ChatGPT, Dall-E, Midjourney: What is generative AI all about? How should you respond?
This new generation of generative AI tools (ChatGPT, Dall-E, Midjourney, and Stable Diffusion) has been grabbing headlines inside and outside of tech and business circles. Is generative AI the gamechanger it is claimed to be? How should you respond? Skip ahead for the cheat sheet version of events.
This is ChatGPT responding to my request to write an article about how to respond to generative AI. Click here for the full text of ChatGPT’s article.
So I'll admit it - I'm as excited as the next person about the new generation of Generative AI tools that have come out in the last few weeks and months (okay maybe more excited than most, since AI is my business). Since these tools came out, I have wiled away many hours co-creating art for my website with Dall-E, and chatting with ChatGPT, asking it to create vacation itineraries for me, recommend speech to text tools to me, and numerous other questions of varying seriousness and importance.
This is all very entertaining, but is it really revolutionary?
In a word: YES!
What is Generative AI?
Generative AI is the umbrella term to describe AI that generates its own new content (such as text, images or music) that is similar to human-generated content. It operates by being “trained” on large quantities of human-generated data, and applying AI techniques to create new content from that*.
The most famous example of generative AI is ChatGPT, which generates human-like text content. ChatGPT can provide answers to questions, and take orders to create written outputs such as emails, documents, screenplays, and code (with further applications still being “discovered”). While there are numerous articles being written about ChatGPT’s capabilities and new tricks people have used to get ChatGPT to generate increasingly entertaining outputs, my suggestion is that you spend the 30 seconds required to register a free account and start engaging with it (seriously, it is not that hard, and think of the bragging rights).
Some of the other well-known generative AI tools include:
Dall-E, Midjourney, Stable Diffusion: to generate digital art images from text descriptions.
Lensa AI: to generate digital art images from photos.
The common theme with all of the tools cited (with the exception of Lensa AI) is that you type in some commands or descriptions via text, and out pops a code snippet or an image (whatever you asked for, limited only by your imagination).
Why is it such a gamechanger?
Basically, ChatGPT is a gamechanger because it impacts anything to do with reading and writing. When it’s hooked up to voice to text and text to voice, it also impacts anything to do with speaking and listening as well. This is pretty much everything. (Okay, we still have taste, smell and touch, but they’re working on AI for that as well).
Let’s think about the impact of the reading and writing a bit more deeply - with ChatGPT, we now have an AI that has effectively “read through” all of the data on the internet** and can tell you what it understands of what it has studied. Because of its exhaustive “reading”, it has learnt how we write, and is also now able to write content on command. Think about your day, and how much of it is spent reading and writing, and then you get a better idea of why things are already changing with ChatGPT (cue: mic drop).
To be clear - ChatGPT is not free from errors, and it does generate answers and content that aren’t always correct. Some naysayers may claim that it’s not that much of a gamechanger because we still need to validate what we get from ChatGPT. While that is true, I’d argue that having draft content created by ChatGPT that we can then edit, tweak and personalise is already a gamechanger. For those of us (ahem) who dislike the burden of starting with a blank sheet of paper, having a first draft (however flawed) cuts many hours of effort from our work day.
Furthermore, this is just the raw version of ChatGPT, i.e. the version that can be used directly by typing into your web browser. The next generation is where you hook up ChatGPT with other tools in order to compound the power of ChatGPT (like the voice to text and text to voice translators I mentioned earlier). Again the possibilities are still being explored, but imagine what your workday would look like if you integrated ChatGPT into your office toolkit, and got it to write replies to clients, meeting invitations, email reminders, meeting minutes, and so forth (given Microsoft’s reported plans to invest $10B in ChatGPT, this isn’t completely farfetched).
Did I also mention that ChatGPT is currently free and available to anyone with an internet connection?
Why is it also terrifying?
ChatGPT is exciting because it gives everyone AI powers. It is also terrifying because it gives everyone AI powers without any rules or regulations.
For example, there are no rules about the disclosure and ownership of writing created by ChatGPT. This has already started raising concerns about ChatGPT-generated content claimed as one’s own (e.g. in a classroom or conference setting), with some schools implementing outright bans of ChatGPT due to its “negative impacts on student learning”.
While OpenAI is working on approaches for people to identify ChatGPT-generated content, this doesn’t cover the case where content is co-created with ChatGPT (e.g. where the initial version is created by ChatGPT and the final version is edited by a human). Not only will ChatGPT’s involvement be harder to detect, but there remains questions not only about how much credit to give to ChatGPT, but how much credit we should give the writer who had to ask ChatGPT the right questions, and refine the writing before getting to the finished product.
Bigger picture, it’s also terrifying to think about how ChatGPT could displace writers, journalists and all of us who make our living from writing and generating content. If we look at what’s happening with AI-generated art (due to the availability of tools such as Dall-E), there are definite concerns about how to credit the artists who created the work that AI-generated art is based upon, how to copyright and evaluate the art that is AI-generated (vs human-generated), and protect those whose work can be substituted with AI art. Beyond the initial protests and bans, a nuanced approach to policy and regulation is needed.
Finally, what happens when money comes into the picture? Right now, ChatGPT has been brought out as a free tool since it is in “research preview” mode, but OpenAI has started piloting a paid premium version of ChatGPT. If usage of ChatGPT (and other generative AI tools) becomes limited to those who can pay for it, then how do we make sure that those who can’t aren’t unfairly disadvantaged?
How should I respond?
Learn as much as you can about it (reading articles like this are a good first step). Look at both the opportunities and threats and how they apply to your circumstances.
Be ethical. While I’d encourage experimentation with the tools, if you do decide to use content created by ChatGPT or Dall-E (or others), then please PLEASE be ethical about it. That is, disclose the use of AI and give the proper credit to the tools that you used. This is especially important if you are using ChatGPT for work related items.
Supervise your kids. Technically, children should not be on ChatGPT but reality can sometimes be different. While it does exercise some self-censorship, there is no child-friendly version of it.
Things you can say to sound smart about AI
ChatGPT (and the ability to generate all sorts of words and images) is a complete gamechanger and could be the start of AI revolutionising our lives.
There is no sector/industry/aspect of our lives that can't benefit from being able to read or write more effectively and efficiently.
Most companies (including ours) should be revisiting their business approaches so that they can factor in how generative AI can get integrated into their business and operations.
While I am happy that generative AI is accessible to millions, I have serious concerns that nobody is taking responsibility for governance of these tools.
We need to make sure we are not displacing industries with these tools (i.e. artists who are getting displaced by generative art).
I would never use generative AI without properly crediting the tool AND doing loads of testing and refinements.
I have started using ChatGPT to write draft <emails|documents|code> since I do a lot better at refining existing <emails|documents|code> than creating from scratch.
I am limiting/heavily monitoring my children's usage of ChatGPT because I don't trust its ability to provide a safe environment for my kids.
I am only using Dall-E/Midjourney in limited doses because I am concerned about how these tools displace artists.
I am not using Lensa AI because it is clear that it has not been disparity tested on women.
Footnotes
*This definition was created in collaboration with ChatGPT.
**ChatGPT was trained on data up to 2021.
