People Powered Robots

The Dirty Little Secret of Artificial Intelligence

Machines don’t teach themselves. We are not yet at a point where a computer can simply “look” at a situation and understand what parts of the situation are valuable to learn. Because of this — the dirty little secret of the artificial intelligence community is that human beings are HEAVILY involved in the process of teaching computers.

What does this look like?

First off — many AI companies are simply claiming to be AI when in actually they’re just a sweatshop of mechanical turks entering information into a database. But that’s not what this post is about! Those who are not that deceptive tend to have a system where human beings are able to validate or invalidate the work created by the intelligent machine. At Zero Slant we follow this second trend.

Quick background — Our AI creates stories from spikes in social media. For example — it will notice that there are a lot of social media posts at the Mercury Ballroom. It will then analyze these posts in order to learn what is going on. Once it understands that the event is a concert, our AI will title that story as the name of the band/artist and it will list keywords that it believes accurately represent the story. The human overseers at Zero Slant (myself included) take a look at the story created by zerobot. If the information is correct it is “published.” If it is incorrect, we flag the story, and zerobot learns to be better for the its next creation.

Screenshot from our CMS where we tell zerobot his business

If you’re interested in how this learning process works, check out my other piece —

In this way, human involvement in AI is largely meant to speed up education. Like people — intelligent machines learn through a loop of data input, and output validation/invalidation. Telling a computer that a task it completed was done correctly is similar to affirming an action performed by a child for the first time. The concept that they have succeeded is required for them to grow and learn.

Won’t take long for the robot to learn how to win this.

Is It Really AI If A Human is Involved?

Is it really journalism if an editor is involved? The function of people in the ever-expanding world of artificial intelligence is to check to make sure that the actions taken by these bots accurately reflect what their human counterparts would provide. The human is to the AI as the editor is to the journalist. It’s not to say that the journalist is untrustworthy, but rather that a second pair of eyes can be helpful in making sure that the outcome is as amazing as possible.

AI = code = human error

Human Involvement Improves the AI

No software is perfect right out the gate. AI is no different. In addition to acting as the editor — the human partner to the AI is constantly in search of bugs. Looking for inconsistencies and patterns in the actions taken by the intelligent machine that do not reflect the desired outcome. At Zero Slant our time checking the work of our AI, zerobot, showed us a pattern that proved to be a huge issue that was causing our intelligence to become stupider and stupider over time because of some choices we made with the original code. Without a human supervisor zerobot would have stupified himself into uselessness in a matter of months. While AI acts on its own and make decisions without human supervision, we must always remember that the intelligence is simply code — and code is created by humans (currently) — and humans make mistakes.

Google = Proof

Google is the greatest AI company on the planet. Through home grown intelligence and acquisition, no other company compares. After all, the Google search engine is completely built on machine learning. YET, even Google has people manually look at the results of searches in order to improve upon them. The job “Search Engine Evaluator” is ever-present in a Google search.

The Not So Dirty Secret

Though there are plenty of companies claiming AI and secretly delivering human-created results, involving humans in the AI process is not a dirty task. It’s actually exactly what needs to happen in order to take AI as quickly as possible to the point where the human onlookers are no longer needed. If you’re building a new AI, you’re involving human oversight. The humans help teach, help edit, and help discover bugs in the machine intelligence. At some point even these tasks will be able to be done by a computer, but for now — people to the rescue.

As always you can reach me at