Gen AI: Too Much Spend, Too Little Benefit?

Background

Last month, Goldman Sachs published a research report questioning whether the investment in generative AI is excessive, relative to its benefits.

Despite attempting to present a balanced argument, the article was swiftly picked up by media outlets to portray an overall tone of skepticism. Before I share my personal opinion, let’s have ChatGPT summarise the 31-page report for us.

Report Breakdown

Written by ChatGPT

Skeptical Views on AI Investment

1. Daron Acemoglu (MIT) Estimates only a quarter of AI-exposed tasks will be cost-effective to automate within the next 10 years. Predicts AI will impact less than 5% of all tasks and will increase US productivity by only 0.5% and GDP growth by 0.9% cumulatively over the next decade. Argues that AI's transformative impacts on new tasks and products are not guaranteed, and advancements may not happen as quickly or impressively as expected.
2. Jim Covello (GS Head of Global Equity Research) Believes AI must solve complex problems to justify the high costs, which it isn't designed to do. Points out that unlike the internet, AI has not yet enabled low-cost solutions to disrupt high-cost ones. Skeptical that AI’s costs will decline sufficiently to make large-scale automation affordable due to the complexity of building critical inputs like GPU chips. Doubts AI will significantly boost company valuations, as efficiency gains might be competed away and the path to revenue increase is unclear.

Optimistic Views on AI Investment

1. Joseph Briggs (GS Senior Global Economist) Estimates generative AI will ultimately automate 25% of work tasks, raising US productivity by 9% and GDP growth by 6.1% cumulatively over the next decade. Believes AI's potential for cost savings and the likelihood of long-term cost declines will drive more AI automation. Incorporates the potential for labor reallocation and new task creation into productivity estimates, drawing from historical records of technological innovation.
2. Kash Rangan and Eric Sheridan (GS Analysts) Remain optimistic about AI's long-term potential, even without a clear "killer application" yet. Argue that current AI capex spend as a share of revenues is not markedly different from previous tech investment cycles. Note that investors reward companies tying AI spending to revenue generation and see potential returns from this AI capex cycle as more promising than previous cycles.

Constraints and Concerns

1. Chip Shortages Toshiya Hari, Anmol Makkar, and David Balaban (GS US Semiconductor Analysts) argue that chip shortages, especially in High-Bandwidth Memory technology and Chip-on-Wafer-on-Substrate packaging, will constrain AI growth in the next few years.
2. Power Supply Issues Carly Davenport and Alberto Gandolfi (GS Utilities Analysts) expect AI-driven power demand to increase significantly, posing a challenge for the aged US power grid. Brian Janous (Cloverleaf Infrastructure) warns that substantial investments in power infrastructure are needed, but these won't happen quickly due to regulatory and supply chain constraints, potentially leading to a power crunch that could limit AI growth.

Report Conclusion

• There is a significant divide in opinions regarding whether the massive investment in generative AI will pay off.
• Skeptics highlight the high costs, slow expected productivity gains, and potential supply constraints.
• Optimists focus on long-term potential, historical precedents of technological cost declines, and broader economic benefits.
• The outcome largely depends on future technological advancements, cost reductions, and overcoming supply constraints.

What Do I Think?

Personal Bias

You can call me an optimist. I’ve been a lifelong mathematics and science fan, having pursued STEM throughout secondary school, through university, up until today in my role as an AI Engineer. I’m not a humanities student, I’m not an artist, I thrive on logic, data, and innovation.

This background naturally inclines me towards a belief in the transformative power of technology. So, when discussing the potential of generative AI, my perspective is inevitably shaped by my faith in scientific progress and the possibilities it unlocks. Keep this bias in mind as you consider my views on the investment in generative AI.

Is AI Overhyped?

Absolutely.

The vast number of people who discuss AI or claim to understand its technical intricacies far exceeds those who truly do. This disparity creates a situation where consumers' expectations of AI are inflated by speculation rather than grounded understanding.

In the short term, this has led to governments, large institutions, and companies worldwide placing immense hopes on AI automation as a driver of economic growth. Does that sound overhyped to you?

Is AI Worth The Investment?

After highlighting the overhyped nature of AI, you might wonder how it can still be a worthwhile investment. To understand this, let's rewind to November 30, 2022, when ChatGPT first appeared on our screens.

The significance of this conversational assistant cannot be overstated. Many of us have used ChatGPT for various tasks at work — I even used it to write 38.89% of this article. Realising that ChatGPT can perform even a fraction of your job with just one click can be daunting, yet it can also be liberating.

The true value of AI extends beyond immediate economic figures or financial reports, as highlighted in the GS Report. The time saved by Generative AI isn't often reflected in these metrics, nor are the 'higher value' tasks it enables.

The greatest value of AI lies in its potential to transform education, assessment, and our overall mindset. Despite anxiety about job displacement, the threat of automation has already driven significant changes in these areas.

Generative AI excels at rote memorisation and imitation, while humans excel at critical thinking, creativity, and emotional intelligence. 

Yet, our society continues to assess intelligence based on rote memorisation, leading to highly structured and repetitive tasks.         

AI is worth the investment if we are prepared to rethink our approach, embrace the challenge, and consider AI's potential rather than its current limitations.

In the long term, AI is absolutely worth the investment.

You may notice that my response above contained no reference to economic facts and figures, which, admittedly, is not my forte. Instead, I've presented the perspective of a technology optimist, outlining the idealistic theory behind AI's potential. I've shared my belief that AI will inevitably prove its worth in the long term. But why aren't the numbers adding up right now, and why might they continue to fall short in the coming years?

Why Isn’t Your AI Investment Providing Short-Term Value?

• Disconnect Between AI Champions and Technical Reality: AI projects are often spearheaded by ‘champions’ from non-technical backgrounds, leading to a gap between their expectations and the technical challenges involved. This disconnect can result in overhyped capabilities and missed opportunities to pursue more impactful solutions.
• Challenges with AI Maturity and Data Preparation: AI maturity presents a significant hurdle. Much of my time is spent preparing data to be compatible with AI models. While companies like OpenAI aim to streamline this process, their solutions often fall short for many organisations. In practice, a large portion of your investment in Generative AI ends up going toward data engineering rather than the AI model itself.
• Misaligned Productivity Expectations: Advertisements often claim that AI software can boost productivity by 5x, but these claims are often idealistic. In reality, while AI can enhance productivity, employees frequently spread this increased output over a longer period due to workplace expectations or personal motivation. The real challenge lies in human factors—motivation, engagement, and adaptation to new technologies.

Summary

• Develop a Strategic Approach: Dedicate time to coming up with the best idea.
• Assess and Prepare Your Data: Understand and address the data preparation needs.
• Consider Human Factors: Evaluate how human behaviour and expectations impact productivity.

By addressing these areas, you can better understand why your AI investments may not be yielding the short-term value you anticipated.