Democratizing AI, Bit by Bit

Democratizing AI, Bit by Bit

"Democratization of storytelling”, is how Co-Founder Alejandro Matamala, framed Runway AI’s vision 

“Democratization of technology”, is how Managing Director (North America) Darren Mowry, reframed it for Google Cloud 

Last year, Google made $652m/day aggregating, selling our data. This year, their focus has pivoted to the ‘democratizing of technology’, via open-source. 

Thomas Kurian (CEO of Google Cloud), shared the following in an interview on Stratechery, “we are seeing significant acceleration in how customers want to use digital tools and AI to transform their core business…. we have an open architecture, which gives them the ability to use these services, but across a range of different models. Some from Google, obviously Gemini, but also from partners.”

Using the term, democratization, seems fitting for AI. Google aggregates data and automates workflows to maximize profit - AIM 2.0 (AI Marketing) is set up to democratize data via disruption of automation.

A for-profit business’s goal is generally to maximize their profit against a given time horizon - spend efficiency by revenue over cost, also known as return-on-ad-spend (ROAS). 

Google Ads is set up as an auction. The advertiser sets up a campaign with a daily budget, said campaign then participates in real-time bidding (RTB) and the advertiser pays-per-click (PPC), also known as cost-per-click (CPC), regardless of the bid strategy (this is significant - Appendix). Bid strategies for campaigns focus towards maximizing or hitting a target CPC, ROAS, or CPA (cost-per-acquisition aka lead & Appendix).

The problem is that Google wants to maximize their revenue as well. Google does this by promoting automation through campaign types and bid strategies (Appendix). The DOJ has referred to the RTB framework as a “black-box”. Unsurprisingly, large-language models (LLMs) have also been described as a black-box.

AIM 2.0 will be a free application released this quarter. The software will use an api call to fetch your data, run it against the trained model for industry/competitor specific context, and then output a bid target which will maximize your ROAS or conversion volume against given time-horizon. 

AIMs Framework

The term AI is thrown around to the point that it has become laughable; it's provocative, gets the investors going. Every company is trying to automate workflows via fine-tuning a model, and then proudly share it with investors, clients, on earnings calls, etc 

I would compare this to figuring out how to dribble a basketball and then sharing that you can play the sport. 

Rather than fine-tuning (current competitor best practice), AIM leverages a custom, pre-processed, hybrid-LLM trained against datasets to properly compete with Google's blackbox, and which are complimented with Google's data-driven best practices, deep exploration via deep reinforcement learning and policy methods including RLHF, RLOO, etc (Appendix) - this enables robustness through reasoning, essential for game theory, allowing AIM to assign weights appropriately against the noise; dynamics which need to be understood in strategy.

AIM 2.0 is engineered to extract pertinent historical Google Ads data, analyze seasonality, current-market dynamics, auction insights, demographic indicators, time of day and device efficiencies, etc - most importantly, with reason. Basically everything a strategist should do everyday for every keyword, product, ad group, etc (if it were humanly possible, while being profitable - Appendix).

Simply put, when we fine-tune models, Gemini forgets which color vanilla ice cream is (Appendix). 

Why this is important: Value Stacks

Rather than anthropomorphising, let's imagine we are a machine (weird) - our operating system (OS) is tied to an environment, bounded by its weighted values (Appendix). The human OS also is bound by its weighted values, which we can stack and prioritize when needed. For example, many Americans are single-issue voters. If you're pro-life, then a presidential candidate's stance on other issues becomes mostly irrelevant if they don't outweigh your pro-life positioning. The pro-life parameter takes priority in your value stack.

Now let's apply the value stack framework to aggregators (Appendix) like Google and Meta. As mentioned earlier, for-profit organizations aim to generate profit. Historically, companies like Google earned most of their revenue from advertisers who have the technical expertise to use their ad platform. While billions of users indirectly support aggregators, advertisers and agencies have been their most important customers, implying that you are the product (Appendix).

Another framework to consider is the AI Tech Value Stack (refer to the image by Michael Parekh). This stack depicts a shift from the traditional model of aggregating user data (Aggregation Theory - Appendix) to aggregating AI chips. This is evidenced by Project Stargate (the OpenAI & Microsoft joint venture) and recent US executive orders (Appendix). This shift allows Google and Meta to emulate Amazon's approach, which is accredited with both integrating infrastructure, while maintaining aggregation of data, network effect, to maintain their moat (Appendix). 

Key Takeaway

By prioritizing chips to support their infrastructure, along with their current aggregated frameworks (Aggregation theory - Appendix), Google and Meta are strategically focused with AI top-of-mind, and well placed to capture the disruption in data-driven industries (Appendix).

I shared this knowledge graph in mid-February. At the time, I painfully outlined how tech giants are monopolizing data and framing it as a privacy concern. On Tuesday (4/9/24), Meta closed up 46% YTD. This flywheel propelling data monopolization is essentially further increasing barriers to entry thresholds, or more simply, expanding tech giants moats (Appendix).

Conclusion

Sam Altman, CEO of OpenAI, has been reported sharing, "95% of what marketers use agencies, strategists, and creative professionals for today will easily, nearly instantly and at almost no cost be handled by the AI.” 

Considering this is Google’s main customer-base (Appendix), Google is ironically searching, searching for new customers, and have placed their bits on AI start-ups, as clearly reflected through the theme of this year's Google Cloud Next.

Google CEO Sindar Pichai kicked off the event by sharing, “ we’ve known for a while now that AI would transform every industry and company including our own. That’s why we've been building AI infrastructure for over a decade….”

If the customer wants a faster horse, build them a car; a framework around innovation accredited to Henry Ford. The idea of a car was so foreign to us, to understand how to frame its output, we created the term horse-power. 

AI will allow people with agency (Appendix) to actualize their vision. Runway co-founder shared a surprise for the team when they were initially catching traction. Runway's ideal customer profile (ICP - Appendix) had been intended for enterprise-level creatives; rather though, saw their revenue being actualized by non-enterprise, non-creatives. The Google Managing Director responded to this with something I found even more intriguing. 

Historically, tech valuations were based on the software, & more importantly, output capacity from the volume of employees who can build it out at scale. Today valuations for tech (AI) companies are based on vision, & implications of the industry which is impacted by its disruption; video cameras were originally developed to film plays (paraphrasing).

Perplexity is democratizing answers. Runway is democratizing storytelling. Google is democratizing technology. & AIM intends to democratize data. 

Let’s make marketing great again.