"A financial analyst I know recently
asked ChatGPT to write a report. Within seconds, the software generated a
passable document, which the analyst thought would earn him plaudits.
Instead,
his boss was irate: "You told Microsoft everything you think?"
This scene is playing out at
countless organizations. Innovative employees find new uses for artificial
intelligence while their employers worry about losing sensitive data.
Businesses are learning that large language models are powerful but not
private. Before the technology can give you valuable feedback, you have to
offer it valuable information.
Many executives aren't willing to
make that trade. They don't want to compromise their greatest asset, let alone
train an algorithm that could be used by their competitors.
Yet what good is an asset if you
don't have access to it? Corporate data are like gold mines beneath every large
organization, but they're often stored in inconvenient ways. Troves of
company-specific material are effectively wasted because they're poorly
organized and thus not consulted when needed. The data assets of large
organizations require a connective technology to realize their true worth. AI
can become that technology by absorbing an enterprise's data and translating
them into a highly accessible algorithm.
Given security concerns, however,
many organizations won't ship their data to large tech companies. Instead, they
will bring AI inside the tent, to train and deploy it within their own
companies. I call this approach private AI.
Such technology isn't hard to
implement. Take a publicly sourced large language model, such as StarCoder or
Bloom (https://bigscience.huggingface.co/blog/bloom), many of which are available on marketplaces like Hugging Face. Then,
train it on your own data sets, and you'll get a competent model that can
address the kind of situations your business faces. Private AI can handle
communications from customers and will be able to read, route and prioritize
incoming correspondence.
Private AI, like all artificial
intelligence, will require oversight. Expect to keep humans in charge -- as
editors and decision makers -- for as long as you can project. The technology
likely won't pass the Turing test or plan your next dinner party, but it will
help corporations save considerable time on tasks without sacrificing valuable
data assets. With time, it might become like the custom software application: a
commonplace way for corporations to remain autonomous and unique.
---
Mr. Calkins is CEO of Appian, a
business software company." [1]
1. Don't Let AI Steal Your Company's Data. Calkins, Matt.
Wall Street Journal, Eastern edition; New York, N.Y. [New York, N.Y]. 29 June 2023: A.17.