Buyer Beware: Not All AI Is Created Equal

BLOGMar 23, 2021

Matt Smith

We all know the person with a poster of an Albert Einstein quote. Strive not to be a success, but rather be of value, it says. This individual says it inspires them and makes them smarter. I come from US Navy stock, so I became accustomed to hearing the proverbial sea stories, those instructive nuggets of wisdom from experienced veterans, with a good dose of embellishment sprinkled in.

Both these examples remind me of the hype around artificial intelligence (AI), a technology trend that is becoming more and more prevalent today. There are entirely too many companies who represent their core competency as “now powered by AI.” Buyer beware here. It’s one thing to hang a poster or tell sea stories about AI, but quite another to develop and train a truly capable AI engine.

Don’t get me wrong. AI is transforming many workflows and enhancing many of the tasks they perform. But the emergence of “adjective, noun, AI = subscriber retention” is a fallacy; it’s far too easy to claim predictive insight AI capabilities.

Knowing the difference between superficial and authentic AI capabilities stops media and entertainment providers from thinking they are benefitting from something when they really aren’t.

AI Is Like a Body. You’ve Gotta Train, Train, Train to See Results.

Here’s the why. Machine learning is the process through which AI develops autonomous insight-generating capabilities. Humans teach and train the algorithms that comprise ML to build the actual AI. (Unsupervised ML systems then proceed to refine these capabilities, independently of human input.)

ML training is the equivalent of using the correct dumbbell and a repetition of curls to develop a killer bicep. It definitely doesn’t happen by itself. The instructive element is key to functional AI. The better you train it, the more it is capable of. Exceptionally well-trained AI systems can eventually perform tasks unsupervised and derive data patterns, anomalies, and efficiencies that humans may not have envisioned.

Some solutions labeled as AI are trained to perform one task exceptionally well, such as predicting audience content preferences. Others are trained to solve broad business problems. Both types of “AI” fall short for two reasons.

  • The bulging bicep problem: A SaaS solution that’s been retrofitted with a specific AI feature is not an AI platform. It’s developed a single muscle to perfection, while neglecting the interconnected systems that build overall strength. In other words, it skips leg day.
  • The poor conditioning problem: An AI platform designed to serve multiple industries is not adequately trained to understand media and entertainment customers, challenges, and use cases. It’s like training at low altitude for a high-altitude marathon; real-world conditions affect real-world performance.

To be clear, there are some very capable, even powerful AI elements in use today that fall into the categories I describe above. Amazon’s Rekognition is widely used to identify objects within images, with specificity too – when a database of objects such as known faces is provided to the algorithm. This works well for celebrities and previously categorized faces, although this category is not without its controversies because of privacy and bias concerns. Salesforce’s Einstein can predict customer actions, but it performs best for customers who build custom AI models specific to their business environment.

Either of these solutions can be taught to accomplish tasks with specificity, given adequate time and guidance. But to truly go the distance, organizations need AI systems that are already primed to perform the level of industry-specific data analysis required to achieve insight at scale.

Making Media Smart

Applying AI to media and entertainment is the shiny new talisman in the industry today.

With the recent eyebrow-raising headline that global streaming subscriptions surpassed 1.1 billion, providers are scrambling to keep up with all the data-driven elements these services require and answer the questions that arise as a result:

“How many titles are in our library and what groups of people are watching them? What other titles would that strata of subscribers watch?”

“I can see what our revenue looked like for the trailing three months, but I want to advise Wall Street on the next three months’ revenue. Can I predict that?”

All these questions can be answered by properly tuned and trained AI. Authentic AI can also unshackle providers from the myriad of Excel spreadsheets and point tools that fail to help codify the aforementioned data.

Like the Einstein poster and the salty Senior Chief with their sea stories, it turns out there’s a lot of training required to produce meaningful wisdom. Anyone can hang an AI shingle on the door, but buyers would be wise to carefully examine the products they’re investing in.