AI is long past being just text
A pure language model works with text (see module 2). A multimodal model can also handle other kinds of content – images, audio, video. "Multimodal" means exactly that: it processes several modalities, not just one. For you as a decision-maker this matters, because many real problems aren't text problems at all, but image or speech problems.
Understanding isn't the same as generating
For each modality it's worth distinguishing two directions. Understanding (analysis) means AI takes in existing content – describes a photo, reads text out of an image, transcribes an audio recording. Generating means AI creates new content – an image from a description, a spoken voice, a short video clip. Understanding is usually more mature and lower-risk than generating – generated images, voices, and videos raise their own questions (labeling, deepfakes – see module 4 and module 20).
Where it matters day to day
As soon as you stop thinking of AI as only chat, new use cases open up: automatically transcribing and summarizing meetings, having a submitted damage photo described, reading text out of photographed documents, generating product images or first marketing drafts. Often the right modality is already half the solution.
Why this matters for you as a decision-maker
When a problem comes up, it's worth asking: is this actually a text, an image, a speech, or a video problem? Choosing the right modality saves detours. And for generated content, the same care applies as with text: check it, label it, don't trust it blindly (see module 4).