Is This the Year of Artificial Intelligence?

It should be, based on year-ahead predictions from analysts and consultants, but companies first need to get accurate data.
Published: January 7, 2020

I’ve been reading some year-ahead technology predictions. Most, if not all of them, suggest that artificial intelligence (AI) applications should be high on every company’s radar. Automation seems to be a key buzzword for 2020, but it’s related in a sense to AI. That is, computers will automate some decision-making tasks, and data will go from being something that is analyzed to something computers use to figure out what needs to be executed on. If this were really true, I’d be worried.

Why?

Before I answer that question, let me state the obvious: most so-called “AI” systems aren’t true artificial intelligence. That is, they don’t learn the way AlphaZero taught itself to master strategy games (see AlphaZero: Shedding new light on chess, shogi, and Go). They are simply algorithms that instruct computers to execute commands based on data inputs. For example, if someone clicks on certain products online, they can look up what other customers clicking on those items purchased, then recommend those other products.

These systems work well in a completely digital environment in which good information exists. For example, Facebook can track the links on which users click, then show them ads likely to appeal to them based on that data. The problem is that most companies have bad data regarding what’s happening in the real world, and any AI applications based on that data are doomed to fail.

On average, store inventory accuracy tends to be at about 65 percent. If retailers purchase merchandise or reorder based on that data, they are likely to create more problems than they solve. Hot items will not be reordered because AI systems won’t see any sales, but the lack of sales would not be due to a lack of customer interests. Customers might be very interested, but the items are actually out of stock, even though the inventory-management system says four are available on the shelf.

In manufacturing, warehouse inventory is a mess—tools end up lost and the locations of parts are often wrong in database systems. A speaker at a major U.S. airline once told the audience at one of our events that airplane engines, which are massive, become lost in the hangars in which they are being repaired. Paper manufacturers have told me that workers often can’t find rolls of paper weighing more than a ton.

The fact is, the real world is messy. Things are moved around, and workers are often too busy to record every item’s new location each time it is moved. Feeding real-world data into AI systems so the technology can make decisions? OK, so what happens if an AI program decides that a particular airplane engine should be put on an incoming plane because that plane’s engine needs servicing, and the decision is based on the location and status of the engine in the hangar—except that engine was put on a different plane?

RFID systems and other Internet of Things technologies can provide data about the real world that enables AI systems to work properly. If an RFID-tagged engine is moved, an RFID system will tell you it has moved and where it is now. No humans need to be involved, so systems are automatically updated. Companies looking to deploy AI technology to automate real-world decision making need to focus on employing an RFID system first so they can capture the data that systems need in order to make good decisions.

Artificial intelligence might catch on in 2020, but only in a small portion of a company’s decision-making.

Mark Roberti is the founder and editor of RFID Journal.