Ignoring machine learning won't make it go away

Email from Adobe Stock gets my attention

I received an email from Adobe Stock with the subject stating simply “A better way to search is here”. I didn’t think much of it at first, but after seeing the search in action via the example in the email content, I realized this was a promotion for an artificial intelligence functionality included in the Adobe Stock service.

I use Adobe Stock for all the images used in posts like this, as well as most of the stock photography in client websites. As you read further into this article you’ll understand why ignoring machine learning is not a great strategy.

Learn and leverage instead!

I did a little digging and pieced together that Adobe Sensei, Adobe’s artificial intelligence framework, was behind the “better way to search”.

Artificial intelligence replaces countless hours of human work

OK, time for things to get real. Imagine a world without artificial intelligence for a moment. Imagine you are the manager of a project to integrate a large number of new images into your companies existing collection of images.

Your boss, of course, expects all these new images to work with the same search functionality and categorizations as the existing set of images. This is a big task. So you start estimating that work to get back to your boss on how long this will take.

Let’s just make up some numbers to get the picture here. Say it takes a member of your internal staff 5 minutes to properly review and categorize each image based on the existing features like tags, layout, people included, etc… And let’s say you have 10 people on your team.

The next thing you need to know is how many images?

Boss says “About 12 million”.

Ummm… that will take your team 12,000,000 times 5 minutes divided by 10 people. So that’s 6,000,000 minutes or roughly 34.2 years.

“No problem boss, we’ll get that done in a little over 34 years, assuming we work 24×7 with no vacations. Does that work for you?”

That same task – integrating 12 million images into the existing catalog – was exactly what the team at Adobe did when it recently added Reuters Editorial images to it’s Adobe Stock service. And no it did not take 34 years.

I don’t know how long it took, but I do know they did not have a team of people manually categorizing all those images. They likely used their Adobe Sensei artificial intelligence technology to do that in a length of time that is all but trivial. And they’ve added several collections from different sources, and they continue to add more at a blinding speed.

Is the picture getting clearer on how artificial intelligence and machine learning is already impacting our world?

What is Adobe Sensei? Great experiences don’t just happen. And nobody knows that better than us. Using our decades of knowledge in creativity, documents, and marketing, Adobe Sensei harnesses trillions of content and data assets — from high-resolution images to customer clicks — all within a unified AI and machine learning framework. From image matching across millions of assets to understanding the meaning and sentiment of documents to finely targeting important audience segments, Adobe Sensei does it all.¹

Exec uses integrated image search in PowerPoint

Let’s assume the role of an executive in a large corporation tasked with presenting before a large public gathering. We’ll use PowerPoint to create the presentation.

We’ve learned from past experience that finding the right images to use in the presentation comes with many challenges, including several hours of digging for just the right image as well as copyright and other concerns. So we find out who is available from the marketing team to help out with the research and gathering of images.

We get paired up with some up-and-coming young marketer to help in the process. We explain the type of images we are looking for, the content and mood and maybe even colors. The marketer heads back down to their workstation and begins combing through current archives of licensed images as well as starting to search and gather potential images form various online picture galleries.

There are several back and forth exchanges via email: “How about these, are we on the right track?”, “Maybe a little more professional.”, “What’s the budget on this?”, “Why is there a watermark on all these?”, “How long do we have to finish this?”. Maybe that takes several hours or even a few days.

Not anymore.

Adobe, through a partnership with Microsoft, announced a new Add-In for PowerPoint that allows searching and embedding Adobe Stock with the advanced Sensei features directly into PowerPoint presentations.

This is a perfect example of a skill once thought to take the talent and experience of a trained person being replaced (or at least augmented) by artificial intelligence. The benefits of instant results are obvious here. The exchange of words, especially when trying to describe images, is challenging.

We all interpret what we hear a little differently. So if we describe what we want to the graphic designer and they go out and find a list of images that they think are relevant and then package those in a way we can review and give feedback that is an extremely long and error-prone process. This usually would require several back and forth conversations to get to the end result.

