How businesses can use AI research strategically

Tl;dr - You need to understand how AI works and be realistic about its capabilities to achieve a decent result.

OpenAI rolled out its deep research tool to Plus users this week, so I gave it a test run.

As someone who spent four years constructing a doctorate, and has researched professionally, I felt queasy at the thought of this tool.

My verdict?

Deep Research (DR) offers a starting point for further investigation. And it makes the learning process accessible. But there are serious limitations.

Here’s how businesses need to think about maximising its potential -

1) Use AI as a jumping off point - Reports offer high level overviews. Used well, DR can give building blocks for putting together a broader piece of work

2) Clearly define the parameters of the topic - The user needs to set up a very specific research scenario. DR’s clarification questions are helpful in doing that. But you also need understanding of the topic. It returned my report within 10 minutes composed of the headings specified in my brief

3) Be absolutely clear about its limitations - AI can’t produce a point of view. Reports are generic, even formulaic, because AI can’t synthesise in any meaningful way. It’s never going to produce an original perspective

4) Results need to be checked - Sources are extremely hit and miss. It uses unreliable references - for example, any organisation using the word ‘institute’ is bizarrely deemed trustworthy. So check the reference links - it offered me some highly questionable ones

5) Humans need to drive the tech - The tech shouldn’t drive us. We need to become trained about how AI works, and its shortcomings, to synthesise our own results

My key takeaway - Exercise caution. Paradoxically, you need understanding of a topic to be able to define a problem, and to discern quality outputs. And divide and conquer - even GPT can't parse everything.

I’m more excited about using AI strategically across a business system, than for bounded tasks like DR.

Have you given it a test run yet? Thoughts?