Readers: this blog is set in the future (sometime after the year 2020). Each entry assumes there has been a 5th revolution in the US — the Revenge Revolution. More about the Revenge Revolution and author, Entry #1.  Most entries are formatted as conversations. Characters appear in a number of entries, with many entries building on previous conversations.

Occasionally I do a “sense check.”  Auditing one’s own work is problematic but I try to be objective.  Entries #300 and #301 are the most recent standard “sense checks.”   The previous four entries (#310-#313) broke from conversation format.  Characters return with this entry.

Scene: Jordan’s office, Washington, DC.  Continuing conversation with Gelly, Jordan’s assistant.  Conversation began Entry #308.

Gelly: “The last lesson we had on economics, professor, was about how tariffs usually don’t work. As a result, when tariffs are imposed it’s often the working stiff that ends up…well, getting screwed.”

Jordan: “Where’s this conversation headed? And, please, no more of the ’professor’ stuff.”

Gelly: “Alright, no more name calling. But I would like to learn more about AI.”

Jordan: “For AI, you talking about ‘artificial intelligence’ or ‘artificial insemination’?”

Gelly: “You know which one – artificial intelligence.”

Jordan: “You mean artificial intelligence…like Donald Trump’s intelligence — artificial?”

Gelly: “Get serious, please. A lot of people are talking about AI and I really don’t know what it means or how it works.”

Jordan: “First, I’m not an expert on AI and don’t know exactly where it’s headed. But I’m not sure anyone does. What’s even more confusing is the term ‘artificial intelligence’ gets applied to a broad range of what seem to be not overly sophisticated software applications.”

Gelly: “I’m also confused about what I hear people calling ‘machine learning.’ Is machine learning the same as artificial intelligence or part of artificial intelligence?”

Jordan: “Let’s try to simplify the entire category. I know some techies would probably go apoplectic if they heard this explanation but we’re not trying to prepare you to write a thesis at MIT. To get started, think about AI as using more advanced software. And there are a lot of applications. At the same time, the software is designed for…”

Gelly: “Let me try, please. Rather than software for say Excel or Word, the software for AI is a designed so some kind of machine can learn to do things?”

Jordan: “That’s a good way to think about it.”

Gelly: “But does the machine end up writing its own software program? How does it learn?”

Jordan: “Think about how you learn to do some task. Could be learning to use a computer. Maybe something easier like hammering a nail. What’s the first step?”

Gelly: “Usually you watch somebody do it.”

Jordan: “Then what?”

Gelly: “When you first try you’re usually not very good at it.  So you have to practice and practice. If you’re lucky there’s a teacher around to help correct your mistakes. Eventually you get better and do whatever it is on your own.”

Jordan: “You just described how a machine learns.”

Gelly: “So, if somebody shows the machine how to say pick a box off a shelf and load it on a cart, then the machine can learn to do that?”

Jordan: “Basically, yes. The person demonstrates to the machine how to find the box, pick up the box and load it on the cart. The machine records the movements — say with some kind of camera.  Then the machine tries to repeat the movements.”

Gelly: “What if the machine makes a mistake?”

Jordan: “Good question. That’s what different about today’s machine learning and attempts in the past.  In the past, the limitations were software and computer processing. To get the machine…a robot…to do something required a lot of very precise programming. Plus the capacity of machine ‘memory chips’ was limited…intellectually challenged…so the robot couldn’t really do much.  Today’s chips are like Albert Einstein compared to earlier chips.”

Gelly: “I know computer chips have gotten more powerful. But can a machine learn a more complicated task right away?  How would it do that?”

Jordan: “The power of the computer chips is a key.  Let’s say the machine response is not quite right.  The person can do another demonstration and the machine can correct itself.  But let’s say the task is a lot more complicated.  The demonstration can be broken into smaller steps.  The machine then learns each step.”

Gelly: “If I understand correctly, by learning a set of small steps, the machine can adapt  if some part of the task changes.  For example, if the first box is small and the next box is larger, the machine could pick-up either size box and load on the cart.  I understand there’s some range of sizes but, as long as the machine has been taught to pick up a bunch of different size boxes, the machine can pick up any size box within that range.  Right?”

Jordan: “You got it.”

Gelly: “What else can these machines be taught to do?”

Jordan: “That’s the question that seems to be scaring a lot of people. As computer chips continue to be more powerful, the capabilities of the machines will continue to increase.”

Gelly: “Guess the machines are like children. As they get older, they can learn to do more things. But people seem to reach a limit as to what they can learn and do. What about machines? Will they reach a limit or will they keep on learning…and maybe outsmart humans?”

Jordan: “No one knows for sure. When I hear that question, I go back to a fundamental of learning, whether learning for humans or machines, but especially machines.”

Gelly: “You mean GIGO – garbage in, garbage out?”

Jordan: “Exactly. People need to think about what machines are being trained to do. And then think about what happens if the machines turn against the trainers.”

Gelly: “If we go back to machines picking boxes in a warehouse, what kind of bad things can the little robot do? At most, some people lose their jobs. Not nice for the people but not a societal crisis. If we talk about some sort of robot war machine that seems like a lot more risk.”

Jordan: “That’s the point. War machines are a different story. While it seems like a bit of science fiction, very sophisticated war robots could decide to turn around and attack.”

Gelly: “Now, that does seem scary.”

Jordan: “A rogue war robot…or even worse an army of rogue robots…is still an abstract idea. Because the idea of a rogue war machine is still abstract and not yet a reality is why we need to start thinking through what we’re training machines to do…and start now.”

Gelly: “Back to the more mundane. Seems as if machines can replace lots of people, even some who are considered highly skilled.”

Jordan: “Yes these robots can and probably should replace some of these people.”

Gelly: “What does society do with all the people who end up losing their jobs to machines? If a bunch of people lose jobs, won’t that negatively affect the economy?”

Jordan: “Yes, but countries having to manage the transition from an existing to a disruptive technology is not something new. Countries worldwide have faced this very problem for centuries — what about the introduction of steam engines, planes, cars, telephones, internet…and the list goes on. Some countries have managed the transitions well, other not so well.”

Gelly: “Seems to me the United States has managed well so far. But I’m not so sure going forward. Trump’s whole theme seemed to be attracting people who had been or likely to be negatively affected by new technology. Then after Trump, we had another upheaval with the Revenge Revolution. How do we manage going forward? Now, I’m getting nervous.”

Jordan: “Hold on, Gelly. Let me go get a coffee refill, then we’ll talk. Or maybe that coffee robot could get a coffee refill for me.”

Gelly: “Funny, Jordan. But I really am concerned.”