Having tried out ChatGPT I could hardly refuse an invitation to give Bard a whirl. It is much the same sort of thing as ChatGPT, and behaves in similar ways. I have a better sense now of what to ask it to explore its capacities and shortcomings, and so I asked it:
What is an idea for a science fiction story that no one has ever used before?
Here is how Bard responded:
It is difficult to come up with an idea for a science fiction story that no one has ever used before, as the genre is vast and there have been many great stories written over the years. However, here are a few ideas that might be original:
- A story about a group of astronauts who are stranded on a planet that is inhospitable to life. They must find a way to survive and to find a way to get back to Earth.
- A story about a scientist who creates a machine that can travel through time. The scientist uses the machine to go back in time and to try to prevent a terrible disaster from happening.
- A story about a group of aliens who come to Earth and who try to learn about our culture. The aliens must deal with the challenges of trying to understand a different culture and of trying to communicate with us.
- A story about a future where technology has advanced to the point where people can live forever. The story follows the lives of a group of people who have chosen to live forever and of the challenges that they face.
- A story about a future where the Earth has been destroyed and where humanity has colonized other planets. The story follows the lives of a group of colonists who are trying to build a new life on a new planet.
These are just a few ideas for science fiction stories that no one has ever used before. There are many other possibilities, and it is up to you to come up with a story that is original and that is also engaging for your readers.
These are obviously not innovative suggestions. They are anything but, some of the most well-worn tropes. Like ChatGPT, Bard does not answer questions in the sense that we use that phrase in relation to human speech. As I said to someone on Facebook recently, it is because key words conveying factual information are woven into the texts that guide ChatGPT and Bard in making new text that they often get obvious things right. That’s precisely what misleads some people to think these chatbots answer questions. They do not. They imitate patterns in how other people have answered questions and written things that share the same key words, which is not the same thing as answering a question, or at least, not the same thing as understanding the answer to a question. I think this is something we can usefully point to in communicating to students a point that we often try to make as professors. Parroting answers others have given to a topic is not the same thing as understanding that topic. What we see from ChatGPT is not that different from what we see from a subset of our students. The reason chatbots fail our tests is the same reason human students do: an attempt to get by imitating correct answers rather than having a capacity to generate them from our own knowledge and understanding. The key difference between chatbots and human students is that the latter can learn to understand, whereas no chatbot has that capacity.
A lot of people still don’t understand how ChatGPT actually works, so let me link to this article on that topic.
I also found it funny that Bard warns of its potential to not provide factual information, and offers not just a warning but a solution: “Google it”! I suspect that advice comes from Google and was not generated by Bard.
On the other hand, I suspect that ChatPDF and other AI tutors/quizzers like it may prove to be extremely beneficial to educators and learners. I have yet to try it out, but it sounds intriguingly useful! See as well Sal Khan (head of the Khan Academy) on how he hopes AI will be used in learning. David Brin also shared some thoughts. ChatGPT-4 is here but I haven’t been willing to pay to try it. Mary Harrsch did, asking it questions about infanticide in the ancient world. Have you tried it yet?
In concluding let me mention that ChatGPT and Bard are not the only AI chatbots. Have you tried any others?
Also of interest:
How Intelligent (and Conscious and Sentient) is Artificial Intelligence?
Sean Carroll’s Mindscape Podcast: Raphaël Millière on How Artificial Intelligence Thinks
Theory of Mind Breakthrough: AI Consciousness & Disagreements at OpenAI
Yann LeCun: Dark Matter of Intelligence and Self-Supervised Learning
Andrej Karpathy and Lex Fridman talk about Transformers: The best idea in AI