AI Is About Abundance

Lately I am experiencing a bit of AI sentiment whiplash.

At Newcomer’s terrific Cerebral Valley event in November, gradations of opinion seemed to range from “AI is great” to “AI is incredible”.

Conversely, a few weeks later, while discussing AI with my family over the holidays, the overriding feeling was angst that AI was going to replace large swaths of employment for the benefit of higher corporate profits at the expense of regular people.

I’ve previously discussed some of my general skepticism around AI automating away employment en masse, but it’s not enough to take a loss minimization stance that AI won’t be a travesty for employment – I’d like to elucidate why I am genuinely excited for the coming AI-infused world.

The core primitive of AI is replacing human cognition with computation that can be executed by machines. Human cognition is limited by the number of people and how much mental labor they are willing to do. I can do simple addition all day, but if I was asked to I would probably refuse or eventually get distracted.

Computation by machines is limited only by computing power and energy. Moore’s law in semiconductors has delivered decades of exponential growth in computing power and efficiency, and even as it slows in the traditional sense, AI-focused accelerators and improvements in software are pushing the limits of further gains for the specific computation needed for AI. In short, machine computation grows much faster than human cognition and looks likely to continue to do so.

MLPerf Training benchmarks compared to Moore’s Law (Yellow Line)

So what is the point of AI? Why be excited about AI? Because AI is about abundance.

Abundance of Expertise

For somewhere between free and $20/month, anyone in the world can access distilled expertise in a panoply of topics: personal training advice, career coaching, SAT prep, stress management tips, the list goes on. Not only is the information available, but users can interact with the expert and receive customized responses. Yes, LLMs sometimes hallucinate and give incorrect answers, but the alternative for most is generic advice or going without help. The best approach is if an expert can guide the AI, allowing AI to multiply the capabilities of experts to serve many instead of few.

Tutoring is the perfect example. It’s long been known that one on one tutoring helps students to outperform peers. Human tutors are expensive and not universally available, but LLMs are also capable of interacting with students and adapting to their learning needs. A recent randomized, controlled trial in Edo, Nigeria looked at secondary school students studying English. Over 6 weeks, the treatment group interacted with Microsoft co-pilot, guided by teachers, while the control group received only instruction from teachers. The results are stunning. In just six weeks, students in the treatment group outperformed peers by about 0.3 standard deviations, equivalent to nearly two years of typical learning.

Abundance of Labor

There are many tasks we simply do not want to do because they are boring, repetitive, or culturally viewed as low status. I had an internship in high school in which I was tasked to go to a specific monthly report from the government, find a statistic, and record it in a spreadsheet to generate one graphic in a research report. Today, those hours of very boring work are a single AI prompt.

The importance here is not saving interns time (or companies money from hiring interns), but rather that anyone interested in this data can now access it versus only those that could pay for research prepared by those who can hire interns. We are entering a new era for those with curiosity and agency. Compute will be all that is needed to access somewhat capable intellectual labor. Competition will no longer be constrained by the capital to pay for intellectual labor but rather be decided on the vision and coordination of labor.

Abundance of Thought

In physical sciences, as in many other fields, breakthroughs are fundamentally limited by skilled practitioners applying thought and effort. Start with a hypothesis, design and execute an experiment, review the results, and if it doesn’t work, start over again. AI systems that incorporate physical and biological knowledge can transform this into a computational problem.

We have already seen this with AlphaFold. A tiny fraction of the billions of known protein sequences were determined via tremendous experimental effort. AlphaFold developed computation methods for predicting 3D protein structures based solely on its amino acid sequence. Knowing protein structures makes drug design more efficient, helps better understand diseases like Alzheimer’s and broadly advances our understanding of human biology.

AI systems are limited (for now) by the ability to generate results in the physical world. Even so, humans working with these systems intentionally can maximize their discovery capabilities. Researchers, knowing that an AI system can analyze across a massive search space, can design experiments to generate a large amount of general data to be searched and refined computationally. Thus, a series of iterative experiments can become a single experiment with iterative analysis done quickly via computation.

Moreso than almost any other field, scientific and biological discoveries directly improve humanity. I will happily argue why I think improving corporate productivity is positive for humanity, but few need to be convinced that eliminating childhood cancer or curing Alzheimer’s is a goal worth striving for.

Abundance of Patience

This is a bit difficult to quantify, but I believe it is just as meaningful as other forms of abundance. When a human is at the other end of a chain of questions, they have finite patience to deal with repeated questions and social dynamics can prevent an individual from asking for needed clarification. AI systems, on the other hand, have infinite patience to repeat information, drill down deeper on a topic, or refresh someone’s memory.

In the future, we could foresee a child sitting with an LLM that can answer every possible question about dinosaurs and repeat the process every day, over and over again, setting them on a path to a career in science. Or patients in early-stages of memory issues can interact with a system that allows them to refresh daily on details of their life and their family without the embarrassment of asking. Or perhaps an employee, afraid of surfacing their lack of knowledge on a subject, can run an action by a system trained on internal documents to prevent an error before it happens. Each of these has a small impact, but over a global population and a long time frame, the cumulative effect can be immense.

Abundance of Accessibility

In accepting his Oscar, Parasite director Bong Joong Ho incisively noted that “once you overcome the one-inch-tall barrier of foreign language subtitles, you will be introduced to so many more amazing films”. What is true in film, applies nearly universally. Current generation AI models already allow for near real-time translation of many languages. While Google Translate and other machine translation has existed for decades, recent advances in LLMs improve not just the accuracy, but also the nuance and context.

Collapsing the language barrier will unlock knowledge and potential connections for billions worldwide. International business will get easier. Research will become more global. Communities of all kinds – intellectual, cultural, social – will find new connections.

Machine translation capability over time, as measured by BLEU score (Source)

I believe AI is ultimately a technology that will expand what’s possible. It will amplify expertise, accelerate discovery, democratize access to knowledge, and remove barriers that have historically limited human potential. While fears of job displacement and corporate exploitation are understandable, history shows that technological progress, when harnessed thoughtfully, leads to greater prosperity and opportunity on the whole.

The key challenge ahead is not whether AI will create abundance – it will – but how we choose to use and distribute that prosperity and progress. I believe the future will belong to those who embrace AI as a tool for empowerment rather than a threat to be feared. Instead of resisting this transformation, we should focus on shaping it to build a world where intelligence, creativity, and opportunity are more widely available than ever before.