How We Put Our Team Under Artificial Intelligence Control

robot artificial intelligence

The teammates told us how they placed one of the project teams under the management of Artificial Intelligence and how it is to work being controlled by a machine…

The machine learning team has been thinking of introducing the Artificial Intelligence (hereafter ‘AI’) into our team’s workflow for a year. The experiment was in getting the AI into development of our internal project, with a deadline set to 30 days. For the sake of experiment, the company’s management decided to send our Project Manager to our second office in Vilnius, to do some other tasks.

Just so that you can understand, the neural network for our office is an AI, capable of structuring the teammates’ work, helping with decisions on bonus offers, task transfers, closed tasks and smoking out free-loaders. Put simply, this month the neural network is our PM.

AI has been connected to our task management system. We have digitized a vast amount of data for it, such as: teammates’ check-in and check-out times, database of colleagues and their wages, company’s accountancy, task assignment system. Those who’d like to work out of scope will get a wage bonus at the end of the month.

“AI surpasses human in all logical games. Probably, it can surpass us in ‘management games’ as well. Our machine learning team’s task was to develop an AI capable of enhancing the team’s efficiency. When we answer our colleagues’ questions, we often compare our system to AlphaGo, creators of which do not know how it makes decisions on its own. In terms of input, the machine receives a large amount of data, where it sees hidden consistencies. Neural network simply finds an individual management approach to everyone, but which one — even creators don’t know,” comments Vladimir Kliuenkov, machine learning team PM comments.

Probably, a neural network isn’t necessary for our team of 30. But if you think about a company with thousands of employees, every percent of efficiency matters, and it is possible to achieve with the help of a machine only.

There are four of us in the project: Misha and me are coding using Python. Vitaly is a JavaScript front-end developer. Pavel develops Android apps, and Dasha is our outstanding designer.

I’m 20 years old. My name is Mark. I’ve been into new technologies for my whole life. I study physics in Moscow Institute of Physics and Technology (MIPT) and have already been working as a Python developer for a few months. AI has become a special part of my interest. While reading Asimov, watching Matrix and Terminator 2, I didn’t expect to see the time when I will meet AI in my job.

Day 1

AI was connected to our task tracker, Youtrack. Neural network has computed maximum outcome for each teammate and spreaded tasks between us.

Vitaly started to receive tasks for application development with React Native (cross-platform development framework), after two years of concentrating on front-end development with JavaScript. Vitaly has worked on his last smartphone project long time ago already, so he has to recall everything.

However, we will now have two mobile developers in our team — Vitaly for iOS and Pavel for Android. On one side, Pasha will be a bit unloaded and will be able to take some overtime, or spend some more time with his family. In his 26, he’s already married. He and his pregnant wife rent a flat in Moscow region.

“Some additional work is good for me, of course, but some rest is more essential for an upcoming father.” Pavel liked the whole AI idea more and more. On the other side, Vitaly will need to grab some textbooks and do his work in the same time. Smoko.

“We understood that ‘the more you work, the more you earn’ approach will stimulate guys to comply with deadlines and will enhance the overall productivity. As for AI, exemption of the human factor is quite a dare idea. The only understanding that a machine can make a decision to fire someone, based on facts, that you can’t hide anything from the machine, or fool it, theoretically, can increase productivity,” CEO Alexey Spasskiy comments.

Day 4

A nine-hour workday is a torture for many of us. I got used to work starting in the afternoon, and getting used to stand up at 7 AM again is really hard. Being late at work is a subject for a $10 penalty and a red timer on desktop. Work schedule is being corrected in GitHub and task tracker.

Our project team has been always leaving office later than everyone else, we got used to a schedule like this. Now it’s really strange to leave the office at 6 PM. Vitaly with React Native textbook invited everyone to his place to watch Imaginarium. I already forgot how we gathered together at someone’s place.

Day 8

Sometimes task deadlines are not met. There are different kinds of situations, and lots of things do not depend on us. You can’t explain to AI that you’re sick, have a day-off or broke your keyboard.

Dasha is out of office today, she’s got flu. It turned out in more time spent on a task, which makes a penalty from AI possible. Human factor plays a huge role in a small office. A smart PM would transfer the tasks due to the absence of a teammate. Our Skynet doesn’t see it yet.

After the smoke break, a small portion of tasks already waited for me, luckily small ones — 20 to 25 minutes each — some bugfixes.

Day 11

Dasha didn’t manage to close her task in time and got a $20 penalty. What do we get? Dasha feels guilty for delaying the project. I bought her a Starbucks mug. Smoko. Now we have less and less time for a cigarette. We get small tasks, but urgent ones. As soon as I close one, I get another one, and 30 minutes to deliver it.

“By analyzing the completion of tasks by each specialist, AI is absolutely capable of selecting appropriate tasks for them. If a specialist is good at small tasks, machine will delegate tasks to him. For Al, every behaviour regularity is a resource, which should be used for productivity increase, without depending on whether it is an innocent habit, a touch of nature or a weakness,” machine learning team lead Vladimir Kliuenkov comments.

