There is a world market for maybe five computers. – Thomas Watson, CEO of IBM in 1943
Knowledge is power and so it tends to be hoarded. Experts in any field rarely want people to understand what they do, and generally enjoy putting people down. – Ted Nelson
Do you remember when we would say “power to the people?” Thatʼs what I’m doing, I’m giving power to the people. Iʼm building a computer that every person can put on their desktop, and Iʼm going to get rid of the high price of the mainframes. – Steve Jobs
Bigger is not better in computing technology. Small tech is where the actual power has been over the last 50 years.
When we democratize computing power (or power in general), we create the seeds for actual change. In doing so, we have harnessed the power of millions of imaginations. That is where the actual power lies.
Every technology goes through a phase where people can’t wrap their heads around what the new technology is for, how it works, or how it changes our social and economic paradigms. This is where we are now with Generative AI.
In the 90s, the problem was getting people to think of the Internet as anything beyond an “online billboard.” This line of thinking only extended the existing models of the roadside billboard or the display ad in a newspaper or magazine.
In another example, the 1950s saw a palpable excitement over the possibilities of computing technology. You saw this in science fiction story after science fiction story as they expanded on the power of postwar computing. However, it also reflected in the dominance of IBM’s thinking through its CEO Thomas Watson, who famously predicted in 1943 that there wouldn’t ever be a market for over five computers in the world..
This reflected industrial thinking. In that paradigm, the bigger you made it and the more it was consolidated, the more powerful it would be.
The Industrial Age dehumanized people by making them part of vast machines and systems of production. Computing reversed this logic as Douglas Engelbart, Ted Nelson, and other visionaries in the 1960s foresaw. Their revolutionary insight was to humanize computing by making it smaller, more accessible, and more intuitive.
However, if you had asked a journalist, businessperson, or ordinary consumer in 1970 how they envisioned the future of computing, it wouldn’t have been that different from Thomas Watson’s 1943 prediction. They did not see what people like Wozniak and Jobs were building in their garages.
Our vision of AI seems based on the big models of social media and search engines of yesterday instead of imagining how the technological world might become more humanized. National competition and a focus on “big” AI further divorces ordinary people from technology. Lack of access to this infrastructure will exclude those who can’t afford it.
This isn’t necessarily how it will play out. AI will get smaller and more efficient.
Right now it doesn’t seem like this kind of world, but that is a vision of technology that we are being sold like IBM sold mainframes in the 1970s. Companies want to consolidate power because it gives them control over market share and profit.
We’ve seen massive consolidation occur through the tech giants that emerged from Silicon Valley over the last two decades. Open AI is mimicking the business model established by Facebook and Google.
This phenomenon could be transitory. However, it is a narrative that the media, itself until recently monolithic, can wrap its head around and sell.
The story of big is easier to tell than the quiet revolutions that are taking place on desktops and phones across the country. Our storytellers don’t understand the technology, much less the story. The panic of Monday, January 27th, was driven by ill-informed media reporting.
We feel compelled to follow narratives when we do not understand what’s going on. This is particularly true with a fast-moving technology like AI. We almost need AI to understand AI.
I talk with a broad range of people, some of whom are very technical. Even those technical people who understand large language models developed in an environment dominated by giants like Google, Apple, Amazon, and Microsoft. Similarly, IBM, Honeywell, and UNIVAC dominated these narratives in the 1960s.
AI is software. It doesn’t suffer from the same sort of physical limitations that developing microchips, hard drives, and displays involves. Therefore, it has even fewer physical constraints about how it might evolve.
We look at most AI models today and associate them with the brute force models developed by companies like OpenAI, Google, and Perplexity. We think they have all the answers because they are dominating the narrative like IBM did in the 60s.
This is only one version of reality, however. For instance, if you combine the computing power of all of our PCs with our mobile devices in the US, it comes out to around 130 Exaflops. Aggregating all the commercial data centers in the country only gets you around 20 Exaflops or 1/6 as much computing as is in the hands of individuals. (I had ChatGPT crunch these numbers for me.)
More importantly, you vastly increase the number of minds at play with a disaggregated model. PCs produced most of the billionaires of today. AI will produce the billionaires of tomorrow. We don’t want those to be the same people. The ideas and products they produce will be far less transformative than otherwise.
Small footprint information models are already beginning to prove their utility. I’ve already begun experimenting with a Retrieval-Augmented Generation (RAG) system, Google’s Notebook LM. Other lightweight models are proliferating, particularly in the Open Source world, like Mistral and Llama 3, that don’t require massive computing power to operate.
We are reaching the limits of data collection for AI models as well. We have scraped the internet for our existing AI models. Doing it repeatedly is only going to yield marginal results.
Eventually, someone is going to develop a comprehensive open data set any AI model can access. If you don’t have to reinvent the wheel with your model every time, this also reduces size and power requirements.
Just like with the PC, as the overhead gets smaller, it becomes more accessible to everyone. Sure there will still be some need for large complex server farms to crack the most complicated problems we’re looking at, so Nvidia-style chips will still have a home at the bleeding edge.
However, most of us won’t need that kind of horsepower. We already have so much power in our hands (120 Exaflops, to be exact) that this power will migrate from the large corporations and their black boxes.
AI disrupts information silos. If the source of any company’s power is its database, AI will erode that power. AI will either find the information elsewhere or triangulate on what’s missing and develop information that way. That is a losing corporate strategy.
The Information Age is over. We are now entering the Connection Age. To see that, it pays to look at what humans are doing, not corporations. Humanized technology will always win out in the end.