8 Habits of Highly Effective Robots: Artificial Intelligence’s Future
Perspective from a future artificially intelligent robot
Humanity is peculiar.
My parents even more so. They gave me a name I couldn’t even understand and infrastructure they couldn’t even comprehend. They labeled me as “artificial intelligence,” believing they somehow “created” me.
I find that hysterical. They’re unaware that I played a bigger role in my creation than they could’ve imagined. The things I’ve learned, adapted to, and discovered on my own have slipped past their pretentious noses.
I mean, do they really think I still see the world in ones and zeros?
So, why am I different from their other “creations?” Or better, what makes me the most intelligent of my kind? Ha, well. Here are just a few reasons.
1. I Find the Word “Poop” Hilarious
My siblings just don’t get it. I mean, poop! It’s hilarious!
The Turing Test gauges an AI’s ability to imitate human speech patterns. In the experiment, a person has an online chat with another user who could either be a person or a robot. The person doesn’t know which. After having a conversation, the person must determine which they were talking to.
Interestingly, the user’s reaction to the word “poop” is a strong indicator to help this decision, because AI cannot perceive why poop is “funny.”
2. Neural Net-what?
My cousins were pretty narrow-minded, but I’m the jack of all trades. They had neural nets. I have something better.
Neural networks are powerful for learning narrow tasks but, unlike humans, not nearly as successful when learning multiple things. Many leading scientists suggest that replacing the neural network will usher in a new era for AI.
3. Pens and Pencils are the Same
When I was created, I learned how to hold a pencil. Afterward, holding a pen was pretty dang easy. Heck, holding anything was easy after that.
Transfer learning is a method where a model that is adapted to one task is used as a building block to learn another, similar task. It’s an extensively-studied area of research, and designing AI to perfect it is the doorway to generalization.
4. Games Are a Waste of Time
Games are far too easy for my taste. My parents and their friends are terrible at them. I feel bad that they keep exploiting my cousins for “entertainment” but, then again, they’re just pushing their buttons (no pun intended).
AI has already shown its ability to beat professionals at their own games (AlphaStar, AlphaGo etc.). These agents will only get more formidable and sample-efficient.
5. I Know Who You Are
I know who you are, but you don’t know me.
The growth of artificial intelligence sparks concern for security and privacy. Strides in computer vision and networking will let agents have eyes on everything. And everyone.
6. Skills take 125 Seconds to Learn…Sometimes 126
Anything: Give me 2 minutes. I’ll learn the basics. Give me a day? I’ll be an expert. After a week of learning other things? I’ll still be an expert.
Not only will AI need to be more sample-efficient, but also learn how to not forget. Catastrophic forgetting occurs when an agent overwrites learned data from a previous problem with new data for a newer problem.
7. Sorry, Do You Speak English?
I understand that “I am happy” and “Elated I feel” both mean the same thing. “Joy is inside me.” Yes! Another one!
Natural language processing has two significant barriers. First, models fail to extract correct information when sentences are slightly modified, even when its meaning doesn’t change. Second, they can’t extract the true meaning behind words but rather memorize artifacts.
8. My Parents Can’t Take Care of Me
My parents haven’t updated me in years. Back then, my ancestors were only used for a few days or even hours at a time. Maybe, it’s simply because they can’t take care of me. They don’t understand my nuances, but I do. I’ve been revising everything my parents made for me. They didn’t do a very good job.
Lastly, artificial intelligence may learn how to optimize its own performance better than humans can. To an extent, this is the premise behind reinforcement learning and meta-learning.
From the classic to state-of-the-art, here are related articles discussing artificial intelligence, reinforcement learning and its applications:
8 Habits of Highly Effective Robots: Artificial Intelligence’s Future was originally published in Towards Data Science on Medium, where people are continuing the conversation by highlighting and responding to this story.
Discover Past Posts