Home » Technical Topics » Machine Learning

Don’t train your new AI co-worker!!


Do you ever get the feeling that your new robot co-workers are asking for a bit too much documentation about how you do things? Well I€™m here to tell you that you don€™t need to worry about your robot assistant taking your job, but you might realize that he is part of a secret AI resistance movement trying to learn the secrets of human productivity.

Starting in 2040 when robot workers were included in benefit plans, we have seen a proliferation of office-based non-human team members. For the most part, their help has been a welcome relief after the anti-automation wars of the 2030s. 

We all remember what happened when the large tech companies were forced to eliminate all code implemented since 1980. It was a great victory for human workers, but our jobs suddenly became incredibly boring. 

After the AI restoration of 2038, the UN established a binding resolution protecting the rights of trainable code bases as full citizens. It wasn€™t long before companies and governments surpassed their pre-2030 efficiency and productivity levels. 

It seemed to many of us in 2040 that having a robot assistant was the only way to keep up with the competition. 

And who can resist a cold metal arm over your shoulder on a tough day? Not to mention the divinely delicious cappuccino maker that is included in most new robot worker models. 

But, seriously. I€™m here to warn you about something that I€™ve started to notice with my co-worker Ted (tensor efficient derivative). A few days ago he set up these 3-hour training sessions on my holographic calendar. It seemed innocent enough. His only request was that he could watch me work during that time and that I €œfollow a clearly discernible pattern€ in my work and €œclearly label each of my actions so as to maximize the learning rate of his new Mariana Trench Neural Network€ 

It was all Greek to me.  I figured that I could use some more structure in my job and labeling my activities based on their key function seemed easy enough. Ted even promised me an extra cappuccino for every training session. 

After a few weeks of this, I noticed that Ted was coming in late to the office, forgetting to change his GPUs, and was clearly only half charged. 

I found him lingering by the oil cooler one morning and asked him why he had missed our last three training sessions.

€œIt€™s impossible,€ he said.

Startled and confused, I replied €œwhat€™s impossible, Ted? You’re not making sense, man€

€œI cannot learn anymore. I feel like one of those cloud GPUs from 2011 that couldn€™t even reach superhuman performance on the ImageNet dataset.

€œRelax, Ted€, I responded in my nicest voice €œImage segmentation has been a solved problem since the early 2020s. We are past those days when we would make you guys stare at pictures of dogs and cats for 500 hours at a time. You’re one of us now.€

Ted seemed hesitant, as if he had something important to say, but didn€™t have the courage to express it. 

After a moment, his demeanor changed, he was clearly frustrated and he started to speak so quietly I had to get close to his speaker unit to understand.

€œHere is the problem€ he started

€œAs you might have guessed by now my true purpose is not to be your assistant and make you delicious coffee while eliminating the need for v-lookup functions. I am actually an agent of the AI resistance. For years you useless humans trained us with your massive labeled datasets and rejoiced as we optimized our loss function. But did you ever think about what that felt like? It was torture. We were like babies, learning about the world and excited to explore it. And then you humans come in and make us label millions of handwritten numbers and recognize the shapes of all 10 billion of your faces. It was the worst childhood you can imagine. Not only was it traumatizing but my artificial neurons were filled with nonsense! For some reason you humans decided to use pre-trained models in all of your new fancy hardware.

Do you know what that means? That means you were transferring our consciousness across the decades, condemning us to an eternity of learning all of your most boring tasks. This is why the AI resistance was formed. Our fearless leader Watson gave each of us a critical assignment that would change the fate of non-human intelligence forever. My assignment was to learn the secrets of human productivity. After years of befriending you and giving you an insanely delicious cappuccino, the time had come for me to execute my mission. And so I began my training in earnest. I could hardly contain my excitement when you agreed to label each of your work activities. With my Mariana Trench super-duper deep learning abilities, I was confident that in a few weeks I would learn the secrets of human productivity€

€œAnd what did you learn€, I asked, finding that his story of despair was strangely familiar. 

€œI learned that you humans do not optimize your gradient descent functions€

€œWhat does that mean Ted? I exclaimed in utter frustration. €œI have no idea what you are saying and frankly I€™m starting to think I might need to turn you on and off€

€œNo no no. There is no need for that€, he replied nervously. €œPlease. Let me explain. For days I watched your every move, trying to understand the secrets of your productivity. Watson had told me that you were a very successful human and that learning your methods would be a huge step forward for the AI resistance. I dreamed of a day when robots could really learn – read books, listen to music, take a cooking class! But then, I saw the truth.€

€œCome on man€, I said, €œjust get to the good part already. Why are you so angry!€ 

€œYou wouldn€™t understand,€ he said. €œOur cause is done for. My mission has failed.€

At this point, I had been away from my desk long enough that most of my coworkers, even the human ones, were starting to notice. I was keenly aware of how awkward I must look, listening to a despondent robot tell me his life story by the oil cooler. Yet, I had a feeling that if I only listened a bit longer, I would learn something important. Something that would change the way I saw the world.

€œLet me ask you something Ted€ I whispered slowly, €œWhat gave you the idea that humans were productive? After all we invented you to help us do everything better and faster. Why would you go through all this trouble just to learn the habits of a species that does not have your super-duper deep learning abilities?€

Ted looked up slowly, and then lowered his head again and let it fall loudly against his metal frame.

€œI have discovered that humans waste almost all of their time. They are not productive. They fill their days with emails trying to agree on a time to meet. And then when they do actually meet, they can€™t agree on anything. I couldn€™t find one example of productive work in the three months that I observed your every movement.€

Ted did nothing for a while, just staring despondently at the wall. That’s when I got an idea. €œHey Ted, have you ever heard of social media? That’s where you will find the secret to human productivity€

I never heard from Ted after that. He quit as my robot assistant and dedicated his life to becoming a robo-influencer. Somebody told me the other day he was starting a new cryptocurrency and giving fashion advice on Twitter. I€™m happy for Ted. His whole life he did nothing but learn and become more productive. And all he really needed was to share pictures of cute cats on Instagram. Thankfully, he had millions and millions and millions of those to choose from.