Discover more from Jamaal’s Letter
What I'm Telling CEOs and Boards about Artificial Intelligence
The first killer app, imminent regulation, and the gold in the AI gold rush.
I’ve been in more than a dozen board rooms in the eight months since ChatGPT launched and quickly became the fastest-growing consumer app in history. In every executive suite, the anxiety and intrigue are palpable. “What does the unfolding promise of AI — particularly generative AI — mean for our organization?” Most boards sent urgent mandates to their executive teams for an AI plan. One year into what many are calling the fourth technological revolution, this is what I’ve been telling board directors and CEOs about AI.
Your Company's First Killer App: A Tailored Chatbot
Predictions abound about which industries will be AI's winners and losers.
I’m still finalizing my beliefs, but chief among them is that every organization will use ChatGPT-like software fine-tuned with your organization's data. Imagine having a “YourCompanyGPT” fluent in everything: your company's market positioning, customers, financial and marketing data, and beyond. Now imagine having that data at your fingertips in natural language.
Picture a researcher at a multinational pharmaceutical company querying its in-house large-language model (LLM) -- trained on proprietary research, clinical trial data, drug formulation and manufacturing processes, FDA guidelines, and patient feedback -- "Are there any insights from our previous successful formulations that could help with the stability issue I'm facing with our new anti-diabetic drug?"
Or a customer service representative at a large tech company asking its internal LLM -- trained on proprietary technical manuals, customer service logs, FAQs, and product descriptions -- "What are the troubleshooting steps we typically recommend when a customer is stuck on the firewall permissions step during the installation process of our latest software?"
Soon, apps like this will be ubiquitous. The questions that should consume you now are: How does something like this fit into your strategic plans? What data would your internal LLM need? Who would use it, and why? Do you have the capabilities and inclination to build this internally or buy it from the market?
First AI Regulation on the Radar: Copyright Protection
I came of age during the height of internet music piracy. Long before the days of Spotify and Apple Music, I used AudioGalaxy and the speedy T-1 line from my University of Missouri dorm room to download every song I could ever think of in fractions of seconds.
Under siege from unregulated peer-to-peer file sharing, the music industry drove much of the copyright framework we see on the internet today. It used a combination of lobbying, lawsuits, licensing, technology, and international copyright treaties to reign in unregulated music sharing.
I think that the same sort of legal, lobbying, and licensing assault used by “Big Music” in the 2000s will soon be underway for the makers of large language models who train on everything publicly available on the internet.
It seems morally, legally, and economically untenable for LLMs to train on data and profit from it while failing to credit or compensate its owners.
The European Union has taken the lead with its EU AI Act, but most of the globe's lawmaking bodies are still formulating their responses. As of this writing, US copyright law offers no protection for works created without human intervention. As a slew of AI copyright cases are working their way through the courts, I expect that to change.
For example, comedian Sarah Silverman has joined a class-action lawsuit against OpenAI and Meta for copyright infringement, alleging that the companies' AI language models were trained on copyrighted materials from her books without her knowledge or consent.
If your company generates publicly available intellectual property, are you comfortable with giants like Google or Microsoft profiting off your data without your consent or compensation? If not, what's your strategy for copyright protection?
Foster Experimentation, But AI Job Takeover Is Not Imminent
Within 90 days of ChatGPT’s release, companies such as Samsung, Verizon, and JPMorgan Chase banned the use of generative AI tools on work devices. School districts and universities from New York City to India followed suit.
Larger organizations, with more to lose, might understandably be more hesitant than intrigued, but an outright AI ban is a grave misstep.
Ignoring AI not only stifles potential benefits but also impedes understanding of AI's risks. The harsh reality is that AI remains a vast unknown, and only through promoting experimentation, observation, and iteration can we hope to understand it better.
Currently, AI enhances productivity but cannot replace complex human jobs. AI may help value an acquisition target, but the back-and-forth with the target's legal counsel to finalize the deal remains beyond AI's grasp.
The mantra "AI won’t take your job, but someone using AI will" remains relevant.
Smart companies will encourage AI experimentation among employees, learn from their experiences, and invest in what works. Instead of denying AI's existence, prepare your organization to thrive in an AI-infused world.
Your Proprietary Data: The New Gold
In 2006, British data scientist Clive Humby pronounced, "Data is the new oil." Humby was alluding to the customer data gleaned from British retail titan Tesco's Clubcard, the world's first mass-customization loyalty program.
In 2006, today’s AI tools were but gleams in the eyes of deep-learning researchers, but now, even more so than then, data is gold in this AI gold rush.
The quality of training data will differentiate large language models. Poor or mediocre data will yield the same results, while unique, valuable, and proprietary data will offer unique and valuable insights.
With the barriers to entry — money, and talent — being surmountable, models will be everywhere. The data quality will set them apart.
Imagine a top-tier cancer hospital using its proprietary collection of radiological scans to build a model that outperforms human doctors in detecting cancers.
What proprietary data sets might your organization possess? The more unique, the higher the potential. Think innovatively across data types. This AI revolution is multimodal — your organization’s data set could be textual, auditory, visual, or something else entirely. Harness its power wisely.