Intro to AI

Attention-grabbing artificial intelligence (AI) advancements have made headlines over the past several months, ranging from articles about the ChatGPT language bot passing business and law school exams with flying colors to the potential threat of AI replacing thousands of workers. While there is still much uncertainty surrounding AI and its impact, the concept has existed for decades, if not centuries. The original concept of artificial thought traces back to Greek and Egyptian antiquity, the first scientific paper on neural networks (the architecture of today’s AI) was published in 1943,¹ Alan Turing published the hallowed “Turing test” for machine intelligence in 1950, and Joseph Weizenbaum invented the world’s first natural language processing (NLP) chatbot, ELIZA, in 1966.

At a high level, AI is defined as “a machine’s ability to perform the cognitive functions we associate with human minds, such as perceiving, reasoning, learning, interacting with an environment, problem solving, and even exercising creativity.”² And according to Andreessen Horowitz, AI is “the application of mathematics and software code to teach computers how to understand, synthesize, and generate knowledge in ways similar to how people do it.”

AI applications are prevalent in our daily online interactions and are useful in a variety of fields. We interact with AI whenever we search on Google and receive curated weblinks, see targeted advertisements on Facebook or Instagram, get a Saturday night movie recommendation from Netflix or Hulu, use facial recognition on our phones, or translate a phrase from English to French via Google Translate on summer vacation in Paris. In each of these cases, AI collects and analyzes existing data, detects patterns, and provides a response. The infrastructure and funding for these AI applications, known as analytical AI, have existed since the 2000s and 2010s. The recent and widely publicized round of AI development is focused on generative AI, which is a particular large language model of AI that generates brand new creative content in response to a prompt. Thus, in the following sections, we turn our attention to generative AI and where Brown Brothers Harriman’s (BBH) investment portfolios have exposure to this rapidly developing technology.

Generative AI: What Is It, and What Are Its Applications?

Generative AI is “a category of machine learning where computers can generate original content in response to prompts from the user.”³ Machine learning is a subset of artificial intelligence that takes the latter one step further: While artificial intelligence responds to specific, human commands, such as when we tell Siri to send a text or Alexa to play a specific song, machine learning allows a bot to “learn from data patterns without human direction.” Popular generative AI applications, such as Character.AI and OpenAI’s ChatGPT and DALL-E, rely on advanced machine learning models to scrape the internet and generate original text responses or hyper-realistic digital art instantaneously.

The applications of generative AI are vast, and we expect that they will become increasingly useful in our lives in the coming years. Generative AI can have a profound impact on industries including:

• Content Creation: Automating graphic design and marketing campaigns and completing them at a speed and cost multiple scales cheaper and faster than humans.
• Gaming: Revolutionizing user-generated content (UGC) capabilities in video games (think more customizable games like “Roblox” or “Minecraft”), as well as significantly increasing the number of games produced each year.
• Education: Creating affordable customized curricula and making one-on-one AI tutoring widely available for kindergarten through higher education students.
• Ecommerce: Streamlining internet search and ecommerce shopping discovery capabilities with more curated recommendations; enabling effective, automated customer service and administrative capabilities for companies.
• Biotech: Powering pharmaceutical drug discovery and development.

We believe that these use cases are just the tip of the iceberg. The most significant implication of these applications is that AI now has the power to produce creative, original products – an ability once limited to humans – at a fraction of the cost and time. Therefore, there is a perceived threat that AI may replace some roles in editorial, video production, software development, and graphic design, among others.

The nearby table shows a tentative timeline for how several models of generative AI may further develop over the next decade and how currently human-led tasks could be made more efficient, or even displaced, by generative AI in the future.

The opportunities for value creation in AI do not end with the mentioned examples, which are only the possible capabilities of the final applications of generative AI. Beneath the application layer, there are multiple levels of the generative AI “technology stack,” and each level is crucial to making the end-user facing generative AI application possible.

