Artificial Intelligence AI

Artificial intelligence is software that works a rule out from examples, instead of following rules that a person typed in advance. That is the honest, narrow definition, and it is a good deal smaller than the marketing around it. A spam filter, for instance, learned what junk mail looks like from millions of messages that people had marked as junk, so nobody ever sat down and wrote the rule.

In an ordinary business it turns up in four quiet jobs that nobody calls clever. It sorts the messages coming in, and it guesses how much of something you will sell next Friday. It reads a scanned supplier invoice and pulls the total off it, then writes a first reply to a customer that a person checks before sending. None of that is a robot, yet each one gives somebody back an hour a day.

Here is the thing worth knowing before you buy, because most products with AI on the box do not learn anything at all. They are a long list of if-this-then-that rules, which is respectable software but a different thing entirely. The test is simple, so ask the supplier what the system learned from and where those examples came from. If nobody can name the data, then nothing was learned. Linkysoft gets asked to check these tools before a client signs, and the answer is often that the product is a rule book with a new label.

The second thing to expect is that it will be wrong sometimes, because an AI answer is a good guess with a confidence attached rather than a certainty. So the real design work is not the clever part at all. It is deciding what happens to the uncertain cases, who gets to see them and who fixes them. When Linkysoft scopes an AI system, that question takes more of the first meeting than the model does, and clients are usually surprised by it.

There is a cost people miss too, because sending your customer records to an outside AI service is a decision about your data and not only about software. That is why it is worth having your security people in the conversation early. Most of what makes today's systems work is one method underneath, called machine learning, and knowing that helps you read a sales sheet. Linkysoft prefers to put AI inside the system a team already uses, so nobody has to learn a second screen.