Adobe Sensei returns the results instantly – it really is amazingly fast given that the search is customized and covers millions of images. Then, you can adjust based on what you see and get another result set of immediate feedback. How many images could a human review in that time frame?

I recently switched to this approach with our clients on selecting graphics and photos for new websites. It has been very successful. I used to follow the above the process, where I was serving the role of the marketer. I would spend hours combing through online image galleries trying to select what I though (more like I hoped) was what the client was looking for.

This was often based on a brief meeting or phone call and a couple of words or phrases around the theme or mood the client was trying to evoke. This was not an efficient process. There were often statements like “We’re not quite happy with the images.” Unfortunately all that tells me is I have to guess again after spending more hours digging through images not really even knowing how to adjust the search. It’s trial and error and is inefficient when working with clients on a fixed budget.

Now I just give the client a link to Adobe Stock and have them send me the pictures they like. It’s awesome and they get the images they want and I get to skip the trial and error. When it comes to photos that evoke emotions – unfortunately you only know it when you see it. Cutting me out as the middle person has been very effective.

Drag and drop image matching

The powerful Visual Search that Adobe rolled out back in November allows drag and drop image search. You take an image you have that is similar in content to the images you are looking for and drop it into the search box. The search returns images that are similar and then you can further filter the list of results based on all the advanced features we’ve discussed.

This allows you to quickly find professional alternatives to an example photo you may have that is copyright protected and just not quite professional.

This technology is far from perfect, but it gets better over time, often on it’s own. That’s through machine learning, we’ll get into the details of machine learning later. Just know that the more images the artificial intelligence reviews, and the more selections people make, the smarter it gets.

So what exactly is AI / machine learning?

There are several articles online that provide an explanation of artificial intelligence and machine learning and some are listed at the bottom of this post. My goal is to explain these things in simple terms and relate them to real world examples happening today and what is expected to arrive in the near future.

Artificial Intelligence is the broader concept of machines being able to carry out tasks in a way that we would consider “smart”.

And,

Machine Learning is a current application of AI based around the idea that we should really just be able to give machines access to data and let them learn for themselves.²

So machine learning is just smart machines learning for themselves. Simple right?

Actually no. You might be thinking that definition doesn’t really help explain what is going on and frankly that’s a bit scary. I agree. Let’s work on clarifying that with some examples, while still keeping it as simple as possible.

Imagine using our graphic designer example from above and that we have worked with the same designer several times on a lot of different projects. Over time that designer has noticed a few trends in our feedback. For example the designer stopped sending any images with a lot of purple in them because for some reason we have never selected any image with a lot of purple in it.

That is fairly standard behavior for any good designer, to save time they will stop suggesting any images they know you will not select. However, up until recently, detecting similar trends, things that were not specifically programmed, has been practically impossible using software.

A very simple example of machine learning would be if Adobe Sensei picked up on the same trend as the designer and stopped returning images with a lot of purple in the results when we do a search. No one programmed the computer to do that.

The smart machine learned on it’s own by detecting a trend.

It would be impossible for a programmer to write code that would look for every possible interaction on every color and decide what action should be taken. But a smart machine can learn by watching behavior and if given a set of possible actions the machine can learn and adjust.

The designer in our example could learn over the course of months or years the preferences of a few people. The keywords being years and a few. Adobe Sensei with machine learning can do the same…for millions of people in a matter of weeks, days or even hours. It’s a big deal, and that is a very simple example that really just saves a little time and some back and forth.

Machine learning is a method of data analysis that automates analytical model building. Using algorithms that iteratively learn from data, machine learning allows computers to find hidden insights without being explicitly programmed where to look.³

Cloud-based gives you the power

I’m sure you’re familiar with the cloud by now, but what’s the big deal? The big deal is the possibility it opens up to accessing information and services anywhere at anytime and on almost any device. Early days of graphic design may have included opening a CD-ROM of images that were purchased for licensing. Generic images that a graphic designer would review and then select the best based on a very limited result set probably categorized in only one way. No search and no filters.

