thinking through ai by analogy with computer history
The past few days, artificial intelligence has been on my mind, likely due to all the news coming out. We have had a new midjourney model, a new GPT model, and solid efforts to run Stable Diffusion and Stanford's Alpaca models on laptops. Is this it? Is the march of AI consistent and unavoidable, culminating in the singularity?
I thought I might write down some thoughts on it, to help me process it all. Perhaps these might age poorly. Whatever.
At some point this week, I felt that the machine learning models are getting eerily close to artificial general intelligence (an AI which for all purposes acts like a human), and that society would be completely different in 10 years. Perhaps society may even fall apart. I'm not sure about that anymore.
I do think it will have an important effect on our society, but it will be similar to the impact that computers have already had on our society. This analogy can help us think through the effects and to break through some of the hype.
For instance, will jobs just disappear? No. A couple certainly will. Personal assistants have been slowly replaced by personal computers, for instance, and I expect this trend to continue with artificial intelligence. Likely a lot of data entry jobs will disappear, as well as some call center jobs.
By and large, jobs will change to accommodate the new tools. Most major US movies are now animated with computer graphics. The animation industry has changed, but the artists haven't disappeared, they're just doing a different job. Architects almost exclusively use computer-aided design, after a few years of hesitation.
Many jobs will also be created. The advent of computer led to the rise of software engineers and now the rise of AI will lead to the rise of AI engineers. Already specialists in engineering data flow architectures, deploying large AI models, and prompt engineering are in high demand. AI desperately needs its own field of study, separate from computer science.
It's hard to say whether there will be less or more jobs. My feeling based on this reflection is that there may actually be more jobs overall.
Will the jobs be paid the same though? Or will the productivity gains be siphoned into the wallets of corporate leaders?
As you can see in the chart above, productivity and pay have been diverging since 1972. What caused it? Nobody knows for sure, but I'd wager computers played an important role. Suddenly workers could perform more tasks with less training. Each worker could be more productive, but also more... dispensable. It didn't help that computers improved the hiring process itself.
So here we are again, finding more ways to improve worker productivity. Unfortunately, the underlying structures have not changed and the rise of AI may only further amplify the inequities rooted within our society. It doesn't help that the AI is trained on inherently biased human data, so the effects may be borne out not only from how we use these new tools but also in how they behave.
All this talk of productivity, but I wonder too if AI could actually give us more ways to express ourselves. The advent of computers have brought us video games, digital painting, and electronic music. What will AI bring?
It feels like early days, mostly people are using it to remake experiences that we are familiar with, much like early movies were re-enactments of plays. Each iteration of technology has allowed us to capture more of the world and generate more realistic alternatives. I wonder if we can push AI to create deeper experiences for people. At the individual level, perhaps something like video games but with richer settings and higher levels of interaction. AI dungeon is an early example of this. At the group level, we could emulate the feeling of a performer being in the room and interacting with the audience. AI can add richer interactivity to any aspect of our static media.
commercial vs personal
Computers started off huge and would cost millions of dollars to build and maintain. Gradually, we figured out ways to make them smaller and how to manufacture them cheaply. Now we all have access to powerful computers within our homes.
I think something similar will happen with the hardware and software powering artificial intelligence. Currently, OpenAI is leading the deployment of large language models. They have a lead on the model development, as the latest models have a lot of parameters and take up a lot of computing power. However, it's likely that the models are not fully optimized and could be run with much less computing power. At the same time, the computing power available in each computer continues to increase year by year. In time, we will see the rise of personal large language models. It is already happening in some places.
OpenAI feels like the IBM of computer history. They are controlling the technology and trying to sell it to businesses. This is a great strategy and certainly works for them. However, I think we'll see the rise of something like Apple but for AI, a company selling personal AIs directly to people for their own use. It's beyond me to imagine what effect that will have on our society, but I imagine people will experiment even more with applications of AI in their lives.