Artificial Intelligence Begins to Realize Its Potential – Nextgov

Artificial Intelligence Begins to Realize Its Potential – Nextgov

In my previous column, I looked at the problem of artificial intelligences forcing hardware to consume too much power, which could lead to an unsustainable spike in demand at data centers in this country by 2025. To test out their appetites for more power, I employed several advanced artificial intelligences, and also their close cousins machine learning, cognitive computing, deep learning and advanced expert system technology. For that column, I only measured how much power they consumed, but my original intention was to actually test them out to show some innovative things the technology was accomplishing. I am circling back to that effort now.

For many years we have been reporting on the technology of artificial intelligence, about how it’s being built out and made more efficient, or how it can be paired with other technologies like quantum computing to become even more accurate. At the same time, the government has been keenly focused on AI ethics, ensuring that our newly created smart programs and machines don’t go rogue or make mistakes that could get people hurt. The Defense Department now follows five ethical principles when using AI, while the intelligence community has its own artificial intelligence ethics guidelines.

There is still a lot of learning to do, but at this point we have pretty much covered the basics in terms of building out smart AIs and related technologies, power consumption issues aside. That is why we are starting to see a lot of interesting reports on projects that make use of AI, like figuring out which fishing boats out in the ocean are using forced labor or planning how we can safely get people to Mars.

These are some of the most interesting AIs that I have collected over the past few months, what they do and how well they perform.

COBOL Colleague

The federal government invested big in the COBOL programming language back in the day. It was designed for business, finance and administrative systems within both private companies and government organizations. Today, however, it’s not actively used for any new projects, having been replaced by more efficient languages. Most COBOL programming today is used to maintain existing systems written in the language that can’t easily be replaced or recoded. The problem is that nobody is learning how to code in COBOL anymore, and most programmers that already know it have retired.

That is where Colorado-based startup Phase Change Software and their COBOL Colleague AI comes into play. Instead of trying to teach COBOL to modern programmers, it scans existing programs written in the language for vulnerabilities and problems, and zeros in on exactly what lines need to be fixed.

“There is certainly a skills shortage, however, the real problem is that the knowledge of the application is disappearing,” said Steve Brothers, COO of Phase Change Software.

Deploying an AI to look at code is almost like hiring a skilled human programmer. In the case of COBOL Colleague, it won’t make changes to the code on its own but will show where any changes are needed. Then skilled programmers, even if they are not totally familiar with COBOL, can make the necessary fixes.

ToxMod

ToxMod is an interesting artificial intelligence made by Modulate Inc, a company that specializes in innovative AI. ToxMod is designed to regulate live comments in voice chat rooms and is able to distinguish subtle differences between, say, someone using an explicative in frustration, and someone using it as an attack or as part of a hateful tirade. To give ToxMod a real workout, it’s being deployed in the ultimate toxic environment, the chat rooms of video games.

“Since ToxMod can differentiate something like honest frustration expressed in a toxic way from malicious intent, it can also advise matchmaking or reputation algorithms to improve the player experience,” said Carter Huffman, Co-Founder and CTO of Modulate. “Additionally, each game’s private ToxMod instances learn over time about their community’s specifics, on top of ToxMod’s universal core algorithms which evolve and improve automatically behind the scenes.”

ToxMod can listen to and understand emotions, volume, inflections and other factors to determine if speech should be flagged. If hateful speech is detected, site moderators are alerted along with an audio clip to back up the AI’s claim. This will let moderators check the AI’s work while identifying bad actors and preemptively resolving a problem before it grows into something more serious.

The Test

This last one is mostly just for fun, but I was so impressed that I felt like I needed to include it here. The Test is, on the surface, a series of three games available for less than $2 each on the Steam gaming platform. The games are kind of bizarre in nature. Players sit in front of a demonic-like figure at his desk and are asked a series of very personal questions. The questions consist of typical personality type questions like “If you found money on the street and knew who it belonged to, would you return it?” But there are also a series of very strange scenarios and off-the-wall questions like if you were starving at home would you eat your pets, if pink is a prettier color than red, or if would you stop a zombie apocalypse if you could.

Behind the scenes, developer Randumb Studios is likely using an expert system as opposed to a true AI to track results and prepare advice for players. The questions were so strange that I really didn’t think much of it, though I did try and answer honestly. But the final results, especially for the second game, really floored me.

It told me that I was working too hard and that I needed to take some time to recharge my batteries because I was not doing anyone any good if I was spread so thin that nobody was getting my best. It advised me to allocate two units of personal time for every one unit I spent working for others, which would be a key to both my happiness and success moving forward. 

The strange thing is that I was really thinking about this exact same thing over the past two weeks, especially with the holidays approaching. I was worried about burnout and keeping a good work and life balance, and had spent quite a few evenings contemplating that exact topic. But I never told the game this, and don’t see how it’s bizarre questions led it to that conclusion. 

I guess that is the magic of expert systems. It’s why those little 20-questions toys that ask you yes and no questions can always guess the song or the movie star that you are thinking about. But the leap that The Test made with me was a lot bigger than that. I still don’t know how they did it, but am very impressed with the results. 

The really amazing thing is that we are really just scratching the surface about what AI can do. In the near future, even the projects highlighted here will seem trivial compared to what is possible.

John Breeden II is an award-winning journalist and reviewer with over 20 years of experience covering technology. He is the CEO of the Tech Writers Bureau, a group that creates technological thought leadership content for organizations of all sizes. Twitter: @LabGuys

Source

Leave a Reply

%d bloggers like this: