It's the dying weeks of 2022, but it feels like the birth of a very exciting time for technology as Artificial Intelligence leaps into the mainstream consciousness - with potentially seismic implications. There's so much happening in the space it's hard to filter the meaningful from the noise. Will ChatGPT make developers obsolete? Will Stable Diffusion kill the art industry? In this series of posts I'll filter the ongoing developments for you, providing a product perspective on what's happening, what we should be worried about, and what are the opportunities.
What is artificial intelligence (AI) - is it machine learning (ML)?
Artificial intelligence is the endeavour to create a computer which can effectively think like a human can. This means interpreting vague inputs, being creative (or at least appearing creative), and continuously learning. Machine learning is how the AI evolves, meaning ML is effectively a subset of AI.
How does it actually work?
Basically, all of the public works available on the internet are used to "train" the AI so that it learns what things are. If you've ever solved a Google reCaptcha, you're actually helping the machine learn!
The process of machine learning can effectively be boiled down to a type of evolution for code, where the machine creates its own logic, humans are used to verify/correct the logic, and the machine keeps the logic that works, and discards the logic that doesn't. For a more in-depth explanation, I recommend this video from CGP grey (~9 mins).
The system is then left with bundles of logic that are incredibly complex (it wasn't a human that wrote the code, it was a machine), but that achieve a given target outcome.
What can it do?
This depends on what "it" you're referring to. If you're referring to chatGPT - it can do things like write you a poem and even write you code - try it out here. If you're referring to stable diffusion or other image-based AI like Dall-e - then it can generate images based on a text prompt.
We'll delve a bit deeper into the practical applications for product managers in upcoming posts, so stay tuned, but if you're eager to see some examples;
- Here's someone who used chatGPT to build a Twitter bot - without knowing programming
- Here's someone using AI to create a marketing plan.
- Here's a service that uses AI to generate profile pictures of you in different styles - currently blowing up the internet - profilepicture.ai
What can it not do? (at least not yet)
When it comes to answering specific questions, these tools are very good at appearing accurate. However, what's happening under the hood is the machine is simply trying to predict what would be the best word or phrase to put next, based on the data it's been trained on.
In this video from MKBHD he talks about "The truth about AI getting creative" (~15 min), where the AI seems to do a fantastic job of writing him a script for an iPhone review video, but on close examination, the AI got a some of the details wrong.
In this post from Intercom, CEO Des Traynor and Director of Machine Learning Fergal Reid discuss how the recent advancements might apply to their realm of customer support. When testing, they found that chatGPT was great at sounding like it's giving the right answer to a support question, but the URL it provides to the user is actually completely incorrect, in that, it doesn't even exist.
Fundamentally, AI in its current state is pretty bad at saying "I don't know". It will do something known in the industry as "hallucination" - where the response you get could sound great, but be factually incorrect. Read about Meta's AI hallucinations here.
What jobs are at risk?
Well, none directly just yet. I think what will happen for now is that these tools will become part of a lot of people's workflow. Developers can write code quicker (see Github copilot), copywriters write ads quicker, and product managers can write PRD's quicker, etc. But as discussed earlier, the veracity and accuracy of the content still needs to be verified by a human.
For now, the AI are shortcuts for the initial phases of creativity, not replacing it wholesale, at least not yet.
What are some opportunities?
There are a few interesting ways this new technology could impact your product or features.
- Search/Questions. ChatGPT has been trained on the open web and can answer conversational questions reasonably well (accuracy aside). You could train an AI on your proprietary data and let your users ask it conversation questions like "email me a report of my sales figures for each month so far this year", for example.
- Assistants. There are already companies out there like builder.ai or gong.io that use AI in an assistive way - essentially what our old friend clippy was trying to do back in the day, but much more effectively! If your product has an onboarding or configuration flow that requires lots of input, perhaps a more conversational approach would work better.
- MaaS (models as a service). As Chamath Palihapitiya mentions in the All-in podcast, models as a service could replace SaaS in industry, where a "single use model that allows you to solve a function" becomes a company's key product offering. Think of an AI that specialises in expense management, or SEO, or social media management, etc. Your company probably has direct access to very proprietary data around your specific market and use case. This could be commoditised into an AI model which you sell (or license out).
Products/services/APIs me and my team should be aware of?
- openai.com - the home of chatGPT
- stableboost.ai - for playing around with image generation
- zapier.com/apps/openai/integrations - use Zapier and openai to create no-code ai tools!
The technology is advancing incredibly fast - and there's a lot of hype around it because, at least on the surface, it appears like it'll put a lot of people out of a job. But it won't, at least not yet. ChatGPT is not accurate enough, still "hallucinating" answers. Image generative tools are interesting and impressive, but they're not doing anything yet that will specifically put masses of people out of a job.
There is a tidal wave of opportunity coming though, and as product people we need to keep track of those opportunities.
Coming next time
A sneak preview at some of the content you can look forward to next time. Get in touch if you've any suggestions or requests on twitter or hit reply to the emails.
- Going deeper on how chatGPT can help product managers in their day-to-day tasks (assuming it comes back online in time - currently experiencing overwhelming traffic!)
- Understanding the GDPR implications of using AI with your user's data
- Answering the question "Do I have to bring the AI in-house or are there APIs I can use?"
Did you find this post helpful? If so, I'd appreciate if you'd share it with your team or anywhere on the internet you think makes sense. 🙏🏼