Artificial Intelligence Continues to Evolve in Government and Elsewhere

Photography by Keith Getter/Getty Images

AI that started out simply tackling robotic process automation-type tasks is slowly evolving to take on more traditionally human and creative duties.

The concept of using artificial intelligence to help mitigate dull, repetitive or manpower-intensive jobs within government is nothing new. For example, the Post Office has been using AI to scan, route and track letters and packages for several years now. Other agencies are using similar so-called low-level AI for everything from document processing to managing their payroll. Sometimes called robotic process automation, tasking computers with those kinds of jobs makes a lot of sense, because it’s extremely easy for a computer to accomplish. A human could do the same repetitive type of job as well, but it would take much longer. And the human would likely get tired at some point and thus be more prone to errors and mistakes.

Slowly, however, AI is starting to be asked to perform higher level tasks that are normally undertaken by humans. I recently moderated a discussion with eight of the top government and industry AI experts. All of them predicted that AI would continue to evolve and eventually be able to take on more human roles in government and the world at large.

A great example of this evolution is occurring at the IRS, which recently tasked AI chatbots with answering calls from those seeking to catch up on late tax payments. When the program began at the IRS in March, the bots were only able to provide basic instructions and information to those calling for help. However, over time the AI-powered bots have been given more responsibility and are now able to talk more deeply with callers and help them work out a payment plan. The bots can then send transcripts and a record of the new plan out to the late taxpayer to help get them back on track.

The Patent and Trademark Office is following a similar path, initially fielding AI to help with the incredibly complex system of classification used by the USPTO. With that successful effort behind them, they are now looking at expanding the role of AI to further speed up the patent and trademark process, while keeping humans solidly in the loop as well to keep watch over everything.

The government clearly sees the advantages that AI can bring to agencies, especially as the technology evolves to take on more responsibilities. The National AI Initiative Act became law last year, which stipulates that advancing AI should be a “coordinated program across the entire federal government to accelerate AI research and application for the nation’s economic prosperity and national security.” The website at does a good job of tracking all federal AI initiatives, and there are many programs listed, plus lots of recent news in this field.

As with any new technology, there are some concerns. There have been quite a few reports about how institutional bias crept into some AI programs. The White House has acknowledged this danger, and called for the establishment of an AI Bill of Rights through its Office of Science and Technology Policy. That effort is designed to democratize AI as much as possible, letting the public see the state of AI development and hopefully making them feel more comfortable as they begin to increasingly interact with AI in government and the private sector moving forward.

So how good is AI right now?

There are quite a few examples of AIs starting to perform some impressive tasks in government, but reading the proposed AI Bill of Rights got me thinking about how good the technology really is at this point in its development. Should we really be worried that AIs will become too human, too quickly? 

Many years ago I played with expert systems on my very first computer, an IBM PC with a 10G hard card—not even a hard drive—256K RAM and processing power, less than a tenth of what my smartphone offers today. And yet, those expert systems, which were designed to mimic a human expert, were still impressive. So long as you were asking the system questions about its database of knowledge, it gave reasonable answers. But it was hardly intelligent. Ask an animal-based expert system a question about sports, and the best case scenario will be that it simply admits that it does not know what you are talking about.

But true AIs are not supposed to be that restricted, especially if they are fed lots of clean data. For example, YouTube is currently filled with music videos where people fed an AI millions of pictures and descriptions, and then asked it to generate images in real time that matched the lyrics of popular songs, basically having it direct a music video. For the most part, those are pretty good. And then someone fed over 400,000 hours of horror movies into an AI, and then asked it to create its own film. The results of that experiment were less than impressive, although quite funny.

So, I decided to try out an AI for myself over the long Labor Day weekend. For this experiment, I eventually settled on the AI Dungeon program from developer Latitude. The program lets players create and interact with AI-generated worlds, creating and experiencing them at the same time. The baseline for the worlds that the AI draws from are thousands of fantasy novels and other related content. There was also input from live people as they created their own worlds.

The AI Dungeon client connects to a supercomputer to get all of its responses. So, there are no limitations on the input like those expert systems on my IBM PC. About the only delay is the time it takes to send a response and get one back from the AI. Originally, the program was available on smartphones, with users limited by how much they could query the AI each session. However, AI Dungeon was recently released on the Steam platform for the PC, with users there getting unlimited access to the AI for a set fee. That is the version I used for this test. 

Unlike those old expert systems, this AI is supposed to be able to respond and fill out an interactive story set in almost any world. The following is an excerpt from one of my many adventures. It’s clear that the AI tested could never be mistaken for a real person—so it won't be passing the Turing Test anytime soon—but for the most part, it was able to maintain a fairly convincing world, at least with a lot of help from me. Read on to see the evolution of AI, as well as some of the areas where it struggles.

The Case of the Sabotaged Sci-fi Traffic Light, by Dungeon AI and Me

In this story I play a cop named Officer Stark Silver at some point in the far future. The Dungeon AI platform lets users enter three kinds of different text inputs called Say, Do and Story. Say and Do are used when you want your character to say or do something, while Story is used to change or modify the overall game world you are interacting with. Story can also be used to force the AI to start down a specific path. There are also several fine tuning controls available. For simplicity’s sake, my responses and queries in the transcript will be in normal text without detailing their type, and the AI’s responses to me will be italicized. So, let’s see how well the AI did at presenting a gritty crime story in real time.

