to when computers can be research assistants, yet in five or ten years, they won’t need us at all, and could write their own science fiction novels. Will computer programs win the Hugo Award for best novel someday? And after that, a human and machine co-authors might write a more thrilling novel of wonder.