Hard Decisions for DIKW

There are existential and ethical issues with enterprises adding AI to or replacing traditional workflows.

This is going to happen, and there’s no need to deny it.

Unlike NFTs, crypto or metaverse bubbles, it’s real.

As long as organizational cultures and leadership refuse to invest in people, process, and tech to improve data quality of information libraries that are used as training data, expect eventual model collapse.

Data governance and strategy is what the leadership team does 5 minutes after dismissing the data on the dashboard generated by AI.

Model collapse is real.

Enterprises will begin selling/licensing their proprietary data for training purposes.

There is a lot of valuable LLM training data in private GitHub repos, PMO issue trackers, JIRA tickets, Slack chats, internal PowerPoint presentations, and proprietary knowledge bases.

But, without human stewardship it’s still crappy by any measure.

Garbage In, Garbage Out = model collapse

Some enterprises might employ a form of internal stewardship, along with cheap outsourced labor (read: mechanical turk) to massage the swathes of data.

It’s small fraction of the Russell 3000.

The outsourced workers will be a given a 100-page long document provided by a consulting firm, which derived it from other engagements.

There is no feedback loop between hand-on users and the conceptual product, marketing, and PnL owners.

For most, enterprise culture and the leadership is not invested nor incentivized to improve.

So, GIGO.

Generative AI may become automated “Big Data”.

It can generate outputs disconnected from valuable business outcomes.

Many executive boards approving investments are not engaged — nor willing to roll up sleeves — enough to tell the difference.

Generative AI democratized user access in a way that other tech hype cycles did not.

The possibilities feel tangible and it’s value as creativity-augmenter and tedium-reducer are real.

The DIKW pyramid is also real.

Biases are real.

GIGO is real.

I wonder if “Human Fixer of Machine Code” is a job title in 5–10 years.

The majority of people and enterprises will make easy decisions.

I’m not yet convinced that they are knowledgeable and wise decisions, though.

YMMV