Thread by @galgiacomelli

📌Do you wonder why so many AI projects fail in the real world?

Every AI developer should read this!

💥THREAD💥

#AI #DeepLearning #100DaysOfCode #100DaysOfMLCode #Python #javascript

We all imagine AI developer as a geek solving math equations all day. The funny thing is AI developers imagine exactly the same. We just want to be left alone with our models and develop in peace, but the real world doesn't work this way.

Business needs change often and we need to embrace it if we want it or not

The change in business needs means we have to change the goal of our algorithms to maximize the value we generate in the company. To give you an example imagine a company producing cookies

They want to eliminate burned cookies from reaching their customers. The company hires AI developers to make a model that will detect burned cookies on the production line and mark them for employees to remove them.

The developer comes back from his hole and presents a solution that works with 99% accuracy. They quickly realize, If it works with 99% accuracy we can fire 2 workers. Cokie company happily pays 40.000 implements said solution and saves 5.000 per month on salaries

After 4 months, the cookie company changes light bulbs in their factory to save on cost. The company saves $3.000 per month on a lower electricity bill. Changed light conditions, cause that AI often marks perfect cookies as burned and the company is suddenly without stock.

A day without stock costs the company $20.000. The company calls AI developer and complains that the software broke down and they need to fix it. After 7 days developer comes back with the solution that works.

The company pays $15.000 upgrade and moves on with $160.000 loss, counting the days without stock. In the next year, they change the color on the walls, color of their production line, and every time the cycle of software upgrade and company loss continues.

The company’s manager blamed developers for making bad software. Developers blamed the company for not following the obvious rules. Wouldn’t they just be better off keeping their workers employed and never bother with AI? Actually not.

The problem arose from unmet expectations, from both sides. The company’s manager imagined AI as a magical tool that will save on cost, he believed media hype and not what AI can do, but he was not the only hype victim.

Developers unknowingly omitted AI’s limitations because they read a research paper and implemented a solution that worked in a highly controlled environment. Nobody bothered to talk to each other to explain how to use the technology best.

Being an AI developer is not just building models with the best accuracy. A big part of our job is to talk with people and figure out what they need and want. With AI we are mimicking humans and we have to treat it as such.

Like every employee, AI has to receive training to know what is going on. Like with people, training does not end on the day we got a job, but it lasts as long as we live.

This is why every good AI developer and business manager has to keep in mind that only through communication we can figure out what we need and want. They have to listen to each other, know their limits and express everything they need.

Remember that AI is “not set and forget” but “set and adjust”.

Do you agree?

@threadreaderapp
unroll