I have been catching up on readings and found three particularly interesting articles and websites addressing Artificial Intelligence (AI) concepts that may be useful.
AI Experiments: Visual Learning
First, Google’s recently launched AI Experiments allows anyone to explore the complex world of machine learning concepts using several different “games.” The most interesting of these are the “Quick, Draw!” and “What Neural Networks See.”
Quick Draw! allows you to draw sketches and the AI identifies what you are drawing. The compelling aspect of this (apart from the “parlor trick” nature of it) is that you can view the first, second and third guesses, as well as see what other sketches were used to help the AI detect a particular sketch’s subject matter. With machine learning, the real work is always behind the scenes helping to train the system through submission of ground truth data and tuning the algorithms by observing outcomes.
Our Algorithms and Their Fairness
Second, a recent article on Wired.com focuses on the fairness of algorithms. With the continued, rapid development of AI, the concerns have increased that humans will cede decision-making more and more to our machines without regard to how the machines make decisions. It is easy to see that human-developed algorithms—due to our own perspectives and biases—could inadvertently introduce these biases into things like credit scoring or job applicant screening software. The article discusses the application of “awareness” to detect these potential biases.
What I find most intriguing is that the article draws attention to developing a better understanding of the constructs of how sample sets, used for machine learning, should be selected in order to provide the fairest outcomes. While machines make it easier to develop algorithms, the onus remains on us—as humans—to teach them correctly.
What Automation Means to Jobs
Third, the next article explores the impending robot takeover. Or rather, it won’t pan-out like some have predicted with many jobs lost to automation. This article is written by a professor at Northwestern University. It addresses the lack of productivity gains and references his book, The Rise and Fall of American Growth.
Essentially the author argues the lack of productivity increases that we saw with the previous era between 1920 and 1970 is a good thing since it drives job growth. He maintains we are nowhere near robotic automation on a scale required to dispense with airline pilot jobs, service sector jobs and the like. Few, however, have been able to accurately, much less roughly predict how technology truly can disrupt markets. He makes some important points worth thinking about.
Automation: Scenario Building and Planning
All of these articles and sites are thought-provoking for different reasons. Based upon my own direct observations of how effective machine learning automation can be, scenario building and planning will help businesses better understand the specific impacts of automation, both adverse and positive. Beyond the hand-wringing and the hype, the best approach is with eyes wide open.
If you found this article interesting, you might also find this ebook on AI applied to recognition and capture useful: