There’s an admission I’ve heard from all the people I know who practice applied artificial intelligence (AI), i.e., those who actually build real products and applications out of AI components, and it’s this: putting AI into the real world is just as much an art as it is a science. This means a couple of different things, in practice. In this post, I’ll explain the first way in which good AI is an art.
Put simply, it’s that a lot of non-artificial intelligence is bound up with successful AI. In other words, humans are very much involved, within an overall process that is boosted and assisted by the AI, but not wholly dominated by it.
People are critical to building good AI. Whether they are called “knowledge editors” or “curators” or simply “administrators”, humans must be involved, somehow, in the pre- or post-processing of whatever data or content is analyzed by an AI tool, in order to render the results usable.
As an example, take Web-based help-desk applications – the kind that let you submit a trouble-shooting question and (hopefully) receive an answer soon thereafter by email. One example might be a questions such as, “Why is my dishwasher beeping loudly in the middle of the night and won’t stop?” The best of these systems not only use AI to call up a probable answer to your trouble-shooting question, they also run this answer quickly by a real human support person. Most of the time, he or she clicks “approve” and you are emailed the answer. But often he or she will add a helpful link, type in an additional tip or, in some cases, delete the entire answer and compose a better one.
What’s happening here is that the AI and the people are working together. The AI is letting the company scale to handle millions of customers while delivering an answer more quickly to all of them. However, the human ingredient is still very much in the mix, ensuring correctness and completeness. It’s a pretty good arrangement.
So what about when it’s 3:00 am, the human help desk is unmanned and you need an answer right now? Or what if it’s 3:00 pm, all the human support staff are busy, the wait time for a response is 90 minutes, and again, you need an answer right now? In such cases, relying on the AI system is your only hope because you’re cut off, for the time being, from any form of human help.
Actually, no, you’re not. The knowledge base, and the logic engine that runs on it, might stand a good chance of answering your question instantly in an “unmanned” state, given that it has been getting groomed by human beings for months and months. It’s been codifying human knowledge, training itself and/or getting tuned by the human support staff.
Yes, those “artful” humans who have been editing, adding, deleting and correcting its “knowledge” over time, even when they are not directly accessible to you, are speaking to you indirectly, when the AI-enabled system extracts a bit of their intellectual labors and brings it to you, in that trusty email you receive at 3:05am, just five minutes after you’ve requested help. You open the email entitled, “How to make the Whirlpool K9030A Dishwasher stop beeping in alert mode,” and jump for joy. Now, you can get some sleep just like the human experts whose knowledge you are receiving are already doing. Hurray for AI—artfully employed!
At Federated Media Publishing, a stack of AI tools creates candidate collections of URL’s that bear relevant topics with respect to particular advertising campaigns. But the AI doesn’t act completely autonomously. Human editors are constantly monitoring and tuning the topic mix. Even for a particular campaign, the subject matter to be targeted can be revised before and after the launch of the ad campaign, while it is watched over by a caring staff member.
This human configuring and monitoring is an art in itself. While we could never scale our patent-pending Conversation Targeting product across our entire network without the use of AI, we’re glad that there’s room for the art of real human beings to be exercised within the process as well. It’s really the best way to accomplish scale while maintaining quality.
There’s another way that art enters into development of practical AI applications. Stay tuned for that in Part 2.