People often think of robots when they hear AI. Then, they think about how robots can be made to reason like we do. “I know! Data analytics! No, wait, machine learning. No, data science.”
Okay, some cool buzzwords there. Now, let’s actually see what role data is meant to play in building “artificial” intelligence.
Let’s say you want to build a robot that can tell you if you like Game of thrones. It needs some information about you to be able to answer that – what kinds of shows you normally watch, what genres you like, what the statistical popularity of Game of thrones is and so on…that, right there, is data!
If the robot could obtain this data, it could perform certain statistics on it and determine with a probability of being right, that you either do or do not like Game of thrones. If the data is perfect, this probability would be 100 percent. As inconsistencies and dissimilarities occur, the probability will take a hit.
Okay, great! Now, let’s ask the same robot if you would like Lord of the Rings or the Miles Vorkosigan series. The robot has data for Game of thrones and no data for the other two. Without any data, it is unable to tell you if you would like the other two.
Now, think of yourself. If someone knew you like Game of Thrones, they would, at least, be able to guess that you might like Lord of the Rings or the Miles Vorkosigan series based on the kinds of things they figure you’re looking for in a story; or, the similarities and differences across these shows.
See the problem with data? A robot that knows you like something but cannot use that knowledge to guess at other information is not really intelligent. It knows you like sports, but can’t begin to figure out if you have high energy and fitness. It knows you used money to pay for everything you’ve ever bought, but can’t tell if you need money for your next purchase. Data without context is quite meaningless when building intelligence. What we need is a script.
A script tells you that you watch Game of thrones; that you watch it for the intrigue, for the complexity of the characters, for its unpredictability and whatever else. Based on this script, the robot can tell you that Lord of the Rings has excellent characters, it has intrigue, it isn’t as unpredictable but it certainly keeps you guessing enough. Likewise, Miles Vorkosigan has large-scale political intrigue, it is breathlessly paced and has compelling characters and artistic expression.
So, real artificial intelligence uses scripts, not just data.
Let’s do another example. “Jon ate at a restaurant, then he saw a movie”
That’s the only data you have. Did Jon eat food? A data driven robot would say “No idea”. Sigh…
But a script-driven robot could guess that people normally don’t eat mud or, straws at restaurants, and that they normally eat food at restaurants. Exceptions occur, but these exceptions are usually so uncharacteristic that even a human being wouldn’t think of them. A guy walks in and starts eating the cloth and humping the table, an exception; uncharacteristic enough that nobody was expecting it.
For fun, I’ll let you come up with a representation for the restaurant script. It could be a graph, a chart, a sequence of if-else statements, anything you want.
Here’s an example for the script.
If Jon went to a restaurant,
He sat down
If he sat down,
He looked at the menu,
If he looked at the menu,
He liked the lasagne,
If he liked the lasagne,
He ordered at least one serving,
And so forth…
Based off this script, a robot, from just knowing that Jon went to a restaurant and then a movie, could tell that he ate food at the restaurant, that he liked lasagne, that he probably also liked pizza and quiche, that he knew how to read, that he recognized a menu, and so much more.
The idea of scripts, an emerging cognitive-science based approach, focuses on an index of knowledge that the artificial intelligence engine can draw upon to provide specific information, and abstract to provide general information. It’s a simple, but powerful idea and it opens up truly exciting possibilities for the field.