Nowadays, artificial intelligence is everywhere.

There are generative AI tools for creating advertisements and AI-powered platforms for managing advertising campaigns. Your refrigerator—and possibly even your toothbrush—allegedly has artificial intelligence built in, at least according to the packaging.

Advertisers looking to implement AI-powered technologies and consumers considering purchasing a new "smart" Oral-B toothbrush share similar concerns: What do AI claims in a product really mean? Does it truly matter?

Are you an advertiser, agency executive, or ad-tech developer seeking to integrate AI and/or ML into your business? This article will highlight key areas to focus on.

AI vs. ML: Understanding the Basics

AI and ML are often confused.

One reason is that "artificial intelligence" has become a more marketable term, and companies with products that primarily rely on ML often label them as AI to appear more attractive.

Machine learning (ML) is the practice of training algorithms to process information and identify patterns on a large scale. AI, on the other hand, is a broader field of software development that simulates advanced intelligence or enables systems to generate new ideas and iterations independently, rather than through human programming.

Many AI-driven products rely on large language models (LLM), such as ChatGPT from OpenAI, Google Gemini, or Claude by Anthropic. These models process massive amounts of content—text, images, and videos—to interpret user queries and generate coherent responses.

The emerging AI space uses a variety of specialized terms: deep learning, neural networks, and virtualization, just to name a few. This complexity can be overwhelming. For example, a generative AI chatbot like ChatGPT is also a neural network—a type of machine learning model that enables software to detect non-obvious patterns across large data sets.

The complexity of neural networks and the connections that computers identify within vast data sets make it difficult to clearly explain how ML-based products work. In other words, it’s often unclear why a generative AI chatbot provided one answer instead of another.

As a practical example, consider an AI-powered advertising product—Google’s Performance Max.

PMax doesn’t simply retarget audiences or find lookalike users; it leverages a neural network to determine a person’s likelihood of converting, even in the absence of obvious signals that a marketing specialist would consider relevant.

However, you cannot ask PMax exactly how it determined a potential target. Even Google engineers working on the product sometimes don’t fully understand why the system makes certain decisions.

What is Bot Traffic?

PubMag • 2024-12-07

AI or ML?

So, the difference between AI and ML lies in the level of human involvement required in their processes.

For example, once an AI-based system is deployed, it does not require human intervention to operate, whereas ML-based systems need users to make decisions.

It is likely that a technology provider using AI or ML is itself relying on another company's AI platform—such as OpenAI. A provider offering human-assisted services and a customizable model is likely using ML, whereas a product delivered as a ready-made solution accessible via an API is a true AI-based solution.

However, AI and ML providers differ significantly from companies that merely use algorithmic modeling and standard datasets to optimize campaigns—even though the latter often market themselves as AI-driven marketing solutions.

You might wonder: Do advertisers evaluating AI or ML technologies really need to dig into such details?

Perhaps they should not overlook these differences. While AI and ML are distinct, both represent a significant advancement over earlier algorithmic products that could never achieve a comparable level of understanding.

"If a vendor's product is based on JavaScript, and their client service specialists do not know how to use Snowflake or data clean rooms, which use ML models for advertising, that’s a red flag," a senior CPG marketing analytics executive told AdExchanger.

Advertisers should also push for transparency regarding which large language models (ChatGPT, Anthropic, Gemini, etc.) a provider is licensing.

Unknown Intelligences

Beyond understanding AI vs. ML terminology, there are other critical aspects for ad-tech vendors, agencies, and advertisers to consider when working with these technologies.

"Advertising technology is an industry where ‘everyone is an analyst, but not everyone is a technical expert,’" Kamakshi Sivaramakrishnan said in an interview with AdExchanger in 2023, when she sold her startup Samooha to Snowflake.

Adopting ML and AI-driven technologies often means learning new programming languages, such as SQL and Google BigQuery. A major part of Samooha’s value to Snowflake was its machine learning functions, which easily translated ad-tech developers’ code into SQL-based query results.

Processing speed also varies between AI and ML applications. While advanced technologies typically work faster, executing complex AI and ML tasks can take significant time.

At a developer Q&A session at the Amazon unBoxed conference in Austin last October, ad-tech engineers complained about long delays when using Amazon Ads ML—an AWS-powered solution known as Amazon Marketing Cloud.

Google’s cloud-based clean room product, Ads Data Hub, suffers from the same issue, they added.

However, what appears to be slow processing is not a flaw, Amazon’s technical managers explained to developers; rather, the delay is an essential part of the decision-making process.

AI and ML models analyze vast amounts of data and can autonomously run simulations as part of their response formation.

“AI products are designed to function more like human intelligence,” Amazon product managers said. “They need a second to think.”

And advertisers are responding to these technologies with the same level of scrutiny, carefully evaluating the AI and ML-driven options available to them.

The Use of AI in Ad Fraud and Disinformation

Digiday • 2024-06-18