GEN AI: TOO MUCH SPEND, TOO LITTLE BENEFIT?

In the rapidly and ever-evolving landscape of artificial intelligence, voices of skepticism like those of Jim Covello from Goldman Sach, which question the return on investment and the potential of AI to solve complex problems, are increasingly being challenged by the latest advancements in the field. Companies at the forefront, such as OpenAI, are not only contradicting these doubts but are also setting new benchmarks for what AI can achieve across multiple sectors.

Jim Covello’s cautionary stance on AI investment hinges on the belief that the technology yields too little benefit for the financial outlay it requires. This perspective, rooted in a traditional assessment of technological gains, overlooks the transformative strides being made by AI technologies that are now redefining capabilities in industries as varied as healthcare, finance, and creative fields.

Recent breakthroughs from OpenAI, such as the introduction of GPT-4, have showcased an unprecedented ability to handle complex, multidisciplinary issues with a finesse that is fast approaching human-level competence. This goes a long way in debunking the myth that AI technologies are merely costly experiments rather than practical utilities that can streamline operations, enhance decision-making, and even pioneer new creative realms with technologies like DALL-E 3.

Moreover, the continuous enhancement in AI models and the integration of these technologies into commercial applications demonstrate not only a reduction in costs over time but also a significant potential for creating new business models and revenue streams that were previously unimagined. As such, contrary to Covello’s assertions, the evidence increasingly supports a view of AI as a crucial driver of future economic and social transformations.

This shall we say, rebuttal seeks to explore and counter Covello’s skepticism by presenting recent developments in AI, illustrating how they are not only proving him wrong but are also paving the way for a new era where AI’s potential is limited only by our imagination. By examining the facts and shifting the narrative, I aim to provide a clearer picture of AI’s burgeoning role in shaping the next generation of technological advancement.

With that said, let’s get into it.

Jim Covello’s skepticism about AI’s potential return on investment and ability to solve complex problems appears increasingly outdated in light of recent developments and announcements from leading AI companies like OpenAI.

OpenAI has made significant strides in AI capabilities that directly challenge Covello’s assertions. For instance, GPT-4, released in March 2023, demonstrated remarkable problem-solving abilities across various domains, including law, medicine, and programming. This contradicts the notion that AI isn’t built to solve complex problems.

Furthermore, OpenAI’s continuous improvements and new model releases show that AI’s capabilities are rapidly expanding. The company has introduced specialized models like GPT-4 Vision and DALL-E 3, which have pushed the boundaries of AI in image analysis and generation. These advancements indicate that AI is not only solving complex problems but also creating new possibilities in various fields.

Contrary to Covello’s claim that AI technology is costly without providing low-cost solutions, many AI applications are already offering affordable alternatives to traditional high-cost solutions. For example, AI-powered legal research tools and medical diagnosis assistants are providing cost-effective alternatives to time-consuming manual processes.

The argument that AI’s costs will never decline enough to make automation affordable seems short-sighted. The tech industry has a long history of rapidly decreasing costs as technologies mature. While current AI infrastructure is expensive, ongoing research and development are likely to lead to more efficient and cost-effective solutions over time.

Covello’s doubt about AI boosting company valuations overlooks the transformative potential of AI in creating new business models and revenue streams. Companies that successfully integrate AI are not just gaining efficiency; they’re often developing entirely new products and services that can significantly impact their market position and valuation.

Lastly, the claim that models trained on historical data can’t replicate humans’ most valuable capabilities ignores the rapid progress in AI’s ability to generate novel ideas and solutions. Recent models have shown impressive creativity and problem-solving skills that go beyond simple pattern recognition of historical data.

It’s worth noting that Covello’s perspective, coming from a traditional financial institution like Goldman Sachs, may be influenced by the potential disruption AI poses to the finance industry. AI’s capabilities in data analysis, risk assessment, and predictive modeling could significantly impact the role of traditional financial analysts and potentially reduce the need for some of Goldman Sachs’ services. This context makes Covello’s skepticism somewhat self-serving, as widespread adoption of advanced AI could challenge the current business models of firms like Goldman Sachs.

In conclusion, while Covello raises some valid concerns about the current state of AI technology, his arguments seem to underestimate the rapid pace of AI advancement and its potential to revolutionize various industries, including finance. As AI continues to evolve, it’s likely to address many of the limitations Covello highlights, potentially reshaping the landscape of global business and technology in ways that may indeed challenge traditional financial institutions.

Daron Acemoglu’s assessment of AI’s economic impact timeline appears overly conservative when compared to recent announcements and progress from companies like Tesla and NVIDIA, particularly in the field of robotics and AI.

Elon Musk’s recent statements about Tesla’s Optimus robot directly challenge Acemoglu’s assertion that AI won’t materially improve tasks requiring real-world interaction anytime soon. During Tesla’s Q1 2024 conference call, Musk announced that Optimus is making substantial progress and could be available for external purchase as early as the end of 2025. This timeline is significantly shorter than Acemoglu’s “10 years” prediction for transformative changes.

