Ben Collins Digital Ben Collins Digital Icon

The Next Stage of Artificial Intelligence

When computer scientists set out to create artificial intelligence, they misunderstood something very important. The goal should not have been to create something independent of humans that could do things that were apparently intelligent, but instead, the big idea was training machines to copy what we already did as humans and to improve upon our already established actions.

Fortunately, the application of trained robots became commercially viable. In factories, stock markets, and many other fields, computers are doing a good job of doing what we would ordinarily do ourselves. Just faster. And more accurately. I see this path as entirely separate from the Herculean attempt to write a program that can "think." And this path grows wider each day, with more entries and more impact than ever before occurring.

Soon, it will be very difficult to detect if anything a computer does is a person or a machine. It has nothing to do with whether the "thinking" program can fool a human with text. It has everything to do with acting like a human to the point that if you tried to write a detection script, it would have too many false positives. Browsing, gaming, whatever happens on the internet with a supposed human being behind the keyboard is suddenly and permanently in question.

It began with CAPTCHA, the most annoying thing anyone ever had to do online. Google's notorious CAPTCHA was so bad, it felt like failing a human test when you simply could not understand the squirly letters. There is a client of ours who purportedly destroyed his laptop after failing a CAPTCHA one too many times. We gave up on text CAPTCHA a long time ago, resorting to asking a simple question "What year is it?" Our much more complicated question "Lakes and streams are made of what five letter word?" was completely overkill as all the website crawling contact-form bots just give up too fast. Why bother when you have a million other contact forms to crawl today? Not only that, it makes sense to pay cheap labor to do it in case someone actually replies.

These "robots" are not "artificial intelligence" per se, but more commonly thought of as "scripts." The scripts are quite adept at using browsers, just like you can imagine a big robot welding machine perfectly putting together car doors on the assembly line, these scripts use browsers much better than a human can. I suppose the grand daddy script of them all is the Googlebot, ingesting the entire internet over and over again, snipping apart pieces of web pages, comparing them to past versions, and stuffing them in massive databases.

One major issue for game developers is quickly advancing play-for-you scripts. Does your game include a "grind" that takes many hours before you can get to the "end game"? Just make a script do it. The game developers are at a total loss for how to stop them without accidentally banning actual users who just seem to act like scripts. It's easy to inject some randomization into your play-for-me script, which wreaks havoc on any kind of detection system. I hypothesize that it is now impossible to write a script-detecting script that can reliably detect a script being actively developed by humans.

That's the trick. Humans are helping Artificial Intelligence along now, at a rapid pace, and it's happening way outside the brute force measures that in all respects failed to materialize anything highly promising. Who cares if Watson can win Jeopardy if you can't even teach it to laugh? With a very simple "If this... then that" system, you could tell Watson "If you hear someone laughing, then play this here laugh track." That's what's about to happen to the Amazon Echo and Googlina (my name for their voice interaction Google Now system).

Both Alexa and Googlina have some canned personality responses that seem to reflect the cardboard-like humor of the engineers who built them. It knows the answer to "What's the meaning of life" is 42. Ha ha. That's original. Yet, when the API is unleashed, I will be able to devise any number of answers I want to any number of questions. As can any other developer. You will be able to download a "Jokes" app. You can have a "Control Nest" app. Here's a few other ideas:

It will usher in a new form of creative writing. Humans will be thinking of ways to extend these devices far beyond the makers who just tried to get the thing working. Voice recognition is stronger than ever and finally both Amazon and Google are (supposedly) poised to take the leap and open it up to the world.

I can understand their hesitation. Once you do this, you no longer have a centrally controlled, single Artificial Entity. Instead, you have a population of configured entities, all serving their human owners based on their individual preferences.

At its core, these systems are language interpreters and if-match, then-that programs. Did we ever really need more to create true artificial intelligences that we can alll relate to?

More in Mind