Now, you’re on your phone taking a picture of a flower and you drag and drop that into Adobe Stock, also on your phone and you filter a list of resulting images based on your picture. High-quality professional images of the flower you don’t even know the name of are returned for easy one-click licensing and use. If you happen to have editing software like access to PowerPoint or any of a number of editing products you can choose and open that image directly from your phone – anywhere with internet access. It’s a big deal.

Continuously improving with minimal oversight

The graphic designer from our example above had a manager, who also had a manager. The time spent working on finding us the perfect image required approval and oversight. Imagine keeping a team of several designers productive and happy and busy. It’s take a lot of work. People have needs and wants and all that goes with being human.

I’d wager that Adobe Sensei has yet to file any complaints with HR to date.

I’m sure there have been several technical difficulties and challenges, but once the artificial intelligence is configured and in place it requires almost no oversight. It just runs, day and night, no complaints.

How Google is Remaking Itself as a “Machine Learning First” Company

When the team began testing Smart Reply, though, users noted a weird quirk: it would often suggest inappropriate romantic responses. “One of the failure modes was this really hysterical tendency for it to say, ‘I love you’ whenever it got confused,” says Corrado. “It wasn’t a software bug — it was an error in what we asked it to do.” The program had somehow learned a subtle aspect of human behavior: “If you’re cornered, saying, ‘I love you’ is a good defensive strategy.” Corrado was able to help the team tamp down the ardor.⁴

Opens possibilities that were just not possible with human workers

It’s not just the replacement of human effort with a vastly superior and ridiculously faster technology. It is opening doors to new possibilities, new features that weren’t even possible. Imagine the process of tagging images, except the boss now wants your team to also include a list of all the significant colors in the image, several different levels of how fuzzy the background is, and labels like flower names or dog breeds. Your response “forget it”.

This is becoming trivial work for image categorization with artificial intelligence using machine learning.

Expanding features is easier

Software that performed complex functions, especially involving images (large file sizes) used to require installation on a local server or on your desktop computer. To make changes to that software was a very time-consuming and expensive process. Changes to the software were not quick and therefore took a lot of effort to prioritize, define and then program and test to make sure that everything worked as well as possible before the CD-ROMs with the software were burned.

If a significant mistake was made that could be devastating. Expanding features was anything but easy.

With this type of advanced software now in the cloud and given the availability of high bandwidth it is much easier to expand features. To program a change and test it, even to just a select group of users, is so much faster now it’s often easier to pilot test several changes and let the users decide what is best, rather than going through the process of prioritizing and basically guessing what people want.

The advantage of cloud-based technology is critical to keeping a competitive edge. Adobe is rapidly moving all of it’s licensing to the cloud-based model, not just the example of image search used above.

TrademarkVision uses machine learning technology in its image-recognition tools to determine if a new company logo is acceptable or if it violates any existing trademarks. What used to be a cumbersome, time-consuming, and not always accurate process, is now fast, efficient, and so accurate that the European Union’s trademark offices now use this system.⁵

What’s next in the near future?

Is machine learning coming for my job?

The answer to that question is starting to get a lot of media coverage, and there is no simple answer. Although there are some early estimates on jobs that will be replaced within the next 5 to 10 years, no one knows how this is going to turn out.

There are examples of technology threatening jobs in the past and how things turned out OK, but there are also many reasons why it could be different this time. We are in uncharted territory in many ways, but ignore machine learning at your own peril.

Forrester forecasts that cognitive technologies such as robots, artificial intelligence (AI), machine learning, and automation will replace 7% of US jobs by 2025.⁶

One relevant example is from Google. Google recently announced that is is giving away software for managing the hiring process to companies under 1000 employees.

Yes, giving it away.

Disrupting an entire industry, again. Disrupting the software that is sold to small and medium sized companies as well as the recruiting industry. Although small to medium sized companies (here listed as under 1000) typically do not rely on recruiters, I know that is not true. They do.

So the good news is that although machine learning may take your job if you were in that market – it will help you find another job!