Musk stated, “We are able to do simple lab tasks, or at least I should say, factory tasks in the lab. We do think we will have Optimus in limited production in the natural factory itself, doing useful tasks before the end of this year.” This indicates that AI-powered robots are already capable of performing real-world tasks in controlled environments, with practical applications expected in the very near future.

Tesla’s latest demonstrations of Optimus show significant advancements in autonomous learning, adaptation, and task execution. The robot has been shown sorting objects autonomously, self-calibrating its limbs, and even performing yoga poses. These capabilities directly contradict Acemoglu’s claim that AI won’t be able to materially improve tasks requiring real-world interaction anytime soon.

Furthermore, Musk emphasized the potential economic impact of Optimus, stating, “I think Optimus will be more valuable than everything else combined. Because if you’ve got a sentient humanoid robot that is able to navigate reality and do tasks at request, there is no meaningful limit to the size of the economy.” This vision of AI’s economic potential is far more ambitious and immediate than Acemoglu’s conservative outlook.

NVIDIA, a key player in AI hardware, has also made announcements that suggest a faster timeline for AI’s economic impact. The company’s advancements in GPU technology and AI-specific hardware are accelerating the development and deployment of AI applications across various industries, including robotics.

In conclusion, while Acemoglu’s cautious approach has merit, the rapid advancements and concrete timelines provided by industry leaders like Tesla and NVIDIA suggest that the economic impacts of AI, particularly in robotics and real-world task automation, may materialize much sooner than his 10-year prediction. The progress demonstrated by Optimus and other AI-powered robots indicates that we are on the cusp of significant transformations in various industries, challenging the notion that AI’s impact will be limited to pure mental tasks in the near future.

As Allison Nathan states:  While AI technology cannot perform many complex tasks well today—let alone in a cost-effective manner—the historical record suggests that as technologies evolve, they both improve and become less costly. Won’t AI technology follow a similar pattern?

The assertion that AI models have nearly exhausted the internet’s data and that we’re approaching a data scarcity is overstated and fails to account for several key factors.

Firstly, while large language models have indeed consumed vast amounts of internet data, the internet is not a static resource. New content is continuously generated at an unprecedented rate, with over 2.5 quintillion bytes of data created daily. This constant influx of fresh data provides an ongoing source of training material for AI models.

Secondly, the development of synthetic data generation is not just a far-fetched concept, but a rapidly advancing field with significant potential. Synthetic data can be created to fill gaps in existing datasets, simulate rare scenarios, or even generate entirely new training examples. This approach is already being used successfully in various domains, including computer vision and natural language processing.

Moreover, the idea that we will soon have AI building models using PhD-level knowledge is not just speculation, but a reality that’s already unfolding. The paper “Foundation Models for Decision Making: Problems, Methods, and Opportunities” (https://arxiv.org/html/2311.02462v2) highlights how AI models are increasingly being used to make complex decisions and solve problems that previously required high-level human expertise. This includes tasks in scientific research, engineering, and even AI model development itself.

The paper discusses how foundation models are being applied to decision-making tasks across various domains, demonstrating capabilities that rival or even surpass human experts. For instance, these models are being used in scientific discovery, drug design, and complex problem-solving scenarios that traditionally required PhD-level knowledge.

Furthermore, the concept of AI developing AI is already being explored. Researchers are working on meta-learning algorithms that can optimize the architecture and training process of other AI models. This could lead to a scenario where AI systems are not just consuming data but actively participating in their own evolution and improvement.

The paper also touches on the potential for AI to generate its own training data through simulation and self-play, a technique that has already shown remarkable success in domains like game playing (e.g., AlphaGo and MuZero). This approach could potentially address the concerns about data scarcity by creating virtually unlimited, high-quality training data.

In conclusion, while it’s true that AI has consumed a significant portion of available internet data, the field is rapidly evolving beyond simple data consumption. The development of synthetic data generation techniques, the application of AI to complex decision-making tasks, and the potential for AI to participate in its own development suggest that we are not facing an imminent data crisis, but rather entering a new phase of AI evolution that could lead to exponential advancements in the field.

The Gloves are Off:

Goldman Sachs, we need to talk. Your skepticism about AI, particularly as expressed by Jim Covello and Daron Acemoglu, feels less like cautious analysis and more like the self-serving hand-wringing of a traditional financial institution nervous about losing its grip on the market. Let’s break this down.

1. The “High Cost, No Benefit” Argument

Jim Covello’s Take: AI tech is costly and must solve complex problems to justify its expense. He questions the ability of AI to deliver low-cost solutions.

Seriously, Jim? Have you been living under a rock? OpenAI’s GPT-4 and its successors have made waves in industries from law to medicine to coding. GPT-4 is solving complex problems daily. Remember when the internet was new? It started pricey too. But innovation drives cost down. AI is on the same trajectory. It’s not about the cost now; it’s about the potential and inevitable cost reductions. Moore’s law, anyone?

2. AI’s Economic Impact is “Limited”

Daron Acemoglu’s Take: AI will only boost productivity by 0.5% and GDP by 0.9% over the next decade.

