It seems like you can’t turn anywhere these days without hearing about AI. Everywhere you look, technology companies are unveiling new ideas, and within every sector, businesses are looking for ways to jump on trend.

However, for every groundbreaking new use of AI, there are countless empty promises and hollow technologies. So how can you spot the differences? How can you find truly transformative tech in a sea of flashy illusions?

Why AI Feels Like Magic

I recommend you start by finding that line between hype and reality. AI carries incredible promise. However, the version of AI in our media and imagination may not quite align with reality—at least not yet.

Our cultural fascination with AI is deeply ingrained. We’re prewired to glom onto anything that promises breakthroughs, especially when the technology is just beyond the average person’s understanding. And it’s hard to ignore the real and practical appeal of automation and intelligence at scale.

But I think our expectations of AI are perhaps too high, and I find that many people trust the technology in a way that makes them vulnerable to a type of “magical thinking.”

ChatGPT is a perfect example. The generative AI tool has excellent grammar, an authoritative voice and is eager to please its human user—so eager that it has been known to provide false information and even urge users toward acts of violence and self-harm.

Through our own internal testing with popular generative AI tools like ChatGPT, we’ve seen firsthand its tendency toward hallucinations. ChatGPT-generated text often contains incorrect information and fabricated citations. Simply stated, it can be a poor research tool.

The Reality Check: What AI Can Actually Do

This isn’t meant as a takedown of ChatGPT. Even with its flaws, it has practical value. But we must understand the potential shortcomings of AI before we can seize this value.

Whether you know it or not, you’ve already interacted with this technology in countless real-world settings. Customer service chatbots handle your inquiries. Marketing professionals use AI tools to generate ad copy and personalize ad campaigns.

Furthermore, healthcare networks use natural language processing (NLP) for medical transcription, diagnostic imaging and patient communication. And financial services use machine learning algorithms for risk modeling, financial forecasting and detecting fraud.

The Limitations Of Current AI

In spite of these practical applications, there are limits to how much faith you should place in AI:

False Confidence

Generative AI sounds authoritative. But there are currently few guardrails in place to keep AI-generated content honest. Again, ChatGPT has a regular tendency toward hallucinations.

Lack Of Tacit Knowledge

You can “train” AI with information and rules, but you can’t give it real-world, industry-specific experience. This experience is what makes us intuitive business leaders, nimble attorneys and groundbreaking scientists.

So while generative AI can provide generalized content, when you probe for deeper nuance, context and insight, you’ll usually find responses increasingly repetitive, unsatisfactory and generic.

Speed Over Quality

With costs rising for labor, energy and resources, businesses are understandably enthusiastic about the speed and scalability of generative AI. However, these priorities often overshadow quality control, fact-checking and depth. I worry that we’re learning to accept mediocrity in exchange for fast results.

Inherent Bias

AI systems are defined by their training. If there are racial, gendered or linguistic biases in the data or rules, they will be reflected in the way AI tools behave. These biases can lead to unfair outcomes and deepen existing inequalities.

Spotting AI Snake Oil

How can you tell when some “AI-powered” innovations are too good to be true? Look for red flags like:

• Vague and unsupported claims about “AI-powered” processes
• Absence of verifiable results or data-driven benchmarks
• Lack of transparency into how the AI works

This is why my company takes the time to carefully test any AI tools we onboard into our own toolkit. Without mentioning vendors by name, we have often encountered supposedly groundbreaking AI applications that fall well short of their promises. This has included hiring software that promises to predict employee performance with a single questionnaire, marketing tools that claim to have cracked the code for social media dominance and law enforcement tools that use “predictive policing” to identify crime hotspots.

These all sound great in theory. But our internal auditing sometimes finds that there is little science to back these claims and limited utility to deliver on their promises.

How To Evaluate AI Tools for Your Business

So what should you be looking for in an AI solution?

• Alignment: You’re not looking for the trendiest AI tools on the market—just solutions that actually align with your business goals.
• Necessity: The best way to ensure you’re not buying into the tech just to keep up with trends is to identify internal problems in need of solutions first. Ensure that any AI technology you choose actually helps solve a specific problem. If it doesn’t, stay off the bandwagon.
• Training: Look for a vendor that is transparent about how their AI model is trained, evaluated and updated to ensure accuracy and prevent bias.
• Metrics: Experienced vendors will provide clear metrics and benchmarks for tracking outcomes. Some popular benchmarks include computer vision, which measures the ability of AI to identify images accurately, and Massive Multitask Language Understanding (MMLU), which tests the ability of AI to “think” logically and solve complex problems.
• Human Oversight: Be sure any AI solution you choose has well-established human in the loop (HITL) procedures for ensuring accuracy, fairness, quality control and depth of knowledge. Human oversight is the one criterion that we never compromise on.

The Power And Promise of AI

AI carries incredible potential to transform human work, creativity and innovation. But we must approach this potential with careful thought and a healthy dose of skepticism.

Audit your current tools. Ask tough questions about what you actually need. Choose value over novelty. And find a technology partner who is experienced, accountable and transparent. Most importantly, don’t think magically; think strategically.

See full article on Forbes here