By Chaitanya Iyer*
In May 2020, OpenAI, a California based Artificial Intelligence (AI) research company, first described its new AI language-generating system called GPT-3 (Generative Pretrained Transformer 3). Since then, hundreds of media outlets have written about the system and its capabilities. Social Media has been abuzz about its power and potential. The Guardian even carried an op-ed piece entitled “A robot wrote this entire article. Are you scared yet, human?” All this hype around GPT-3 made me intrigued. So I decided to put it to the test. But there was a slight hindrance to my plan, I could not get my hands on GPT-3 as it is only accessible through a private beta, so I used the next best thing available – GPT2. The prompt I gave to the text generator was “Write a passage on procrastination in about 400 words” and it came up with this:
“Stop Procrastination Before It Stops You
Don’t be that person who puts things off. Part of getting better is knowing your triggers and not giving them even the slightest chance of occuring. The first week is always the toughest. Remember that your To-Do List Isn’t A Ladder. You can start anywhere on it. You can get those numbers of items on your To Do list from anywhere. They could be things you thought you would never do, but because they are important and have impact, they are on your list. The last item on my To Do list was to write and I am amazed that I actually did it. I had many reasons to be afraid of doing it. If you’re one of those thinking “Why in the world would anyone want to read my meandering thoughts about my mundane life?” then you’re not alone. I’m the same way. I thought writing would be a waste of my time, but it ended up being the thing I least expected. When I look back on it, I have never had a greater sense of accomplishment. The best way to stop procrastinating on a task is to just do It and not complain. There is an old adage that says “Complaining is like a coffee stain. Once it’s there, you can’t get rid of it.” And we’ve all been guilty of doing it. If you’re not happy with something, say something. Most of us can’t even admit to it. But it can be done. When you’re stuck, admit it. And do something about it. Don’t just complain and feel bad about it. Productivity is key to being satisfied and happy. Many people have tried and failed to do better at it and so they give up. But if you want to have fun and be satisfied then you must find a system that works for you. If you don’t, you will never really learn how to have fun, get more done, or be satisfied. If you don’t enjoy your work, you’re working way too many hours, and not developing a healthy work – life balance. Get your own hobbies, find the things you’re good at, and do them to the best of your ability. And remember, a balanced life is a happy life!”
When I saw this result I was shocked, this looks exactly like something you would read in a self help blog online. And this wasn’t even the most advanced text generation software out there. GPT-3 has nearly 100 times more parameters than GPT-2. If you Google GPT-3 you will find countless other examples of such passages being churned out by the software. It can even write prose in the style of famous authors. But does this mean that AI can think like humans and has sentience? Simply put the answer is no. GPT-3 does a great job at imitating human writing but it falls short in showing comprehension skills and simple reasoning abilities. It is also unable to identify questions that make no sense like “How many eyes does my foot have?” In essence, GPT-3 is like a kind of vast, eclectic scrapbook created from millions and millions of snippets of text that it then glues together in weird and wonderful ways on demand.
GPT-3 is quite impressive in some areas, and still clearly subhuman in others. However this is not something new. Since the 1950s, AI has repeatedly overpromised and under delivered. And like most advancements before it GPT-3 also added fresh fuel to the raging debate on AI and its power. The increasing ability of machines in recent years to replicate or even supersede human abilities in complex tasks has been impressive. This has sparked concerns of millions of people losing their jobs and a robot uprising that will obliterate humanity. And the people who are perturbed about the capabilities of AI are not luddites but eminent tech entrepreneurs like Elon Musk and Bill Gates. This begs the question, why do we need to research AI safety? In the near term, the goal of keeping AI’s impact on society beneficial motivates research in many areas, from economics and law to technical topics such as verification, validity, security and control. Whereas it may be little more than a minor nuisance if your laptop crashes or gets hacked, it becomes all the more important that an AI system does what you want it to do if it controls your car, your airplane, your pacemaker, your automated trading system or your power grid. The biggest problem with AI would be if we fail to fully align the AI’s goals with ours, which is strikingly difficult. If you ask a smart car to take you to the airport as fast as possible, it might get you there chased by helicopters and covered in vomit, doing not what you wanted but literally what you asked for. Most misconceptions regarding AI are related to the myth that machines can’t control humans. But keep in mind Intelligence enables control: humans control tigers not because we are stronger, but because we are smarter. This means that if we cede our position as smartest on our planet, it’s possible that we might also cede control.
*Freelance developer based in Nagpur