A Peek into the Future with ChatGTP and DALL-E 2

OpenAI

I first learned about DALL-E from the New York Times, in an April 2022 piece called Meet DALL-E, the A.I. That Draws Anything at Your Command. DALL-E (and it's successor, DALL-E 2) are AI systems that generate images from natural language prompts. Having read about with great interest over the past few months, I was delighted to finally get to test drive DALL-E 2, and the OpenAI chatbot, ChatGTP, this past week. Like the results, my impressions are... mixed. Here's what I feel like I've learned:

 

These technologies are very good at some things, and extremely bad at others.

Like any human-computer interface, the quality of the input (prompt) goes a long way toward determining the quality of the output (image/text). At the same time, they have clear strengths and weaknesses.

 

The content that AI is producing is almost always a starting point, not a finished product.

DALL-E 2 helpfully provides four different images for each prompt. Even then most are spatially distorted, or just plain weird. In my testing, about 1-2 images out of 50 are worth saving. This reveals something important about AI generally: that it's a tool more akin to a screwdriver than a lamp with a genie in it.

 

It's tempting to conceive of these types of AI as "thinking." They aren't.

A great deal of research shows that, when we perceive something to be like a sentient being, we assume that it is. ChatGTP in particular can mimic the patterns of speech so well that it can seem as if it can understand the ideas behind our language. But it doesn't understand the ideas--it only recognizes patterns. With ChatGTP, this manifests itself primarily in humor, which it simply doesn't get. Like this non-sensical example when asked to "tell a joke about a pirate and a bath tub":

Why did the pirate take a bath rub with him on his voyage?

Because he wanted to make sure he had a good buccaneer!

It can replicate the structure of a joke, but has no concept of the meaning of words or what makes something funny. That said, if the user understands this, he or she can leverage AI's pattern recognition and regurgitation to produce something interesting or even useful.

 

AI reflects the good and bad in society, and can be used to produce good and bad content.

These AI products have "learned" from an enormous corpus of content, including content that is racist, sexist, ableist, and includes all kinds of other biases. Because it doesn't understand the content, it can't judge what is good or bad, and because it can't judge, it can't filter out the bad stuff. 

Just as a a household tool like a screwdriver can be used as a weapon, users can manipulate AI into producing ugly and shocking content. There will be a constant arms race between companies creating this software to establish guardrails, and users trying to circumvent them, either out of curiosity or ill intent. Now that the technology is available, this will be a central tension going forward, and the current environment encourages smaller, newer organizations with less to lose to release early, while more established organizations move more slowly to manage risk.

 

 

Summary

These technologies will be hugely disruptive. I can see them revolutionizing many industries as a tool for stimulating creativity and reducing the time needed for humans to perform certain tasks. I believe that many types of content, which is already cheap to produce, will become saturated with mediocre, AI-produced content, as well as intentional misinformation. We are already struggling as a society to agree on basic truths, and this will make it even more difficult. 

Just for fun, here's my favorite thing that ChatGTP has produced:

Prompt

Stand-up comedy routine in King James Bible English

Result

Good evening, fellow pilgrims! I hope thou art having a blessed day.

I must say, it can be quite trying to be a righteous man in these modern times. Everywhere I turn, there are temptations and distractions that would lead me astray.

Take technology, for example. The other day, I was trying to read my Bible on my e-reader, and I found myself more interested in playing Candy Crush than in meditating on the Word of God. It was a humbling experience, let me tell thee.

But I persevered, and eventually I was able to put down my phone and focus on the holy scripture. And it was then that I realized - the devil truly does work through technology!

But I shall not be discouraged, for I know that with the Lord's help, I can resist the temptations of the world. And if thou ever find thyself struggling, just remember - the devil may have an iPhone, but God has a hotline.

Thank thee for thy time, and may the Lord bless and keep thee. Amen.