For example, generative AI developers depend on the hardware (chips) and cloud infrastructure at the bottom of the AI tech stack to compile and run complex AI code. As such, the recent rush of funding and interest in generative AI also creates significant trickledown economics for companies that support the bottom of the stack, including NVIDIA, Taiwan Semiconductors, and Amazon (via Amazon Web Services). 

Preliminary Generative AI Tech Stack

AI Investment Opportunities and Risks

BBH’s Investment Research Group (IRG) has spent countless hours considering the opportunities and threats that AI presents, with particular focus on our investment universe. We believe that AI opportunities fall into three main categories across the AI technology stack:

• Tech AI software (the AI applications)
• Tech hardware (chips and semiconductors)
• Companies seeking to leverage AI to streamline their processes, such as U.S.-based retail companies that have integrated AI into their supply chains or those that have partnered with AI developers

Within the first category, some of the most notable frontrunners in the AI application development space include software companies Microsoft, Alphabet (Google), and Amazon, which we own through several of our public equity managers.

One of those frontrunners, Google, has been investing heavily in AI for years and declared itself as an “AI-first” company six years ago – a year after Google’s DeepMind AI unit beat the world’s best human player at the board game Go. Most of Google’s products, whether their flagship Search function, Maps, Advertising, YouTube, Gmail, Chrome, Assistant, Photos, or Translate, would not exist today if it weren’t for the company’s pioneering work in traditional AI. In fact, the “GPT” in ChatGPT stands for Generative Pretrained Transformer, a technology brought to the world by Google in 2018 with its release of BERT (Bidirectional Encoder Representations from Transformers).

In addition to having exposure via public equity managers, BBH clients who are invested in venture capital have additional exposure to AI application developers and AI-powered companies. We have invested with one of the leading venture capitalists in the AI space, and the company has funded several promising AI-related startups.

AI-related portfolio companies owned by invested clients include:

• OpenAI: OpenAI is a San Francisco-based AI research laboratory with products including ChatGPT-4 and DALL-E.
• Character.AI: Character.AI is a platform that allows a user to create and chat with intelligent and personalized AI.
• Genesis Therapeutics: Genesis Therapeutics is an AI and machine learning-powered drug discovery company.
• Anduril Industries: Anduril is a startup that provides AI-supported defense hardware and software for U.S. and allied defense agencies.
• AKASA: AKASA is a technology platform that leverages AI and machine learning to provide end-to-end automation services for repetitive administrative tasks in the healthcare industry so providers can focus on patient care.
• Replicate: Replicate is a software platform that helps make AI usable at scale by using cloud technology to keep all open-source AI models available and easy to use from one place, which draws in AI app developers.
• Rewind AI: Rewind AI is a secure video recorder and smart database that has advanced compression and automatic speech recognition search capabilities that facilitates searching for specific content.

Through our public equity managers, we also have exposure to the second layer of the tech stack, including semiconductor designers and manufacturers who generate value by producing the hardware necessary for AI computing, such as NVIDIA, Taiwan Semiconductors, KLA, ASML, and Texas Instruments.

According to a recently published Wall Street Journal article, NVIDIA is one such semiconductor designer who is experiencing significant tailwinds from recent AI developments. The company reported a projected 64% year-over-year jump in revenue (quarterly revenue totaled $11 billion) in the second quarter of 2023, its highest quarterly sales to date, attributable to high demand for its chips in AI data centers.⁴ While the long tail of AI’s impact on the semiconductor industry is still uncertain, NVIDIA’s recent strength may have positive implications for Taiwan Semiconductor, which manufactures the chips that NVIDIA designs and sells.  

Within the third category, many of our portfolio companies have the potential to develop, explore, or incorporate AI technology into their business models in the coming years, and many have already begun doing so.

Mastercard, one of our top holdings across client portfolios, recently announced that it is using AI technology to enhance fraud protection. Another top holding, Clarivate, an analytics company that operates a collection of subscription-based services, uses generative AI to ensure it provides customers with the highest-quality integrated public and proprietary content and insights.