Daron, are you just a glass-half-empty kind of guy? Look at the rapid advancements. Tesla’s Optimus robot is making real-world progress and could hit markets by 2025. That’s not pie-in-the-sky dreaming—that’s tomorrow’s reality. And let’s not forget NVIDIA’s advancements in AI hardware. They’re not just dabbling; they’re revolutionizing.

3. The “No Killer Application” Fallacy

Covello and Acemoglu’s Shared View: There’s no clear, transformative application of AI yet.

Patience, fellas. Every major tech cycle starts with infrastructure before the killer apps emerge. Remember the early days of the internet and smartphones? The apps came, and they changed the world. AI is no different. We’re seeing incredible use cases in creative fields, software development, and customer service already. The “killer app” will come, and it will blow your minds.

4. The Competitive Pressure Dilemma

Covello’s Concern: Companies are spending massively on AI due to competitive pressure without clear returns.

Welcome to capitalism, Jim. Companies invest in future potential. The ones that hesitated during the internet boom were left behind. AI is a game-changer, and the smart money knows it. Look at the investment in AI infrastructure—it’s laying the groundwork for future dominance. It’s not irrational exuberance; it’s strategic foresight.

5. AI Will Never Match Human Cognitive Abilities

Covello’s Skepticism: AI will never replicate human nuance and outlier understanding.

Ah, the old “humans are special” argument. Yes, humans are amazing, but AI is catching up faster than you think. The advancements in natural language processing and problem-solving are staggering. AI might not replace us, but it will augment our abilities in unprecedented ways. Dismissing this potential is shortsighted.

6. The Data Scarcity Myth

Acemoglu’s View: AI models have nearly exhausted internet data and we’re approaching data scarcity.

Daron, the internet is not a static library. New data is generated daily at an exponential rate. Plus, synthetic data generation is a thing. AI can create its own training data, simulating scenarios and generating novel solutions. Data scarcity? Please, we’re drowning in data.

7. The Fear of AI’s Dark Side

Acemoglu’s Warning: AI could be misused, leading to negative societal impacts.

Every technology has a dark side. It’s not a reason to halt progress; it’s a call to shape it responsibly. Regulation and ethical frameworks are essential, but they should guide innovation, not stifle it. Fear-mongering about dystopian futures is unproductive.

Just Plain Wrong!

Allison Nathan: “…But even in its infancy, the internet was a low-cost technology solution that enabled e-commerce to replace costly incumbent solutions….”

The Internet did indeed have massive costs in terms of connectivity infrastructure, including the installation of vast amounts of undersea fiber optic cables and other foundational elements needed to create the global Internet as we know it today. Here’s why:
Submarine cable infrastructure: The Internet relies heavily on an extensive network of submarine fiber optic cables that span the world’s oceans. These cables are extremely expensive to manufacture, install, and maintain.

High costs of cable laying: Submarine cables can cost upwards of $40,000 per mile, with trans-oceanic cables often reaching hundreds of millions of dollars. Recent trans-Atlantic cable projects have cost between $250-$300 million, while trans-Pacific cables can cost $300-$400 million.

Massive investment: The submarine cable industry has seen billions of dollars in investment. For example, between 1999 and 2001 alone, there was over $22 billion invested in privately financed cable projects.

Ongoing expansion: The development of submarine cable infrastructure is ongoing, with continued investment to expand connectivity to developing regions. For instance, projects like the Connected Coast in British Columbia, costing $45.4 million, aim to bring high-speed internet to remote communities.

Additional infrastructure: Beyond undersea cables, the Internet requires extensive land-based infrastructure, including data centers, routing equipment, and last-mile connections, all of which contribute to the overall cost.

Maintenance and upgrades: The existing infrastructure requires continuous maintenance and periodic upgrades to keep pace with growing bandwidth demands, adding to the ongoing costs.
While it’s true that the Internet has enabled many cost-effective solutions and business models, the underlying infrastructure that makes it possible has required and continues to require substantial investment. The comparison in the query between AI and the Internet overlooks the significant costs associated with building and maintaining the Internet’s physical infrastructure.

Embrace the Future

Goldman Sachs, your cautious approach to AI might resonate with your risk-averse clientele, but let’s not confuse caution with wisdom. AI is here, and it’s evolving faster than any of us anticipated. The future belongs to those who embrace it, not those who cling to the past. So, how about a little less skepticism and a bit more vision? The world is moving forward, with or without you.

Worth Noting:
Artificial Intelligence (AI) stands poised to revolutionize business models across various industries, and companies like Goldman Sachs are no exception. As a leading global investment bank, Goldman Sachs has already integrated AI into its operations, using advanced algorithms for trading, risk management, and client services. However, this deep entrenchment of AI in their business model can lead to inherent biases when forming opinions on AI’s broader societal impact.

While AI can drive efficiency, enhance decision-making, and unlock new revenue streams, it also poses challenges such as job displacement and ethical concerns. Given their significant investment in AI, firms like Goldman Sachs might emphasize the benefits while downplaying potential drawbacks. This selective perspective can shape public and regulatory discourse, influencing how AI’s role in the economy and society is perceived and managed. Consequently, it’s crucial to critically assess the narratives put forth by such stakeholders to ensure a balanced understanding of AI’s multifaceted impact on the world.