In these early days of the AI era, many companies are simply giving their employees access to AI tools and hoping for the best. But in a recent post  to his always-interesting blog One Useful Thing, Ethan Mollick, an associate professor at The Wharton School, argues that we'll achieve the biggest gains from AI only after we fundamentally reshape the culture and methodology of work. 

To which you might be saying: OK, sure, let's rewire work for AI … but, um, how? 

Mollick offers a few ideas, such as using AI to summarize stakeholder feedback, but it's worth taking a step back to view the bigger picture. How do our current work practices inhibit the use of AI? And how might we overcome those obstacles? 

I'd flag three key gaps that managers can tackle right now: what I'll call the access gap, the ignorance gap, and the consistency gap. 

The Access Gap. Many employees are still barred entirely from AI tools. Sometimes, these prohibitions exist for very good reasons (for instance, most AI tools aren't approved for HIPAA-protected data), but other times, they stem from overblown fears of AI's risks. 

• How to close the access gap: Rather than banning AI tools entirely, craft policies that carefully balance risks and benefits. Maybe you can't trust AI with protected health information, but can your marketing team use ChatGPT to draft press releases? Can your strategy team use it to understand market developments? Can you offer trainings to help your team understand what data they should (and shouldn't) enter into external tools? 

The Ignorance Gap. Most people simply don't know how to use AI effectively. That's no surprise: These tools barely existed a year ago, and few people have received formal education on how to use them. 

• How to close the ignorance gap:  The most obvious step, of course, is training. I've offered my own pass at Prompt Engineering 101; Mollick has shared his guidance via a series of YouTube videos targeted at educators; and organizations such as LinkedIn Learning are also investing heavily in this space.
  But once you've learned a few basic AI skills, education matters less than experimentation. Try using AI to tackle a few items on your to-do list, and encourage your team to do the same. If a request doesn’t work the first time, try reframing your prompt, or try breaking the task down into subcomponents that AI can more readily understand. You'll very soon develop an intuition for what AI typically gets right (and where it tends to flail). 

The Consistency Gap. Even when people do know how to use AI, they may neglect to use it regularly, especially if their work habits have been ingrained by long practice. I fall into this trap all the time: For instance, I know  AI tools are great at summarizing my virtual meetings from their transcripts … but I often forget to turn on transcription anyway. 

• How to close the consistency gap: As an individual, embrace the power of defaults. Depending on your organization’s policies, you may be able to automatically enable transcription for all your meetings, for example. Other ideas: Try making ChatGPT (or another AI tool) your homepage to regularly nudge yourself to consider its uses, or identify tasks from your workflow that AI performs effectively and then reserve a recurring calendar time to tackle them.
  As a manager, your opportunities are even bigger. Identify the work AI does well, and rewire your processes to ensure AI always does that task. For instance, if you find that ChatGPT is great at summarizing news developments about your competitors, perhaps you can assign a junior team member to run a specific ChatGPT query each week and share the results — ensuring you always have up-to-the-minute competitive intelligence.  

In the longer term, as Mollick points out, we'll all have more work to do to fully integrate AI into our offices. We'll need to build AI tools into all of our software, restructure teams to make better use of AI, and more.  

But you don’t have to wait. You can start closing the access, ignorance, and consistency gaps right now. 

This week's other top AI-related links 

Large language models (LLMs) can do much more than chat. Because ChatGPT kicked off the current AI craze, many people tend to think of LLMs as just "chatbots." But in truth, LLMs are a new form of intelligence that you can use in many ways. This post describes an LLM-powered "Research Assistant" that takes a query, breaks it down into subcomponent questions, performs the necessary research, and drafts a final response. The multi-step process is much slower than just asking a question to ChatGPT, but in many use cases, speed matters less than accuracy and comprehensiveness. 

A technical (but really good!) one-hour intro to LLMs. If you're ready to dive deeper into the underlying workings of LLMs, this YouTube video from OpenAI's Andrej Karpathy is a great resource, covering everything from technical fundamentals to security implications. (If you'd rather just skim the slide deck, that's available too.) 

Is "artificial general intelligence" already here? Many pioneers of the AI age have defined their holy grail as achieving "AGI," or artificial general intelligence. OpenAI's mission statement, for instance, is "to ensure that artificial general intelligence benefits all of humanity." But what if AGI is already here? This article from Noema argues that "[d]ecades from now, [tools such as ChatGPT] will be recognized as the first true examples of AGI, just as the 1945 ENIAC is now recognized as the first true general-purpose electronic computer."

Your life, as narrated by David Attenborough. In a wild sign of things to come, a developer has tied his webcam to an AI voice generator that narrates his day-to-day life in the distinctive voice of David Attenborough: "Here we have a remarkable specimen of homo sapiens, distinguished by his remarkable silver spectacles and a mane of tousled curly locks." (Sadly, the real-life David Attenborough finds this "personally distressing.") 

Artificial intelligence (AI) can provide innovative solutions to some of healthcare's greatest problems, but only if we address associated obstacles and risks. Explore five AI challenges and get actionable solutions to help your organization overcome them.