Big Shift In AI Perception

Developers of machine learning are beginning to put a much friendlier face on this technology.


Artificial intelligence, which has been controversial since its inception, is getting a makeover. While fears about massive job displacement and autonomous killing machines will persist, and maybe even grow, AI is being portrayed as a valuable tool for people who know how to harness its capabilities.

Underlying all of this is cheap compute and storage, which has made it possible to draw more data from sensors everywhere, run it through readily available algorithms, and come up with some surprising patterns that are well beyond the capabilities of the human brain to find, let alone process. Algorithms are now being generated on computers, rather than writing them by hand, and they are being shared across broad swaths of the research world and applied to different functions and processes.

The big surprise here is just how much of the world can be interpreted through digital processing. Simple apps on phones now can read emotions, and they can identify common characteristics for different breeds of dogs. It’s no longer about distinguishing between a dog and a cat. More advanced apps can determine if a stationary object is a rock, or a car stopped ahead on the road, and they can triangulate data spatially and temporally.

But AI also comes with several important caveats. One is that no system is perfect. There are always errors, and those errors will show up in annoying and sometimes dangerous ways. People aren’t perfect. But while machines may be better at some functions than humans, they are flawed in many ways, too. They can be hacked and give wrong answers based upon spurious data.

Second, machines degrade over time in different ways. Sensors that provide input become less sensitive. They can be damaged in harsh environments, or just begin to wear out through excessive use. The same is true for the connections between those sensors and wherever data is processed. Even the training algorithms may suffer from patch upon patch the same way software begins to slow down over time. Eventually, the various pieces can interact in unpredictable ways and produce results that are unreliable — and significantly different than the same algorithm running on a similar machine.

Third, no matter how good or complete a machine-based system appears to be, computation isn’t everything. Mathematicians are quick to remind the world that math is the only true science, but AI isn’t pure math. Most of the computations done by AI/ML/DL are approximations, based upon weighted responses to patterns in large sets of data. The larger the data set, and the more complete that data set, the better the results. But that, too, is based on a distribution rather than a fixed number.

No matter what purpose AI ultimately is used for, it still will require people to make sure it works properly and effectively over its lifetime. And despite some very high-profile concerns by some very smart people, there is a growing recognition that AI can be very useful in the right hands and for the right purposes. The big question now is how well that idea sells to a very large number of people, and how much they recognize that they’re probably already using it.

Related Articles
How Hardware Can Bias AI Data
Degrading sensors and other devices can skew AI data in ways that are difficult to discern.
Artificial Intelligence Knowledge Center
AI Top stories, special reports, technical & white papers, blogs and videos
Dirty Data: Is The Sensor Malfunctioning?
Why sensor data needs to be cleaned, and why that has broad implications for every aspect of system design.

Leave a Reply

(Note: This name will be displayed publicly)