A clip from our spring 2025 GenAI workshop for mission-driven teams, with AlignIQ founder Dan Garmat. Recorded April 2025; reasoning models, image and music generation have all improved since.
A tour of where AI’s abilities sit across different kinds of human intelligence, and where the boundary still falls.
Key takeaways
- Linguistic: AI can create, edit, adapt and translate text and find patterns. But it never means anything. Under the hood it’s very advanced autocomplete trained on the internet.
- Logical-mathematical: improving fast with models that “think” before answering. Software engineering is still safe; routine coding is less so.
- Visual-spatial: getting better at images that make sense.
- Musical: it can analyze a song, but what it creates sounds like elevator music. Without human experience, its creative attempts don’t connect.
Try this with your team
For one role on your team, list its main tasks and sort them: where AI could accentuate the work, and where a person has to lead. Discuss the ones you disagree on.
Going further
Our cohort training builds this kind of judgment into everyday practice.
Sources mentioned
- Howard Gardner, Frames of Mind: The Theory of Multiple Intelligences (1983)
More from this series
- Why Purpose Matters More in the Age of AI
- Mapping the Jagged Frontier: Finding Clarity in a Fast-Moving Field
- Finding Your Voice with AI: Why Personal Experimentation Matters
- Risk and Reward: Building a Thoughtful Process for Working with AI
- Aligning AI with Your Mission
- AI Personalization, Search, and Real-Time Support
Transcript
Lightly edited from YouTube’s automatic captions for readability. Wording and order are the speakers’ own.
Human intelligence can be broken down in different ways. We can talk about multiple intelligences, and AI is actually very good at parts of these.
For example, with linguistic intelligence, it’s able to create, edit, adapt and translate text and find patterns. It’s able to do these particular technical skills of linguistic intelligence. But as it says, finding the right words to express what you mean: it doesn’t ever mean anything. It doesn’t have a real perspective. It’s not conscious. It doesn’t have a world model. What’s under the hood is autocomplete, like when you’re texting, just a very advanced version trained on the entire internet. That’s why it’s able to do really amazing things. So the linguistic skills are there. It can accentuate, but not necessarily replace.
Logical-mathematical is similar, and it is getting better. There are approaches people are taking now with reasoning models, where they stop and think for a while and then output something, and this is greatly improving the quality of the output when logical accuracy is required. Computer programming, for example, is completely changing. Software engineering is still safe, but being a coder in that sense is decreasingly so.
Spatial-visual: it’s getting good at creating pictures recently, applying that logical aspect so the pictures start to make a little more sense.
Musical is an interesting case. While it understands music theory and can analyze a song, what it can create is more like elevator music. It’s just not that good at this time, and it could be because it doesn’t have a human experience. When it starts trying to get creative, it just doesn’t connect. It doesn’t make sense.
So I think that gives a sense of where this boundary is, at least on the big picture. These are the areas where there is some jagged edge…