Special AI Edition

A Deeper Dive into Artificial Intelligence

This is a special edition of Nowism. Once every month or so this newsletter will diverge from its regular format and contents to offer a deeper dive on a single topic.

AI: Final Musings and Reflections (at least for a bit)

The idea of AI has been on my mind for the last eight months, and I keep telling myself to stop talking about it. The internet certainly doesn't need more non-experts giving their unsolicited opinions.

However, in the past couple of weeks, I've seen ideas and predictions about AI put forward at events that are irritating and provocative enough to warrant one more disquisition.

AI won’t take jobs. It will take tasks. 

It seems that the more you know about technology or academia, the less you know about how people actually work. Architects don't spend their days only dreaming up buildings. The predictions about AI taking over jobs are becoming increasingly outlandish and are based on a flawed understanding of what work entails. Our jobs often involve doing many things, most of which cannot be done by AI. See this long piece I wrote to explain more.

Technology doesn’t “take” jobs away, it merely changes how we do them and what we do. Automation has only eliminated one job in the past 60 years - that of elevator operators. However, even then, the core responsibilities of the job shifted to being a doorman, providing the most critical values of security and prestige.

Photo by Boris Stefanik on Unsplash

How real are AI's three key potential benefits for businesses? 

1. Increasing productivity is the most obvious benefit of AI. AI can perform tasks that we used to spend time on, from automation to enhancing the way we work, to outsourcing tasks. This seems like a huge benefit of AI.

2. Increasing knowledge - Can AI effectively use the vast amounts of data we give it to provide us with better insights and help us make superior decisions? This is a question I often ponder. While we have long talked about data-driven insights, I have yet to see one. As we move towards a world of AI-driven insights, I am skeptical about how much data can truly help. Personally, I still prefer observation, intuition, and hypothesis testing.

3. Increasing creativity - Can AI bring inspiration and new imaginative leaps to us? Creativity can seem magical and deeply human, yet it can also be attributed to the wonder of unlikely combinations, unexpected links, and leaps of faith untempered by expectations or fear of looking foolish. While AI has already created a car chassis that diverges from traditional designs, it begs the question: is creativity not more effective when it stems from wonder, empathy, and joy?

Perhaps AI isn't destined to rapidly improve. 

We can be certain that computing power will continue to increase, and that models will ingest more data and utilize more tokens. Neural networks will also be improved. And as computers begin to write code and design chips, the rate of improvement for these systems will increase. However, it's important to note that a better system won't necessarily result in outsized improvements.

To the best of my knowledge, the issue with generative text is that it relies on a paradigm of likely combinations of words, rather than being meaning-oriented. Similarly, generative images are based on knowing what things should look like, but not on relationships or context. For example, it might know that a house has windows and a roof, and that roofs in mountainous regions are often steeply pitched, but it cannot understand how people flow around a building or how building codes translate to 3D space.

While technology can improve within the existing paradigm, it needs to operate in new ways to become truly useful. This means being less about statistics or syntax and more about understanding context and meaning. In other words, we need a different paradigm, not just a better LLM. 

We are suffering from survivorship bias on a massive scale.

AI currently generates around 20 million images per day and OpenAI alone produces 5 billion words. It is not surprising that among this vast amount, there are many cunningly brilliant poems and magical images. These outputs are widely shared, liked, reposted, and celebrated.

It is important to note, however, that we are only seeing the most remarkable outputs. The average AI output from an average person is of a different standard than what is showcased.

Will AI make us lazy and what are the longer-term impacts? 

What are the long-term impacts of having an easy solution at hand? Will it allow us to spend more time thinking about interesting things, or will it make us cognitively lazy? While calculators have made us bad at mental arithmetic, many people don't mind because it was once a specific and not very interesting thing. But what if AI has detrimental effects on our most human qualities? What if makes us poor at judgment, writing, thinking, and dreaming?

What will we use AI for?

The original promise of robots was that they would perform the tasks we dislike, freeing us up to cook, write, paint, and create. At present, AI technology is being used for painting, writing, recipe creation, and film production, while we are left to deal with tax returns, chasing unpaid invoices, and waiting on hold for British Airways to reprice a flight. However, the true value of AI lies in automating tedious and problematic tasks, allowing us to focus our time on our strengths. Technology serves as a lever; its impact is less about the force applied and more about the leverage it provides. Thus, how can we integrate AI more deeply and extensively into our business practices?

Tech should do more.

People are drawn to tech's promise of efficiency, especially consulting companies. We use technology to achieve the same goals with less effort. However, what we learn from AI is that people are more impressed with the fact that it is free and fast, and not that much worse than human output. It is common to see people boasting about how it can be used to create (poor) ads or perform (subpar) SWOT analyses. Some people show that it can be used, rather unfortunately, to send outreach emails to millions of people for free, but the quality is terrible.

My final thought on AI is this: How can we use it to do things better? How can we maintain our level of effort and employment, while achieving far greater results for everyone? As Robert Solow famously observed in 1987, the computer age was everywhere except in productivity statistics, and this remains true today. Year after year, we produce a little more, no matter what technology we have at our disposal. How can we leverage AI to unleash a huge impact on productivity, without reducing jobs, but instead creating wonderful new things that change the world?