Self-proclaimed Artificial Intelligence products and services are sweeping the HRTech Conference in Las Vegas this week. In most cases, the new tools make statistical predictions, forecasts, and recommendations. What we do with that stuff is hard to understand.

A flurry of self-proclaimed Artificial Intelligence products and services is sweeping the HRTech Conference in Las Vegas this week. Various powerful experiments in the application of Machine Learning, Natural Language Processing, and Neural Nets are diluting their impact by claiming to be something larger than they are. That’s how the early hype about emerging technologies usually goes.

Underlying this storm of new ideas are a couple of key factors:

1. Processing, which used to be prohibitively expensive, is becoming an on-demand service provided by the likes of Amazon, Google, Oracle, and Microsoft. Each company makes its services sticky by providing the latest in Open Source tools. They have processing to sell and encourage their users to consume a lot of it.  Companies, therefore, are tackling computational problems that used to seem impossible.
2. Storage prices have plummeted. Cloud storage is so inexpensive that most consumers have a few accounts with 5 or 10 Gigabytes of storage space (Dropbox, Gmail, Google Drive, Apple’s cloud, etc.). It’s possible to collect huge amounts of data and store it inexpensively.
3. The tools and techniques of Machine Learning, Natural Language Processing, Sentiment Analysis, Neural Nets, Social Network Analysis, Unstructured Data Aggregation and mining and Super Statistics are evolving quickly. 

Overall, this means that the price of a prediction is falling rapidly. Most of what is happening is a statistics revolution. There is precious little intelligence to be found.

In fact, most of the tools are wooden examples of the worst kind of employee you’d never want to hire. They are very literal. They can not change without retraining, on even the smallest nuance. They require attention and close monitoring. They usually need a support staff.

From these single-purpose analytical tools come statistical forecasts of the likelihood that someone will leave, stay, fit in, make a good employee, do well in a job, need guidance, cheat on their timecards, need to learn something, use a set of benefits, show up for a shift, connect with a key resource, pass screening requirements, become a part of succession plans, or complete a project on time.

We can tell if your job ad will attract or repel people of color, how to make sense of benefits options, how fitness and productivity are related, how to move new learning around the organization quickly, how edge players are connected to the network.

In almost all cases, the new tools make statistical predictions, forecasts, and recommendations.

What we do with that stuff is hard to understand.

We are entering an era where predictions about future states (based on complex statistical analysis of historical data and other factors) are extremely commonplace.  There’s a range of approaches to the delivery of this information.

• In a few cases, the data is presented without information or forecast. Humanyze, a pioneer in the blending of physical and digital maps of organizational networks does this. The deliverable is a model of how things work without any recommendations. This gives the organization full control about any experiment that modifies the network. Humanyze does have long-term experience with simple things that can optimize the network. The software, though, makes no suggestions.
• In the middle are the coaching services (both UltimateSoftware and Ceridian have this sort of offering). These offerings make ‘suggestions’ about what to do under the circumstances. They may take into account organizational circumstances, prior employee behavior, personality type, and/or external inputs. They guide the manager or colleague towards recommended interventions.
• Close to the coaching services but without specific guidance are services that predict likelihoods with a percentage estimate. Will they stay or go? Will they work out? Will they fit. These services are designed to be ‘input’ into a decision making process.
• Finally, there are pure decision delivery tools that reach binary conclusions. These services do things like sort 1,000 resumes into a pile of the 50 most likely candidates. 

Across the solution spectrum, predictions, suggestions and recommendations are being seamlessly integrated int existing workflows.

Just what do you do when you get one? What are the options? How do workers handle the input?