Something is missing in the AI growth debate

"Look. The models, they just want to learn. You have to understand this. The models, they just want to learn. "

Ilya sutskever (circa 2015, via Dario Amodei)

The potential economic benefits of AI are still a topic of debate, particularly when it comes to quantifying its mid- and long-term effects. In the end, this is quite a challenging task. Yesterday, 高盛 published a report (https://www.goldmansachs.com/intelligence/pages/gs-research/gen-ai-too-much-spend-too-little-benefit/report.pdf) that addressed this debate. The report emphasizes the various productivity and GDP growth scenarios put forth by a number of analysts. There is a consensus that fully automating AI-enabled operations can result in significant cost reductions over time. Additionally, the cost of new technologies tends to decrease rapidly. However, as is the case with all transitional phases, the present one is characterized by a greater degree of uncertainty than confidence. Chaos theory physicist Norman Harry Packard named this "the edge of chaos", a region of bounded instability that lies between order and disorder. The widespread impact of small local "perturbations", such as fresh AI breakthroughs or market noise, distinguishes this phase. Almost every week, an update from the major players (OpenAI, Meta, Google, Microsoft, Anthropic and so on) is released, announcing a significant advancement in unlocking latent capabilities. Each of these announcements is a small "perturbation" that alters the reality of our predictions by introducing new future possibilities.

"GPT-4 was merely the continuation of a decade of breakneck progress in deep learning. A decade earlier, models could barely identify simple images of cats and dogs; four years earlier, GPT-2 could barely string together semi-plausible sentences. Now we are rapidly saturating all the benchmarks we can come up with. And yet this dramatic progress has merely been the result of consistent trends in scaling up deep learning."

Leopold Aschenbrenner (Ex-OpenAI)

Upon examining the trends in effective computation (figure below), which cover both physical computation and algorithmic efficiencies as drivers of progress, it is possible that we will soon (2027-2028) have models that are as intelligent as PhDs or experts and can work together with us as fellow employees. Perhaps the most critical aspect is whether these AI systems are capable of automating AI research.

The missing piece I: superintelligence

AI advances are one of the pillars of this revolution. However, the other leg is rooted in our ability to generate value from this technical progress, both in the business sector and in society as a whole. But for the cutting-edge research teams, AI progress won’t stop at human-level. Once the transition from Chatbot to Autonomous Agent has been fixed, their strategies are to move toward what is referred to as superintelligence: AI to automate AI research.

If we examine the plot above, we can see that the AI adoption projections (defined by GDP impact in trillions of dollars—PWC, 2024). Direct effective computation projections are nearly proportional to the forecasted AI impact on GDP. The conclusion is that the current projections we are making regarding the impact of AI in the short and medium term (less than 10 years) are actually based on what we currently know. However, we are aware that the current plans extend further. Based on the changes that have occurred in the last four years (since 2020), there is no reason to believe that these medium-term plans will not be fulfilled. From an echo, researchers are exceedingly optimistic about the feasibility of superintelligence and its achievement in less than five years.

"We don’t need to automate everything—just AI research. A common objection to transformative impacts of AGI is that it will be hard for AI to do everything. Look at robotics, for instance, doubters say; that will be a gnarly problem, even if AI is cognitively at the levels of PhDs. Or take automating biology R&D, which might require lots of physical lab-work and human experiments ... But we don’t need robotics—we don’t need many things—for AI to automate AI research. The jobs of AI researchers and engineers at leading labs can be done fully virtually and don’t run into real-world bottlenecks in the same way"

Leopold Aschenbrenner (Ex-OpenAI)

This could potentially?accelerate the current trends of algorithmic progress, which have been a central driver of deep learning progress over the past decade. This would result in the compression of a decade's worth of advancements into a single year. It is possible that there are some bottlenecks that we must address. Will we have a sufficient amount of data to train these models? or will we have sufficient energy to run them? This will be seen in the coming years.

The missing piece II: creativity

“(ChatGPT) It’s a great creativity tool. It’s great at helping you do novel things. It’s not simply doing the same thing cheaper."

Erik Brynjolfsson, Stanford University, 2023

In light of this scenario, it is impossible to predict the advancements in AI in the coming decade. What appears to be clear is that expectations regarding the impact of AI on the GDP are considerably lower.However, there is one thing that we can observe: the automation of the workforce will reach unprecedented levels, revolutionizing the labor market and the current structure of many organizations. In this hypothetical scenario, a critical issue will be addressed: differentiation. We are all currently fascinated with the AI magic that has the potential to increase productivity to levels that have never been seen before. However, how are we going to differentiate ourselves from our competitors? There is no logical reason to assume that our competitors will follow suit if we are able to significantly increase our productivity. And I believe that the key in this context will be creativity, which is to say, teaching AI to do new and unique things, instead of doing the same things cheaply, which will enable us to gain new competitive advantages and attract new customers.