Artificial Intelligence (AI) is in the news. It seems important. What is it and why should you care?
If you have any idea about what AI is, you probably are thinking of computers solving puzzles and playing games. Perhaps you remember or heard of the famous chess match between world champion Gary Kasparov and IBM's Deep Blue computer in 1997. Kasparov lost that match and was none too happy about it.
William Bader and I attended the match in person. It was thrilling. Millions watched the match over the Internet. It was just beginning to really grow then. IBM's servers crashed under the load. However IBM's market cap quickly rose 30 billion dollars after its victory, another hint as to AI's future importance.
So is AI something you might get your head around? If you do, will your understanding of it make you want to do more with it?
Intelligence is about figuring things out. Artificial Intelligence is about having machines do it. When I told my sainted mother that I was making AI applications, she quickly got it and said, "Well, I guess having artificial intelligence is better than having none at all."
So what about figuring things out? There is a simple traditional problem in computer science that can be used to illustrate this. It is called the Queens Problem, or more specifically, the 8 Queens Problem. It goes like this. Place 8 hostile queens on a chess board such that none attacks another. Unlike a typical algorithmic recipe where all the steps are followed one after another, solving this puzzle requires trial and error. Even with placing queens in an orderly fashion, you end up getting stuck and having to remove queens and restart part of the search.
Figuring it out requires backtracking. You could scale this little chess problem to solving mate-in-two problems as I did during my AI research into chess programming. Or you could go for solving a space filling problem at an industrial scale.
SCS does this with its classified pagination technology, SCS/ClassPag™. Did you ever watch an operator fiddle with classified pages using a manual layout program like InDesign® or QuarkXPress®? Don't fall in love with the fancy graphical user interface or the incredibly fast pointing, clicking, dragging and dropping that the operator does. Doing this is incredibly slow and wasteful of both newsprint and labor. We know. And once you see SCS/ClassPag in action doing hundreds of pages per second and packing pages with near optimal density, you will want to do classified pagination with nothing else.
Here is an example that might help make the difference between data processing and AI programming clearer.
Say you are a baker. You make delicious cupcakes and sell them in a retail setting. You follow a recipe, so much of this, so much of that, cook this long at this temperature, etc. Doing 100 cupcakes is just like doing one, 100 times. The work effort goes up linearly with the count of cupcakes.
Now consider what happens when you open up to the Internet. No, this isn't about apps, it's about complexity. To make things interesting, consider packing 100 cupcakes for shipping. A regular arrangement in a shipping box could make this relatively easy. Not too bad. Now let's up the ante. Perhaps the cupcakes in an order can be of different sizes and maybe keeping cupcakes with one icing away from certain others with different icings is also necessary for shipping. Perhaps some cupcakes can be stacked on others, and others not.
Now the complexity increases. Packing the box for this situation might require considering all the cupcakes at once when doing the box packing. Every new cupcake adds a possibility of needing to redo all the packing. That's what happens with classified pagination. The complexity of finding a good packing rises exponentially with each new ad.
You could pretend this wasn't the situation. A now-gone classified pagination vendor tried to market using an idea they called "Live Pagination". That is, put one ad on a page as each is ordered. Most computer scientists I know would say the one-at-a-time approach loses almost any chance of getting close to an optimal packing solution.
SCS took their classified pagination business from this other vendor. On the way down they claimed their method provided real time estimates of classified area. Our counter was to point out that with SCS/ClassPag doing hundreds of pages per second, a really valid estimate for doing it well could be had in just seconds. QED.
BTW - Mixing ad ordering with pagination will cause heartburn in any developer wishing for an understandable, modular systems architecture.
AI at SCS is not an exercise in academic research. William is the principal author of SCS/ClassPag. There has been a lot to figure out to make SCS/ClassPag the world-wide success it is.
Consider what's involved in packing pages tight. It is just one of many problems that need to be solved. Let's say you start with classified liners composed into little PDF rectangles along with some descriptive information about them. Now you let them flow onto pages, column by column. They are likely to be variable in size, maybe an average of 3/4" in height. Should work nicely, right? Well, not quite. If columns are 21 inches in height and the ads come in random sizes, then on average 3/8th of an inch per column would be wasted. That's one percent!
Our metro customers like the LA Times or Clarin in Buenos Aires may save up to $1,730 or $491 respectively on any day (say Sunday) with a large classified section.
Savings = LA Times circ. 653,868 times 80 pages of classified times $700 per metric ton divided by 2204 pounds per ton divided by 96 broadsheet pages per pound
times 1 percent
(((700/2204.62)/96)*80)*653868*.01 = $1730
Clarin circ. 270,000 times 110 pages of classified times $700 per metric ton
divided by 2204 pounds per ton divided by 192 tab pages per pound
times 1 percent.
(((700/2204.62)/192)*110)*270000*.01 = $491
With SCS/ClassPag the labor savings are even more impressive. At Clarin the job that 14 paginators did is now done by two, one using a computer workstation and another watching. You can imagine for yourself how large the labor savings SCS/ClassPag provides throughout the entire tronc (Tribune) organization.
For the computer scientist, 2-dimensional bin packing is one of those notoriously difficult NP-Hard problems (Google it.) Yet it doesn't hurt to think of it as being like figuring out one of those traditional little puzzles AI researchers have worked with, like the Queens Problem.
If you have read this far and want to see more, email me at