Reflections On AI

I am writing this in April 2023, months after the release of the LLM ChatGPT and its successor transformer models as a record of my reflection at this pivotal juncture where AI has finally gone consumer facing with a conversational user interface (CUI). My future self will probably will not agree with all that is said here, but therein lies the purpose of the reflection.

With the advent of a true CUI to the mass market, people are now cognizant of the potential applications of AI and in particular LLMs to daily life and especially knowledge work. Understandably, there are concerns about how jobs will be affected and how work may become dehumanized.?

What should be of interest is the pace of development, which is going to be significantly in excess of what social adaptation is likely to be able to absorb: The coming changes are likely to be measured in months and few years, whereas social adaptation is measured in decades. Even if education were overhauled on the spot, it will be a minimum of 10 years before a new crop of students emerges suited to the likely new milieu.?Education in actuality adapts nowhere near as quickly.

At the moment, it seems probable that much of what we currently understand of knowledge work can and will be automated. There is a significant preponderance of boilerplate and boilerplate-adjacent work done on a daily basis that will initially experience a massive productivity boost, with redundancies to follow as fewer humans are needed to handle the task load. Some entire job families are likely to be made entirely irrelevant even as new roles open up. Others will require significant general and specialist knowledge, significantly raising the floor of expertise in knowledge work.

While it may sound alarmist, the simple fact is that humans (especially scared humans) are highly resistant to change. In a perfect world with classically rational actors, we can reasonably expect that all made redundant will simply take stock of market demands and reskill to fit the new niches and the optimistic picture of human adaptability will allow a transition to AI-assisted work with minimal fuss. However, in a world where humans have emotions and practical needs, and where education takes time even if freely available, the potential for chaos should not be underestimated.

That said, for those who can adapt I remain hopeful. The additional availability of cognitive processing will free humans up to engage in true knowledge creation. No longer will it be sufficient to simply build careers around mechanical tasking in knowledge work. Rather, the roles of humans will adapt to be librarians, editorial and enquirers.?

Librarians, because even with the advent of AI agents and AGI, the formulation of motivations that achieve alignment with humanity will likely still require filtration through humans who have access to and know how to navigate a library of human experiences.?

Editorial, because whereas matters of taste can be computed, what data cannot be captured but incorporated within daily human lives will prove to be useful in augmenting algorithmic derivations of acceptability.

Enquirers, because the simple accumulation of expertise within knowledge work is no longer sufficient when the copilot can now theoretically have free access to all knowledge with perfect recall. Rather, inventors and those inquisitive minds that can sniff out new knowledge and opportunities will thrive.?

Now that the horn has been sounded, where does this leave us? Certainly not more egalitarianism in the immediate term: the immediate gains go to those with the compute resources and the expertise to use them. In the short term, the disruption will be massive and beyond what we can anticipate today. In anything beyond the short term, the geometric nature of the progression will move out of the reach of human intuition and I think short of government intervention the average person will just have to wing it at best. In the long term, assuming madness does not consume the world, I think copiloting will not only be a massive boost to human intellect but will transform the way sentience treats, defines and uses intellect. The harvest of the future shall be reaped by the most adaptable sentience, so may sentience long prevail.

P.S. This was not written by AI. Yet.