AI Stages of grief, and a few predictions.
AI Stages of grief, and a few predictions.
Here are some of my gut feelings about what is going to happen next in the AI world.
By this point, you probably noticed that I am pretty interested in recent developments in AI. The technology has evolved quite a bit since I wrote about it 1.5 years ago in Blog Post - GPT-3, Artificial Intelligence, and what are they up to?
If you are sick of my takes on AI, you are not in luck: As of this week, I joined a completely new team inside Automattic focused on Applied Artificial Intelligence. My (new) responsibility is to help orient the company with these transformative capabilities, explore how they can be integrated into our products to help democratize publishing and business and explore where the future of human-computer interfaces is heading.
I cannot express how rapid the progress feels right now if you are working with what we broadly call "AI" right now. Since the future is getting released every week, we might play with predicting it.
One of the most impacted professions will be journalists
Large Language Models are very good at producing plausible-sounding word salads. It's a shame that a lot of journalism has recently moved to do exactly the same, as only one of those can win the race to the bottom.
Mediocre journalism is already seeing an impact (with Buzzfeed augmenting their editorial process with GPT), but I think we're moving towards on-demand content produced just for you.
By definition, we hear a lot about opinions and the challenges journalists face. The perceived societal impact of AI will be something we'll hear quite a lot.
Programmers will be one of the most hit as well
Programmers have been in a privileged position recently. I think AI and other factors will end the era of my unreasonably compensated class.
Interest rates: The Venture Capital model works roughly this way: You take a lot of money and pour it into 10 companies. Each one of these companies hires a bunch of programmers, and they try to 100x your investment. Nine out of 10 of these companies go bankrupt, one makes you a handsome profit, and the cycle continues. Now that money (credit) is more expensive due to higher interest rates, and there are other, safer ways to compound that money (bonds, sponsored by interest rates), 5 companies get funded instead o 10. Which means you need half as many programmers to work in them.
Layoffs: Many tech companies have overhired massively during COVID, leading to recent layoffs in Google, Meta, Amazon, and many others. In his article titled "Meta Layoffs", Bran connects it to the previous point as well: "Is it too much to hope that the last decade’s intense infatuation with middle management was a ZIRP phenomenon."
AI. AI is massively helpful in a programmer’s job.
Most of my programmer friends (including myself) are pretty anxious about AI mostly due to absolute uncertainty of what it means for our jobs. I guess we absolutely deserve it for working on this.
We all need to learn to delegate
The discussion about "AI Taking jobs" focuses on each job definition like it is etched in stone: "I do X, and that is one part of my job that AI will never do. Hence I will not be impacted."
The reality will be that we'll all get very cheap but not-too-smart "interns". We will all have to find ways to split our work into different chunks, delegate some to our AI interns, and package results into quality output.
This, of course, means that no one job will be taken over by AI, but in the aggregate, quite a bit of work will be performed by our GPT cousins.
But humanity finds a way to stay employed. Even though we continually make it easier to manage and organize information, more and more people are used to managing new swaths of information.
I am currently half-way through a new paper from the University of Pennsylvania and OpenAI examining the job impact of Large Language Models. While it predicts most of us being affected in some way, it seems like the highest-earners will be most affected:
The projected effects span all wage levels, with higher-income jobs potentially facing greater exposure to LLM capabilities and LLM-powered software.
I have to say it is a bit reassuring. The absolute worst outcome would be further inequality between the blue and white-collar classes.
The AI stages of grief
As a thought exercise, I used the stages of grief to map how we are going to come to accept our AI-assisted future:
Denial. First, the power of the current Large Language Models was ridiculed and denied. “This is a toy, cannot be trusted”, “not good for anything in particular”.
Anger - We are entering this stage. A group of AI researchers is already calling to pause all training and progress while the government of Italy temporarily banned ChatGPT.
Bargaining - I am sure we will see Sam Altman (OpenAI CEO) in congressional hearings.
Depression- AI Doomers are only getting started. We will see discourse about the dangers of Large Language Models comparable to the Climate Debate.
Acceptance - We will accept the productivity boost from better tooling and more intuitive software and move about our lives, never again re-examining the impressive innovation like we don't stop to appreciate electricity.
I cannot say these stages map perfectly - we might as well see Anger/Bargaining/Depression happening simultaneously.
The next AI innovation wave is going to be led by Apple
Currently, Microsoft and OpenAI have really good press. I think Apple is going to overtake them with AI models running on the device (your phone) itself.
You will probably see me writing about AI quite a bit in the coming weeks, as putting these thoughts on the screen helps me structure my own understanding and deal with anxiety.
In the meantime, I am in New York enjoying a pastrami sandwich from Katz Delicatessen. If you have other culinary recommendations for Big Apple, I would happily try them out!
Apple. Plausible.