Day 16

Today Misha, our backend developer, opened the office for me. With 10-pound pouches under his eyes and messy beard, he told me that he has been sleeping in the office for a few days already.

According to the AI stats, Misha is on the last place. To keep up in the rating, Misha decided that he didn’t need to sleep anymore. Generally, Misha is a very responsible guy, and if there are any faults in his projects, his first duty becomes to fix them as soon as possible. He always closes his tasks just in time, and has accustomed himself to strong discipline. If Misha was a robot, he would be an ideal couple to Skynet. But now he’s the last in the rating, which enrages and even depresses him.

Seems like I’m the first to adapt to the new work schedule and I come to work in time (almost), but you can’t say that about everyone else. Pasha is constantly late. From human’s perspective, you can understand that, his wife is soon to deliver, and he’s preparing his home for a baby. But Skynet is not human, and it sees only facts: a teammate is always late, gets his tasks done slower, doesn’t do overtime, therefore, he’s not initiative.

Machine moral appraisal is very controversial. There are some studies around it nowadays, teaching human moral principles to AI.

Remember, how in ‘I, Robot,’ a robot saved sinking Will Smith’s hero instead of a girl in a car next to it. Robot has calculated the survival rate, and selected a grown-up to save.

Fundamental moral principles are added to the machines by people. “AI is and will be as good or as bad as the people who created it.”

Say, there is a pregnant woman working in a pharmacological company. To make things worse, a future single mother. Her KPI in the company drops, and management raises a question if they should fire a pregnant woman or drop the production of drugs what will help hundreds of pregnant women? Which decision is more moral? Which values should we set for the AI to follow? Income? Company size? Capitalization? Or amount of satisfied, gorged and healthy employees?

— Alexey

Day 20

On the general meeting, our machine learning team had its first results of our test. With blood, sweat and sleepless nights, Skynet managed to increase our productivity almost twice. Of course, the team truly hates it, but it doesn’t care — the AI just doesn’t care what we think about it.

It’s not a project manager whom you can move to pity with stories about your dog having eaten the homework or your granny’s birthday. It will not break or ease the grip from our hate. Sometimes managers have fights with developers, and then solicit a truce with a cake. A cake just won’t work here.

Day 22

Today Pasha received an unpleasant letter from Skynet:

Dear Pavel XXX,

We appreciate your work and contribution to team’s progress. Unfortunately, for now company doesn’t require Android developers. We are thankful to you for your work and wish you all success in your following projects.

Payment for completed days has been transferred to your bank account.

The letter was reminiscent of “Up in the Air” episode, where George Clooney’s hero fires one of his employees. But here you can’t show a photo to the monitor and ask, “What should I say them?” Of course, we’ll not fire Pasha. He’s one of those, about whom people say “until he comes, nothing works”, spirit of the company, hands of the office, brain of the team. But to keep the experiment clean, we nominally fired him.

Day 26

We were about to have a traditional match in Wormix, and got a message from AI in the corporate Telegram channel: “Task XXX is due to be completed in 4 hours 32 minutes.”
I forgot that a week ago they added this Telegram bot to our working group. Skynet-bot was there to remind us of unfinished tasks. Of course, no one wanted to relax and play after receiving such messages. Dasha even drew a portrait of our new AI-manager. A fair one, I should say.

Day 28

We didn’t celebrate the last day of Skynet work. There was no Champagne, ceremonial server burning and closing the experiment with Pirates of the Caribbean music. Each of us had our own conclusions. I don’t think that I will continue getting up at 7 AM, and Vitaliy will re-read his textbooks. By the way, I should say that he refreshed his React Native knowledge during these days, and Pavel develops for Android again. We delivered the project in time, even two days earlier.

After disconnecting the AI, our machine learning team studied the stats. Misha was cursing above all, because he should have been on the first place according to real data.

We connected the neural network to our tasks archive. Most probably, the system saw that Mikhail gives the top priority to bugfixes and delivers them the fastest. After analyzing the stats, AI got the trend: employee’s KPI increases in case of getting a low estimate of his work.

In this case, neural network simply sent Mikhail to the bottom of the rating to improve his productivity. Of course, it didn’t cope with the human factor — depression of an individual. A man can break down, just like a machine. But it will take much more time for a fix.

— Vladimir Kliuenkov

After this test, the tasks are being completed faster, nobody’s checking back or specifying. We stopped talking at all! Nominally, of course, everything became faster and more productive. But I think that small offices just don’t need Skynet. There are some small inner workings here, and the work rules are different from big companies.

I think, the guys will agree with me. This confirms the fact that three of us went underground to HeadHunter in search for a guy called Connor, just in case the AI returns. Smoko.
Source

Qubit Labs