But with opportunity comes risk. Several of the primary risks that broad-based AI and automation may pose include:

• Unsafe data systems (such as those that may harbor and misuse consumer data)
• Discrimination by algorithms (such as those used in hiring systems) on the basis of race, color, ethnicity, sex, etc.
• Undisclosed use of automated systems or failure to inform a user how an automated system works
• An inability for a user to opt out of an automated system for a human alternative

Companies are aware of these risks, and many are trying to mitigate them. Unfortunately, some generative AI applications have been trained on nonproprietary data, which leads to questions about data quality, bias, and ownership. Any well-trained data analyst knows that the quality of outputs is only as good as the quality of the inputs.

We believe that companies that are utilizing generative AI technology trained on proprietary data will have a long-term advantage in developing high-quality products that obtain significant customer adoption rates. Doing so, in our minds, is a thoughtful approach to ensuring security and high-quality output as well as limiting questions around authorship.

For example, Clarivate’s AI applications are trained on the company’s proprietary data assets, which are expertly curated and interconnected. Similarly, Adobe’s (another portfolio company) Firefly, a creative generative AI engine that allows users to leverage AI to create and enhance images within the suite of Adobe products, is trained using Adobe’s proprietary images.

We are also watching carefully the state of regulation around AI. Indeed, in the United States, there is not yet a framework for AI regulation at the federal level. While it is incumbent upon federal agencies to develop their respective plans for such regulation, only five of the 41 major federal agencies had established agency-level AI plans as of December 2022. The agencies that have promulgated AI policies include the Department of Energy, Department of Health and Human Services, Department of Veteran Affairs, Environmental Protection Agency, and USAID.

Nonetheless, the Biden administration appears to be supportive of establishing federal AI regulation, demonstrated by their publishing of the Blueprint for an AI Bill of Rights in October 2022. The Blueprint for an AI Bill of Rights provides a roadmap for how the government may be able to protect citizens “whenever automated systems can meaningfully impact the public’s rights, opportunities, or access to critical needs,” and particularly in cases where the risks outlined in this article may materialize.

While it is too early to say when wholesale U.S. regulation on AI will roll out – or if it will take shape at all, given a currently gridlocked Congress and a looming presidential election – an ideal AI policy in the U.S. may look something like the 100-page “AI Act” approved by the European Parliament, which bans any applications that exceed an outlined threshold for risk, requires some vendors to obtain licenses before usage of the AI technology within the EU, and charges fines for any developers who fail to follow regulations.

As AI continues to evolve, it is important to understand the threat that exists for companies that fail to consider how they might leverage AI technology to improve their operations. We are acutely aware of AI’s risks and disruptive potential, but we believe that the technology will primarily unlock enhanced productivity and likely result in increased profitability. Those companies that do not adapt the technology while their competitors successfully do will likely be left behind.

Conclusion

We are excited about the opportunities surrounding AI, and especially those presented by generative AI. We have meaningful exposure to AI companies within our portfolios and believe that if AI changes the world as much as many expect, our portfolios should benefit from this paradigm shift.

As with any new technology, we give much thought to both the opportunities and the risks. We have spent many hours discussing and trying to understand how AI may threaten existing businesses and business models. As it turns out, our investment approach, which calls for investing in high-quality companies with sustainable business models and pricing power run by strong management teams, can provide a layer of protection against the risks of automation and AI. This is particularly the case when a company has established itself as an industry standard, such as our portfolio companies Mastercard, FICO, and S&P Global, among many others. Other companies, such as our portfolio companies Google and Microsoft, are staying ahead of the curve by investing in and further developing AI technology themselves.

Indeed, we can see that several of our largest public equity and venture capital holdings have management teams that presciently recognized the opportunity (and threat) of AI and sought to incorporate the technology into their business models. We should note that part of our rationale behind adding exposure to top-performing venture capital managers was to capture the value being formed by nonpublic companies creating and improving interesting technologies like AI. We are pleased to see venture capital playing its intended role in our clients’ portfolios.

Finally, we are also thinking through ways we might be able to integrate AI to improve our own investment decision-making process. We are in the early stages of thinking through how we might do so, but it is worth highlighting that we also seek to apply technological innovation in ways that can improve our own business.