Thoughts on AI: The Near-Term role of AI in Problem-Solving, Decision-Making, and Ethical Considerations

Don: Today I have the good fortune to be talking with Shaun VanWeelden, Partner at Mercor and formerly of OpenAI and Scale AI

I write for business generalists. Some of these business generalists are taking courses on AI, and there are already so many options; every online education site has at least one program on AI. Some other readers have heard the maxim, "AI won't take your job; someone who knows your job, AND knows AI, will take your job." But what does it mean to "know AI"? What do they need to be thinking about? What do they need to be preparing for in the immediate term?

That's quite broad, and that's the nature of AI in today's business context - there is just so much to look at. So, with that preface, I'm asking Shaun a few questions, but I've asked him to be comfortable reframing the question or even to replace it with a different question - feel free to go wherever you think is best with your response.

Shaun: Sounds good - first question?

Don: How do you see AI evolving in its ability to enhance human problem-solving and decision-making? Like, in real-time, for any manager in any business function - their day-to-day decision-making?

Shaun: Problem solving ... ok. I'm thinking of how we envision the universe of possible solutions. AI is already a vital tool, for automating tasks and for enhancing creativity. However, we're really not yet leveraging its full capabilities to support brainstorming. Our queries are too narrow and specific; we tend not to give AI the chance to come up with ideas we wouldn't have thought of on our own.

AI can offer possibilities we hadn’t considered, acting as both a guide and a collaborator. The key challenge lies in how we interact with AI — users need to improve their ability to clearly communicate their situation. Think about how you might describe your work to a friend who works in a different field - or to your mother. Too vague, and AI will underdeliver; too much context, and the results become noisy.

And the AI developers have work to do here. Today, if you ask ChatGPT to give you multiple answers to the same question, sometimes it will give you essentially the same answer a few times, just wrapped differently. The recent 'o1' release does better, but it can also ramble a bit.

But ideally, instead of us asking ChatGPT to brainstorm, we need it to do that for us automatically. This will make us smarter in our problem-solving.

Something else to be aware of: ChatGPT kinda gets magical when we upload CSVs or PDFs. You ask it to analyze a set of PDFs for commonalities or differences. You can get it to execute code inside the model, operating on your uploaded data. Are many people taking advantage of this? I don't think so, not as much as they could. Is this a use case broadly applicable to lots of users? I'm pretty sure. People just haven't realized how they could be using this.

Don: What are the biggest challenges businesses face when integrating AI into common business processes?

Shaun: I can think of two major sets of issues for broader adoption: Technical and Cultural.

Technically, many businesses have mountains of messy data and inadequate tools for cleaning and organizing information before feeding it to AI. The model will often fail, given crude queries and messy data.

Now, AI can clean the data for you, although the prompts could get complicated. There are other tools designed specifically for this purpose - it might be better to use them and feed their output into your AI model.

Similarly, some people will try to manage something like a JIRA ticket workflow within the model - and it might be able to do that for you, but it might be better to use a dedicated workflow tool. Third-party tools can add value here - use the right tool for the job.

Culturally, I would say, don't make it something big - it doesn't need to be big.

Get leadership visibly behind AI - in a team meeting, have someone share how they use AI to address common challenges. Encourage experimentation by showing how AI can be used in daily tasks. Cool, innovative applications will emerge when lots of people are just trying it and playing with it.

Don: How can AI enhance decision-making frameworks?

Shaun: Sequence matters. What I mean is, it might seem natural to feed data into the model, ask for a solution, and then ask for the assumptions and constraints that were considered in arriving at that solution. But, you'll get better answers if you ask for the assumptions and constraints first, and only then ask for the solution.

This approach clarifies the factors influencing AI’s recommendations, leading to better-informed decisions. Newer AI models align with this approach by focusing on assumptions first, then recommendations, helping users refine the decision-making process.

Don: How do we approach ethical considerations in AI, especially regarding trust and transparency?

Shaun: Ethical AI is a complex, ongoing debate. As just one of many ethical considerations, there are fears surrounding AI’s impact on creative work and its tendency to borrow from existing content without proper attribution. On the other hand, AI also has the potential to empower creators. Balancing these factors is difficult, especially when considering citation and crediting systems - which are just messy. The broader question is how to give people the power of AI without infringing on intellectual property — a challenge ethics experts will continue to grapple with.

Don: Thank you so much for your insights - I think readers will find this very useful!

Shaun: Thank you - my pleasure!

#projectmanagement #decisionmaking #GenAI