Analyst Angle: AI, specious tech races, and the dangers of being first

A lot was made last week of Samsung’s large investment into AI research, and a media-fueled 5G tech race between China and the United States.  While wholly separate subjects, the notion that all sea change-type technology developments are a race could be especially dangerous as the “Rise of the Machines” era dawns.

Samsung’s AI, 5G and semiconductor investment

First things first; let’s not confuse Samsung’s announcement of its technology investment priorities over the next three years as anything dangerous.  The company hit many of the right notes as it outlined investment programs into technologies that will surely shape the next five years of telecom, IT and IIoT technology development. So, while some might take issue with the particulars of Samsung’s strategy, from where I sit the move seems both aggressive and appropriate considering the Korean vendor’s place in the AI and 5G ecosystems. They are behind Huawei, Nokia and others in 5G, and they’re behind Apple and Huawei in AI. They need to spend big on both.

The U.S./China 5G race

All I’ll say here is that for every article in the press about China winning the 5G tech race, there is an industry analyst, or other informed observer cautioning us to pump the brakes. I’m clearly biased in favor of the analyst crowd, but here I’m really with them.  There is no race… at least not yet.

When separate ingredients can make a for a dangerous mix

Where these two seemingly separate news items can sound an alarm bell is not necessarily in the news items themselves, but in the concepts that they represent. 1) There’s a ton of investment going on in AI right now; 2) if there is a 5G race going on, then we better believe there’s a larger AI race being run. Here’s where things could get scary. Simply put, AI, at least on a large scale, is not ready to replace human thinking.

Here’s a link to a TEDx talk about AI gone wrong in an extraordinarily simplistic example.

In my mind, the example is a great reminder of a fundamental tenet of AI. Regardless of how sophisticated the algorithms become, at its root, AI is the product of the data that informs the learning engines. Taking things a step further, to inform a truly “intelligent” machine, massive amounts of data will have to be fed into the system. Logically, when feeding a machine learning algorithm with huge lake(s) of data, unknown anomalies will exist. In the example above, this resulted in a mistaken association between snow and a wolf. In a factory automation, or healthcare, or autonomous driving example, the unintended consequences of an inaccurate data association could be lethal.

Now, the good news that the folks responsible for designing AI systems are well versed in these dangers. In response, they have arrived at a set of best practices for AI deployments that include several steps, one of the most important being multiple iterations of testing. This is wise. However, in R&D cycles that are attached to ROI schedules, “multiple iterations” could be incongruent with time-to-market expectations. As concepts such as 5G become intrinsically intertwined with (some might even say, dependent on) AI and ML, the “race” mentality could result in some very bad things.

Here is where we should do our best to avoid taking the tech race concept too seriously. To be sure, AI will change the world. Whether it does better or worse will be heavily dependent on the course the respective contestants take to commercialization.

Comments are closed.