These Days, Everything Is 'Powered By AI.' Here's How to Tell Hype From Real Innovation. Don't fall for buzzy trends. Make sure the products you're using are actually using the latest tech.
By Harry Guinness Edited by Frances Dodds
This story appears in the July 2024 issue of BIZ Experiences. Subscribe »
Today's tech market is inundated with products that claim to be powered by artificial intelligence. Some are, but some aren't, and it can be hard to differentiate legitimate AI from existing tools hiding behind trendy buzzwords and higher prices.
A big issue is the term "artificial intelligence" itself — it's both poorly defined and encompasses a huge number of subfields. Rather than wasting time debating the true meaning of AI, the best way to spot marketing hype is to understand what modern AI models can do, and what they can't.
To start, you'll need to know whether an AI product falls into one of two common categories: those based on traditional machine learning (ML) and those using newer generative AI (GenAI) models. There are other categories like robotics and computer vision, but you're less likely to find those in commercially available tools — if you need a robot, you're talking to an expert.
Machine learning has been around for decades and powers tools including Google search, Netflix recommendations, spam filters, and market analysis software. Basically, instead of hand-coding a computer program to do a task, you provide it with heaps of data and a few rough guidelines, then allow it to develop its own algorithms that can adapt and learn from human feedback. Adaptability is key. If it's not learning, it's just "machine-learned," says Bob Rogers, CEO of Oii.ai and Intel's former chief data scientist for analytics and AI.
GenAI, on the other hand, has exploded in the past two years. It includes large language models (LLMs) like OpenAI's GPT series, image models like DALL-E 3, and the tools built on top of them, like ChatGPT. GenAI tools can be incredibly powerful in the right situations, but they can't generate truly novel ideas, says Tim Bates, Lenovo's former chief technology officer and a professor of practice at the University of Michigan's College of Innovation and Technology. "GenAI is good at taking information in, analyzing it, and reproducing something," he explains. "But when it comes to the creative part, it only mimics; it doesn't actually produce."
Next, you need to know: Is this worth your money? Custom AI-powered tools are the most expensive, and they're incredibly data-hungry. "If they don't ask you for data, they are probably selling you snake oil," says Rogers. Even if you don't want a fully custom tool, you also don't want to overpay for something that's simply powered by a public tool like GPT-4. Those products will only offer features that are widely available elsewhere (and probably for cheaper). Even the flashiest marketing can't change that.
Related: I Transformed My Company Through AI 15 Years Ago — Here's What I've Learned
Rogers also warned against apps that feature AI-powered text generation when it doesn't make the app any better. If you're using a website builder, it doesn't matter if the stock text is "lorem ipsum" placeholder text or AI-generated nonsense that mentions your business name — because you're going to have to rewrite it anyway if you want it to be effective.
Bates, meanwhile, is leery of any existing AI tool that claims it can replace employees. The error rate and inability of GenAI tools to understand the full context of anything means that allowing them to operate unsupervised could end in disaster. Instead, he sees AI as a productivity booster — something humans can employ selectively to work more efficiently.
And really, that's the crux of it. AI tools are just tools. Any marketing that claims its AI-powered whatever can solve all your problems without your input or being trained on your data is likely hype. But if a product makes boring, nuanced claims? That might be the real deal.
Related: Despite How the Media Portrays It, AI Is Not Really Intelligent. Here's Why.
