On April 18, 2014, the NPR show On The Media produced a theme show that focused on Robots and Artificial Intelligence. You can listen to or download the entire podcast here.
The segment I found most interesting and potentially disruptive was a 14-minute interview with Jerry Kaplan of Stanford University titled Engineering Intelligence (link to the segment podcast here.) The segment made several key points.
- In the late 1950s, to counter customer fear of artificial intelligence taking their jobs, IBM taught their sales reps to say, “Computers can only do what they are programmed to do.” This simple truism is decidedly no longer true.
- The convergence of faster processors, cheaper memory, vast stores of information accessible by the internet, AND rapid advances in “machine learning” algorithms means that computers can now be trained to learn, to make deductions, and to behave in ways that earlier were the stuff of science fiction. We can now create computers that learn not only by training, but by example and experience. Consider, for example, that we are close to cars (and trucks) that can drive themselves more reliably and efficiently than the humans who do that job now.
- One study estimates that over the next 5-10 years as many as 47% of US jobs can be automated. These are not just low-end jobs but higher end ones like medical diagnostician, legal analyst, tax preparer, etc. So more than the lower classes will be affected.
- We have not seriously begun considering how we will distribute the profits the owners of the machines (the capitalists) will extract, especially if increasingly large joblessness and ever-greater inequality result.
- Some argue that one way to “share the wealth” is to guarantee each citizen a minimum income, whether they work or not. Is this one way to share the ever-increasing wealth without letting only the robot-owners have it all as “rent” for their ownership of the means of production?
- It does not matter that these job-bots have no consciousness. Can machines “think” or have self-awareness as humans do? One analyst notes that this is as logical as asking if submarines can swim. Obviously they don’t, but the results are the same.
- Kaplan then raises the danger that humans will attribute “agency” to machines. As learning devices, they can and will use “wiles” to attempt to manipulate humans, even though they have no needs like humans. He used the analogy of squirrels not evolving rapidly enough to learn to dodge automobiles – we humans may find ourselves handicapped by such rapid changes and run down by the equivalent of rapidly-moving 3,000 pound vehicles.
Another 7-minute segment called GOOGLE’S ROBOT BRIGADE explores reasons that Google may have launched a Manhattan Project of Artificial Intelligence. The argument is that Google wants to develop robotics, which unlike “mere” software actually converts information to objects or “stuff,” then use their automated supply chain to deliver the “stuff” while tracking what you want and where you are from their other Google services. This goes well beyond the recent book Dogfight: How Apple and Google Went to War and Started a Revolution by Fred Vogelstein, a good read.