<\/p>\n\n\n\n
I have been reading Columbia University\u2019s legal scholar Timothy Wu\u2019s Age of Extraction<\/a> for a second time. I highly recommend this book for those interested in an analysis of the big online platforms (e.g., Google, Amazon, Facebook). With the passage of time and the new life experiences that occur, a second reading can allow different insights and new topics for sharing. I first read Wu\u2019s book soon after it was published and at a time I was also concerned about \u201cfree\u201d social media and the damage done by manipulative algorithms. I was exploring AI, but like Wu made few connections beyond noting AI companies like social media were dominated by large platform companies. It is these insights into AI platforms and possible alternatives I want to explore here.<\/p>\n\n\n\n Wu\u2019s book is partly about the history of our opportunities online and how the original opportunities for individuals to independently share their insights and explore commercial opportunities were eventually taken over by a few major platforms. This transition was explained in terms of first the opportunities platforms provided, first free everything, and then the transition to the manipulation of time and attention in a way that made money for platforms and for the associated boards and stock holders. So many have been captured because of the network effect (I can\u2019t leave because my friends are all there) and while little can be done about this for the majority, Wu offers suggestions for those who are interested in alternatives. <\/p>\n\n\n\n Is AI different or is it still too early?<\/p>\n\n\n\n Upon my second reading, I am more thoroughly steeped in the short history of AI and the opportunity to explore the possibilities both for those in the education community and for myself. AI platforms don\u2019t seem to have followed the same pattern as social media. AI did not grow from the initial interests and activities of hobbyists and individual explorers toward the integration of everyone by the platforms making everything easier and more powerful. The ideas may have originated with the community of academics exploring various approaches to thinking machines over the decades, but the present large language models required the huge financial resources of major corporations and their opportunity to raise massive sums of money necessary to both explore and build the models we can now access. Even the massive budgets of these companies may not be able to feed the beast that is emerging as the way to develop this field. Governments may need to be involved to limit the redundancy that is at the core of the typical capitalistic approach. I explore the ideas of those who believe China has an advantage because the powers that be there can require a more integrated and less competitive approach<\/a>. <\/p>\n\n\n\n The situation with AI has changed a bit. Progress has been made to the point at which individuals and smaller organizations can create and obtain value from the application of smaller models on equipment as inexpensive as a Mac Mini. These opportunities have captured my time and treasure<\/a> for the past few months. Is it possible we are witnessing a reversal of the typical platform dominant pattern Wu describes?<\/p>\n\n\n\n Here is a short description with a few more details describing Wu\u2019s platform model.<\/p>\n\n\n\n Platforms begin by offering genuine convenience. Leading AI companies offer cheap or free assistants, APIs, coding tools, and cloud-hosted models that lower barriers for developers and users.<\/p>\n\n\n\n Dependency follows scale and network effects. A company can become hard to leave once an organization has built workflows, data pipelines, fine-tuning, agents, evaluations, and staff skills around its models and APIs.<\/p>\n\n\n\n The platform becomes a gatekeeper. AI firms and their cloud partners can control access to compute, model capabilities, distribution channels, app ecosystems, and sometimes the terms under which others can build businesses.<\/p>\n\n\n\n Extraction comes after lock-in. This could appear as higher API prices, usage metering, restrictive contracts, preferential access for the platform\u2019s own products, data capture, or fees are imposed on developers and businesses that rely on a given platform.<\/p>\n\n\n\n Concentrated power shapes the whole economy. If a small group controls the main models, chips, cloud infrastructure, and routes to users, it can influence which firms survive, what work is automated, and who receives AI-generated productivity gains. The power to influence government decisions when AI platforms reach a scale of influencing the economy are possible. Note the present argument that these platforms must be supported to match or exceed advances in other countries. <\/p>\n\n\n\n The process sounds scary when described like this. For me, we are already well on our way. <\/p>\n\n\n\n My personal interest has been in the application of AIs analysis and writing skills to the content I have abstracted from my personal reading and augmented with my thoughts as notes. To do this, I must first upload my collected content (a process typically called injestion) and then interact with this content searching for insights and ways of expressing what such insights might reveal. I am much less interested in interacting with AI based on the existing models than the application to my content. This is most frequently described as Retrieval Augmented Generation (RAG). <\/p>\n\n\n\n As long as you can trade time for speed and power, RAG applications can now be executed on your own equipment<\/a>. While not the equivalent of similar approaches using more powerful platform models, the output is certainly useful to my personal work flow and I assume will only become more so if these efforts on personal AI are allowed to continue.<\/p>\n\n\n\n The issue is whether this process will be allowed to flourish. Will the companies developing these smaller models start selling them or not release them at all? Wu doesn\u2019t address such innovations, but the pressure to commercialize that he describes would make this prediction. While these smaller models are available from several sources, I am presently using QWEN<\/a> which is made available by a Chinese company. I suppose this will eventually raise concerns (see the following comment about dangerous uses), but after some trial and error and suggestions from others models from this company just seem to work better for what I am trying to do.<\/p>\n\n\n\n One thing I already see that worries me is the concern that effective AI outside of the control of big corporations opens up the possibility that dangerous actors will use these less controlled tools to create dangerous biological weapons or engage in hacking activities making our reliance on online resources more dangerous. The approach of raising concerns of hypothetical dangers seems a common corporate strategy – typically it is about children and who can best protect them. I see this more about what we believe is the best approach to experimentation and improvement. Present approaches cost out individuals, small companies, and even university researchers. It seems about who will be able to convince Congress of what? <\/p>\n\n\n\n Source:<\/p>\n\n\n\n Wu, T. (2025). The age of extraction: How tech platforms conquered the economy and threaten our future prosperity. Knopf<\/p>\n