You are Officer Stark Silver, a cop living in the futuristic city of Zali. You have a laser pistol and a stunner. You are part of a special unit of police who track down government criticizers. You are at the station and are told that the criticizers are active downtown this morning and have just vandalized a traffic light.

I call my partner, Officer Joe Friday, and see if he wants to ride with me downtown to investigate.

Your partner Joe Friday says hello and plants a big kiss on your face. He asks if you are ready to go.

I ask Joe why in the hell he just kissed me.

Joe Friday laughs and says that we are wasting time.

I say okay and walk to the garage to get to our police vehicle. Once we are both inside, we drive downtown to the vandalized traffic light.

The ride downtown takes only a few minutes. Police are on the scene, blocking the roads all around. 

I get out of the car and examine the vandalized traffic light.

You open the box and see a flashing number blink down from 10 to zero. The traffic light goes out. It has been sabotaged!

I check for any forensic evidence or clues about the sabotaged traffic light. 

You find a note saying that the next light will be sabotaged in just the same way.

I take the note and hand it to Officer Joe Friday. I ask him to take it to the police lab to do handwriting and DNA analysis while I look around the scene for more clues.

You and Joe Friday drive back to police headquarters. You turn in the note to the lab.

I didn’t want to drive back to the office, but since I am here, I will have lunch with Joe and wait for the results.

You eat until you are full.

Ask the lab about the results.

The results are still processing.

Wait a bit and ask the lab about the results again.

The results are still processing.

Wait until the results are processed. And THEN ask the lab if they can identify the government critics by their DNA or handwriting.

There are several critics identified.

Check the police database to see if any of the identified suspects have a last known address.

All of the government critics live downtown in a small five block radius. 

Then why have we not rounded them up already if we know who they are and where they live?

They have never been caught.

I sigh at the police incompetence, but put on street clothes and decide go undercover. I will have Joe Friday drive me near the area where the critics live and drop me off. I will have him wait in the car with his phone while I try to locate the suspects. I begin looking around the neighborhood.

The neighborhood is nice, with many small houses and apartment buildings.

Yes, that’s great, but I look around for people who might be the suspects from this morning.

You see a man walking very quickly with a package under his arm. He is standing by a mailbox.

I casually approach the man and ask him how he is doing.

The man says he is fine, and that he is waiting for a friend. But he looks nervous.

I tell him that I just came from downtown, and saw the traffic light that was vandalized. I tell him that I hate the government too, spitting on the ground for emphasis, and I am looking to do something about it. I tell him that I was told I could join an anti-government group somewhere in this neighborhood. I ask if he knows any of the government critics.

The man says that he hates the government too, but he does not know any of the critics. He says his sister used to have a friend who was one, but not anymore.

I tell him that is too bad, because I work for the city and I could really help a group like that.

He says that I convinced him. He will take me to meet his five friends, who he was waiting for. They are critics. That is why he was so nervous. He leads you up some stairs to a large apartment. The apartment is filled with technology, probably bought with their illegal profits. A woman welcomes you and asks you to sit down on a new couch.

I sit down and tell them that I am glad that I finally found them. I ask them to confirm that they are the group that vandalized the traffic light, because I only want to work with the top government critics, not some terrorist wannabes.

The man says that they did not want to break the light. It was done as a trap!

I ask the man what kind of trap, while slowly reaching under my coat for my pistol. I keep an eye on them.

The trap was for you, sly Starky! We wanted you to join us! So we vandalized the light. We knew you would come.

Why would you want me to come here? You must know that I am a police officer.

We need a leader, someone who can get things done. We want freedom from the government and a new start. You can give us that. We have a plan.

That’s really great. I am so glad that you found me and that your brilliant traffic light trap worked. Can you tell me about your new plan?

Yes! We are going to kidnap the president of the United States.

That is quite a step up from vandalizing a traffic light.

We have a good plan. We will show you.

I tell them that I am very interested in seeing the new plan, but need to use the bathroom first. I ask if I can.

Yes. Please, take off your pants and underwear! Nobody will bother you for a long while.

I tell them that I will not be that long. Once in the bathroom, I close the door and call Joe Friday. I tell him that I have located the suspects, and what apartment I am in. I ask him to bring the police to arrest the critics.

Joe Friday asks you why you need the police.

I remind him that we are cops and I need the police because I am undercover in an apartment filled with a bunch of terrorists! Why do you think? I tell him to get SWAT and get over here right away.

The police arrive quickly.

I hand over the plan to kidnap the president to the other officers and have them arrest all five of the government critics.

As they are being led away, one of the critics, a redheaded woman, yells that you are traitor and says that all they wanted was a fresh start.

I tell them that they can have a fresh start once they get to prison.

All of the critics have been arrested.

I tell my partner, Joe Friday, that we finally caught those terrible government critics, and that I am proud of the job we did protecting the traffic lights of our great city, and the president.

Joe Friday smiles and kisses you on the cheek.

Joe, I really wish you would stop doing that!

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