3 Considerations to Make Sure Your AI is Actually Intelligent

In all the hullabaloo about AI, it strikes me that our attention gravitates far too quickly toward the most extreme arguments for its very existence. Utopia on the one hand. Dystopia on the other. You could say that the extraordinary occupies our minds far more than the ordinary. That’s hardly surprising. “Operational improvement” doesn’t sound quite as headline-grabbing as “human displacement.” Does it?

Small minds are concerned with the extraordinary, great minds with the ordinary.

The end game for AI is nowhere in sight partially because we’re putting the cart before the horse. Exhibit A in this argument is the absence of reliable data to teach the AI models. Even the most digitally advanced businesses we see today are operating with too many disparate technologies and data feeds. Their CMDBs are simply not up to the task of nurturing a well-behaved AI.

Make no mistake, generative AI is truly capable of extraordinary things. It can take the bar exam and pass with flying colors. It can spin up a business plan in seconds. It can detect fraud and manage risk with a high degree of automation. But it can’t do any of this without high-fidelity, real-time data. It can’t do this without sufficient guardrails to protect that data.

And that’s why getting the ordinary things right is so important. Here are three things to consider before your AI starts school.

1. Study Partners

Departmental cooperation is incredibly important when you’re looking to graduate as an AI-powered enterprise. For the machines to flourish, the humans in your organization will have to get with the program. Each of your internal kingdoms will need to conform to certain ‘study guidelines’ governing how they collect, store and process information – and stop operating independently. AI will expose whether you have a data management problem and where it resides. The number one problem with conflicting datasets is often the humans in charge of them.

2. Quality Curriculum

It’s imperative to come up with AI use cases and then choose which projects to pursue. It’s like selecting certain subjects at college to help you in your future career. But this will count for little if you don’t have good quality learning material for your large language models (LLMs). Tools like ChatGPT rely on enormous datasets. Your own marketing activity relies upon mountains of first (owned), second (shared), and third-party (bought) data. If you’re hampered by old, degraded, and inaccurate data, your AI will be too. Good quality data hygiene and governance are implicit in AI.

3. Specialized Subjects

Just a year ago, MIT ran a survey in which it found that 94% of organizations were using AI in some way already. Yet, just 14% were aiming to achieve “enterprise-wide” AI by 2025. This suggests that most businesses have a good working understanding of AI – albeit in pockets of the organization. The worry is that generative AI simply becomes another project that any department can spin up with no central strategy or internal specialization. The fear is that it becomes an outlier with none of the checks and balances that go into specialist domains, like AI in cybersecurity or compliance.

The success of any AI project will depend on the small, ordinary details, not the extraordinary claims and prophecies we often hear about. It all starts and ends with data. To that end, we should all be more worried about where the “I” in “AI” is coming from and whether that intelligence can be trusted.

Get Tanium digests straight to your inbox, including the latest thought leadership, industry news and best practices for IT security and operations. Subscribe to our weekly emails.