We have been using chatbots for some time now. But this new chatbot ChatGPT has been creating waves since November 2022. It’s using some advanced language models from OpenAI’s GPT-3 family.
If you play with ChatGPT, it feels like soon writers won’t have any jobs. It can write really fast and even with some nuance. But is it though? The human mind is complex and how much we train our language models, it’s almost impossible to conquer that mind.
In my upcoming novel The Songs of Farewell, I cover the story of Humanoid AI investigating human extinction. Imagination has no bounds, but AI has limitations. It can only do those things that it had been trained on. Like ChatGPT, it can come up with the best writing, write code, and solve a lot of problems by answering questions.
Here is the main question – Can it create something out of thin air? Can it write the best novel without telling it or giving a hint? Can it solve the various murder mysteries of the world without knowing any details? Can it come up with a major scientific innovation?
ChatGPT is superlative. It can definitely help in many tasks and can be used for a lot of repetitive tasks. Will it replace Google? That’s a valid question. ChatGPT and Google help in similar ways, but also a little differently. It’s not so easy to replace Google considering it is a free search engine. ChatGPT does not have a revenue model unless it starts going GPT-3 way where it charges for API usage.
As an innovation, we must admire ChatGPT. That brings to me the next set of questions:
What are the possible limitations of AI?
Right now any of these AI models can’t predict any futuristic events. But neither humans can. As humans, we continuously try to predict many events unsuccessfully. Stock prices, horse racing, and betting on various sports events.
There are a number of limitations that I can think of. First AI will be able to solve something it has been trained on OR what it has been fed.
Any of these bots will not be able to understand the context of the question or problem they are trying to solve. Their understanding will remain limited.
At the current set of limitations, we can only imagine that ChatGPT or any other AI model will ever attain singularity. Singularity is complex and none of our current AI advances will take us to that stage. That is good in one way. Singularity can also be detrimental to human society as Ray Kurzweil points out in his book The Singularity is near.
ChatGPT is limited by the information it has. So, it can only do a limited set of tasks (even though there are numerous tasks it will do). The biggest challenge for ChatGPT and users of ChatGPT is to be creative. Creation is dynamic and complex. The human mind in its own way is a complex machine. The diversity and biology of the human mind and brain are not easy to decipher for a single human mind as well.