robot costume with mouth and teeth
PHOTO: Jere Keys

Artificial intelligence (AI) is here and it isn’t what we thought. 

A few decades ago, AI meant the HAL 9000, Marvin the paranoid android and other human-built intelligences that could problem solve and make decisions independent of people. These days, AI is considered just part of the overall world of machine “intelligence” landscape, which includes machine learning, deep learning and artificial general intelligence.

For now, I’m going to step back to the '90s, throw on my flannels, turn on some Nirvana, and use AI to refer to the full spectrum of machine intelligence. If people design, create, teach, and control it, then it falls into the realm of this discussion. And what are we talking about? Teaching our AIs how to behave. We have a lot of information on hand. It is this information that we are going to use to teach the AIs of this world. Whether you are ready for AI or not, now is the time to prep your information for your future AIs to consume.

AIs Are What They Eat

You likely heard the aphorism growing up, “You are what you eat.” To some extent this is true. If you eat a lot of junk food your body isn’t going to be primed to deal with everything life throws at you. This is even more true of AI. Its very existence is defined by the information you feed it.

Consider e-discovery solutions, which live lower down on the intelligence scale. Most work by learning what records are responsive to a legal request by diligent training. They offer up a document and ask a reviewer if it is responsive or not. With time, and a multitude of documents, the system is ready to begin making intelligent guesses. Eventually it can categorize the millions of records you have on hand into one of those two buckets.

If you don’t feed that engine a representative sample of your content, or a large enough one, it won’t learn what it needs to do the job. If the person training the system doesn’t have sufficient knowledge to accurately teach the engine the difference between responsive and unresponsive documents, the engine will learn incorrectly.

AI is only as effective as the information it is fed.

Related Article: Data Ingestion Best Practices

With a Little Structure, Making Sense of Chaos

The e-discovery example above demonstrates how poorly managing your information forces you to invest more time and energy training the engine to get the correct results. When we consider more complex tasks, and more automated learning engines, the importance of having your information in shape is clearer.

For example, auto-categorization of records is a more complex problem, as many record plans have hundreds, or thousands, of distinct records categories. More advanced engines can categorize content in your repository by the topic, creating logical groups. It is an amazing trick to watch, but those groupings don’t necessarily match the record's categories.

The key isn’t to have everything tamed, but to have enough tamed to let the engine take over. Based upon my experience, this level of structure is still missing in many organizations, but putting that structure in place is a small cry from forcing manual categorization of everything. Having a significant set of categorized documents allows the engine to start grouping like content and notice previously overlooked differences. The informed engine can then be pointed at great masses of un-categorized content on share drives, or their modern equivalent Dropbox, and start making sense of the chaos.

Watch Out For Bias

Intelligent agents are more advanced but generally follow a closely defined set of algorithms. These are either developed programmatically or are pointed to a series of past actions to determine most likely future actions. Sources could include web page traffic, profile information, or a broader aggregation of data about a person or case. The more information you have, the better the advice an intelligent agent can deliver. Connecting the information in your different systems to provide that big picture can pay off for the agents, and the humans you have sitting in customer service right now.

All of this presupposes the AI is learning from human actions performed in a completely fair manner. The key word here is “human.” Humans are not perfect. We all bring bias into everything we do every day. When setting up our information, we need to work to remove those biases.

As more subjective judgement enters the decision process, the more bias creeps into the equation.

Examples of teaching AI bias have turned up in facial recognition, business networking and in chatbots, such as Microsoft’s infamous Tay. Some were readily fixed. If left unidentified and unremedied, our behavior is directed by an AI using a biased dataset. The resulting actions reinforce the bias. This feedback loop can make problems even worse.

Related Article: Can Artificial Intelligence Weed Out Unconscious Bias?

Getting Information in Shape

Most of us won't be deploying intelligent agents anytime soon. Tackling basic AI elements such as classifying large collections of emails or business documents is our likely immediate future. Removing duplicated information and getting a subset of your information properly categorized can go a long way to jump-starting any AI efforts.

Removing bias from training data is still important to do today as it can help determine if your organization is treating everyone fairly. The advent of robotic process automation in business process management threatens to reinforce any existing biases. Studying statistics around and dissecting past businesses decisions can help.

The key to finding the bias is to assume it exists and to work to prove that it doesn’t. While it impossible to prove a negative, it will force you to ask the tough questions that can increase confidence in your data. While you cannot remove biased transactions from your systems, you can flag them to be removed from any teaching set.

As you can see, this is no small task. The amount of information your organization is dealing with is growing at an exponential rate. Soon, AI will be the only way for us to effectively manage the information in our organization. You need to start getting a handle on your information today so that when the opportunity to use AI arrives, you can point it at clean, accurate, fair information and watch it learn.

The future is nearly here. Don’t carry mistakes from the past into the future.