Is AI (already) in a race to the bottom?

AI technology, which promised massive differentiation and added value for those who could harness it, has become a table-stakes footnote in pitches. Every tech, media, and agency firm claims to incorporate AI into their products or services. Diageo, Progress, eBay, Meta, Kantar, iFit, Snapchat, and Heineken all within the last month have announced AI partnerships or new capabilities.

Meanwhile, countless start-ups are building AI solutions for no one, engineering solutions without market and advertising expertise. The lack of practical applications seems not to matter, with investors pouring billions into speculative AI startups. xAI, Elon Musk’s new AI venture, was valued at $18m this month, before even shipping a product.

The swirling tempest of AI assaulting our algorithm-generated newsfeeds obscures a major disconnect between AI products and the industry problems they purport to solve. Like the ‘proprietary platforms’ of the 2010s, ‘artificial intelligence’ (AI) has become generic.

Same same, but different

Genericism is the state or condition of becoming non-unique, undifferentiated, or common. Most firms have licensed rote solutions like MidJourney or Amazon’s Recognition without even stopping to think about the problem they want to solve. Media and tech firms reskin ChatGPT to demo chatbots, search, and research tools. Start-ups pepper mention of AI throughout their pitches to raise venture capital. Agency leaders let their Microsoft co-pilots languish in the digital ether.

Many of these mass-market AI models suffer from generic training sets, which make their outputs also generic. The companies themselves recognize the risks of undifferentiated training sets and have begun to sign exclusive contracts with media, brands, and data brokerages in the hope of giving their models an edge.

Compound generic outputs with the reality that any company can license Gemini, Midjourney, Sora, Runway, and co-pilot models now to produce work, and AI quickly loses its differentiated sheen. If the active ingredients in your AI products are the same and anyone can license the same set of AI tools, then the downstream AI-infused products and services are also the same.

There’s a reason one pays less than half the price for Ibuprofen versus Advil. General AI is increasingly at risk of becoming generic AI.

Data scientists as miners

According to my business partner and chief operating and information officer, Vibhu Bhan, data science is the practice of extracting value from data. This makes data scientists the miners in the mountains of data.

The various flavors of AI – machine learning, computer vision, speech recognition, natural language processing, and generative – represent a set of tools with which data scientists can apply to extract value from data. These are the pickaxes with which companies equip their miners.

The dichotomy reveals a fundamental problem with AI in marketing and advertising. Marketing departments and agencies are equipping themselves with pickaxes without the miners who know how to use them. Today, generic AI pickaxes litter the floors of marketing organizations while data gold remains unmined.

Meanwhile, teams of data scientists and engineers are mining the wrong mountains. Engineers are grand at building things, but rarely understand industry markets.

In its most reductive form, artificial intelligence can crunch massive quantities of data quickly, find correlations between disparate sets of data points, and generate outputs from those correlations. The AI alchemy that turns lead into gold requires both the tools and the professionals who know how to apply them. The hard part is figuring out what data to ingest and what outputs to produce.

Brands, agencies, and media firms are sitting on massive stockpiles of unmined, raw material – the data training sets that create differentiated data science outputs through the thoughtful application of AI models.

Our initial experiments with banking, entertainment, and gaming brands blend owned, public, and proprietary data feeds into a single data repository. Several large language models then work in coordination to make the data uniform, indexable, and find correlations. Finally, generative AI summarizes the results which are displayed in a variety of cuts for marketers to review.

Administered by data scientists, this blended approach adds value by differentiating both the inputs (data sources) and the models (multiple AI models working together) to create a truly unique output. Brands should invest in partnerships that build a proprietary blend of specialized models and ownable brand data sets.

It’s time for brands to leave the generic AI behind.