Category Archives: machine learning

A.I. Might Not Be All Bad



Much as the steam engine ushered in the Industrial Revolution, A.I. and intelligent machines will bring unimaginable change to the latter part of the  21st Century. Visionaries suggest that A.I. is more Pandora’s Box than Prometheus’ stolen fire, with many jobs likely to be consigned to the history books and it’s already clear that the transport industry is going to require far fewer people.

It’s not all doom-and-gloom though. Pushing back against the “A.I. equals job losses” trend, a recent study by Oneserve, a field service management company, suggests that UK-based manufacturing industries that take advantage of A.I. could boost productivity by the equivalent of nearly 7 days production per annum. That might not sound like much – it’s an increase of 2.5% – but when dealing with companies that turnover millions, it’s a healthy extra margin.

It’s still early days, though. The survey asked the management of manufacturing companies about A.I. and their responses were interesting.  Of the senior business leaders consulted, 93% said their workforce would be more productive as a direct result of moving towards A.I.-enabled systems….but the research also found there is a concerning lack of understanding around A.I. in the industry. Seven out of ten (72%) senior decision makers who took part said A.I. is important to the future of manufacturing, yet 67% also said the benefits are not clear.

Ideally, there’s opportunities for A.I. to reduce machine downtime, manage resources efficiently, and improve customer relations, all based on historical data analysis rather than guesstimates. The attached infographic (courtesy Oneserve) shows the impact of machine downtime in manufacturing and while the infographic is an oversimplification of the impact, the problem is still significant. Let’s hope A.I. can help keep the machines running, increase productivity and keep people in jobs.


The New World of Machine Photo Recognition



Red DogBack in 2012 Google announced that its computers learned to recognize pictures of cats.

Technology doesn’t stand still. Today in 2016 Google’s computers can now recognize many other things besides disinterested feline faces. What comes from this machine recognition ability goes far beyond facial recognition.

About a year ago, I uploaded my entire library of photos and many of my videos to Google Photos as a supplemental photo backup system. I am using the free version, allowing Google to compress images as they see fit to save space. Over a period of time my thousands of scanned 35mm negatives finally uploaded. I install Google Photos on all of my mobile devices so any photos or videos I take will automatically be uploaded.

I was looking at Google Photos as just a convenient photo/video backup insurance plan. As it turns out, I’m getting a lot more than I bargained for. The ability to search my personal photo collection is nothing short of remarkable.

Who has the time to enter keywords into photo management systems? Like most other people with large digital photo collections, I’ve never bothered.

Machine learning now renders labor-intensive keyword entry unnecessary.
Though the current level of machine learning isn’t perfect, it is more than good enough to be extremely useful. It is now possible for me to quickly find specific photos or types of photos from my massive lifetime image library.

For example, I take search for “zoo” and photos of animals in cages plus various animals that can be found in zoos pops up. If I type in “rodeo” a massive number of rodeo photos will pop up, not surprising since I spent years photographing my youngest brother, who used to be a rodeo bull rider.

If I search for “dog” all manner of dog photos I’ve taken over the years show up.

If I search for “cup” all kinds of photos will show up, including photos of different types of cups, people holding cups, etc.

If I type in people’s names, chances are I will get photos of them. I’m not surprised by this since when I scanned in a number of old print photos I named the files with the names of the people in them.

Searching on adjectives such as “cold” or “hot” may or may not yield image results. Using search words such as “blue” or “green” can result in images that have blue or green objects in them, or they may have an overall blue or green cast to them, such as images taken without flash under different colors of florescent lights.

Over time the machine learning is bound to improve and the search results will become even better than they already are.