If you are interested in tech then you probably got no work done this week. Like me, you’ve probably been chatting with the new chatbot, ChatGPT, the most exciting advance in artificial intelligence since Deep Blue beat Gary Kasparov.
What is ChatGPT?
You probably remember chatting with Eliza back when you got your first computer.
ChatGPT is a bit like Eliza but, while Eliza just repeats back what you said in the form of a question, ChatGPT has a huge library of text which it uses to generate responses. Here’s an example:
ChatGPT’s algorithm is not a million miles from the algorithm that Google uses to make suggestions when you start typing a query. It tries to guess what comes next based on what other people have typed in the past.
The crappy autocomplete on your phone does the same kind of thing but ChatGPT has rules to make sure the answer uses good grammar, makes sense and is on topic.
ChatGPT has its critics
I get a kick out of reading the objections of the naysayers every time there’s a new advance in AI.
“Well, yes,” they say. “AI can do X but it will never do Y.”
Their predictions never seem to last very long.
“ChatGPT just regurgitates what you can get from a Wikipedia page”, they say.
“Computers will never be creative.”, they say.
“Computers have no understanding of the topics they talk about.”, they say.
“Self-driving cars will never be able to drive selfly.”, they say.
“Computers don’t always get it right.”, they say.
Well, OK. They were right about this one. But then again, who does get it right all the time?
Sometimes ChatGPT just makes shit up.
Sometimes it learns racism and other unpleasantness from us humans.
But sometimes its answers are dazzlingly impressive.
You ain’t seen the best of me yet
I think the people who pooh-pooh artificial intelligence are making several big mistakes. Number one, they focus on the stuff that ChatGPT is not so good at. It can’t do maths, they say.
ChatGPT can’t paint pictures either.
Another criticism compares ChatGPT with the very best in society. Maybe the best teachers and the best journalists are safe for a while but not everyone who writes for a living is among the best.
Sturgeon’s Law gives us a good way to think about this. Theodore Sturgeon was a science fiction writer who was once told that “ninety percent of science fiction is crap”. Sturgeon responded that “ninety percent of everything is crap”.
You probably remember from your school days that only a handful of teachers stood out as excellent (ten percent?). The others were mostly crap. Same with paintings, movies, blog posts and Eggs Benedict for breakfast. What percentage of poets are remembered by history? Journalists? Novelists? Professors?
Sturgeon’s Law encourages us to dive a little deeper. If you are a professor of philosophy, you probably compare ChatGPT’s thoughts on Sartre with that of your best students. But, actually, a lot of philosophy students are not among the best (ninety percent?). You should be comparing ChatGPT to your average students, not your best students.
Some of those students cheat.
Think of all the crap writing on the internet: the corporate-speak; the product blurbs for hair removal cream; the local news stories about lost cats; the blog posts about unblocking your drain. Who is going to write that stuff in the future? If you have a job writing crap on the internet, you should be looking for a new job.
All we need to do now is to build an AI that can read the generated copy and we can cut humans out of the loop entirely.
What even is intelligence?
ChatGPT raises questions about what it means to be intelligent. At the moment it just regurgitates the words from other writers. But, then again—who doesn’t?
People will see me and cry
Critics like to point out the limitations of ChatGPT as it is now. Imagine telling 11-year-old Lionel Messi that he is not as good as the grown-ups.
The AI is just getting started.