Following its job hunting initiative announced back at Google I/O, the company today released a new tool to help small to mid-sized businesses recruit talent. Aptly called Hire, the service builds on G Suite and lets employers track candidates’ contact information, resume, calendar invites, and allow business partners to share feedback within the candidate’s profile. Hire can also port data over to Sheets to make candidate pipelines more glanceable.⁷

But it is much less clear when this will happen for many other jobs. The key takeaway here, is that the people that last the longest will be the ones that work with the new technology and learn to use it and grow and change with it.

The people that adjust the parts of their day by leveraging advancements in machine learning and artificial intelligence in general will be more productive and much more valuable to companies as they adjust resources. Be the one they keep around!

Oxford University researchers have estimated that 47% of U.S. jobs could be automated within the next two decades. But which ones will robots take first?⁸

What should I do?

Learn! If this article was your first step in learning about artificial intelligence and machine learning look below for more steps. Keep taking steps. Don’t make the mistake of thinking this is all out of your hands or it is up to the people in Silicon Valley. How it all impacts you personally is very much in your hands.

Ignoring machine learning won’t make it go away!

And don’t think programmers are an exception. There are already machines writing code and this will continue to advance. Another example is the availability of services that utilize artificial intelligence. Like scheduling or quiz software. Wasn’t that long ago that if a client wanted a quiz or a way to schedule meetings in their public website we would program that ourselves.

Now there are several very good services that can be included or embedded directly into their website at a low monthly cost. I could resist that trend and try to bill customers for custom programming – but that’s a losing battle. They are already finding these services on their own and if I try to fight that I will be very short on customers. So I work with those services and leverage them where it makes sense and even promote them, saving my customers money can be a competitive advantage.

The vision for DeepCoder is for a person to be able to merely give it an idea and the AI will automatically write all of the necessary code, without errors, in just seconds. More than anything, it will allow anybody with an idea to potentially build an internet business worth millions.

You’d think that DeepCoder would put a lot of programmers out of a job, but Armando Solar-Lezama, a professor at MIT, doesn’t think so. He believes this will enable programmers to attack more sophisticated problems, while the AI takes care of the tedious dirty-work.⁹

I never liked the phrase “can’t beat ’em, join ’em” but it certainly does apply here. If you dig your heels in and plan to just keep doing what you’re doing the same way, don’t be surprised when you get let go, and maybe even by a smart machine that has used machine learning to figure out the best way to communicate that news to you.

Learn more links

Automation and anxiety. Will smarter machines cause mass unemployment?https://www.economist.com/news/special-report/21700758-will-smarter-machines-cause-mass-unemployment-automation-and-anxiety

Where machines could replace humans—and where they can’t (yet)
http://www.mckinsey.com/business-functions/digital-mckinsey/our-insights/where-machines-could-replace-humans-and-where-they-cant-yet

A Gentle Guide to Machine Learning
https://monkeylearn.com/blog/a-gentle-guide-to-machine-learning/

Soon we won’t program computers. We’ll train them like dogs
https://www.wired.com/2016/05/the-end-of-code/

Nimblesoft Overview

We design, develop, manage and support custom software applications, data analytics, websites, mobile technologies and data integration projects. Our success includes delivering solutions across most industries including finance, retail, legal document/file management, home building, and healthcare. Our headquarters is in the Cincinnati OH area (Newport KY to be specific). We are partner-centric so you can focus on being customer-centric. We deliver quality, cost-effective, on-time solutions.

Our company has delivered over 40 large scale web applications/web sites in the past 9 years across more than 30 different clients including Fortune 500, non-profits, and startups. We have a staff of highly trained, technical resources in the Cincinnati and Indianapolis areas. We also have a large network of creative, marketing, and legal specialists that can be brought into projects when there is a need for more extensive capabilities or expertise.

At Nimblesoft, we don’t have clients or customers, we have partners. The difference is that most companies will come in, write up a proposal for a project and then once it is delivered, hand it off. Our partners remain with us and we continue to provide them with support and leadership on any aspect of their technical business that we can assist. We continue to provide support and enhancements to our delivered projects for years to come. Most of our partners have been with us for several years and we continue to complete projects for them both small and large in size.

Email us (We will not add you to any email distribution lists or spam you down the road.)