[elexhobby]

Furthermore, follow https://twitter.com/_brohrer_/status/1425770502321283073

"When you have a problem, build two solutions - a deep Bayesian transformer running on multicloud Kubernetes and a SQL query built on a stack of egregiously oversimplifying assumptions. Put one on your resume, the other in production. Everyone goes home happy."

[hughrr]

This reminds me of an experience I had watching a company trying to replace a system with ML.

First they marketed it heavily before even thinking. During test cycle they fed the entire data corpus in and ran some of the original test cases and found some business destroying results pop out. The entire system ended up a verbatim port of the VB6 crap which was a verbatim port of the original AS400 crap that actually worked.

The marketing to this day says it’s ML based and everyone buys into the hype. It’s not. It was a complete failure. But the original system has 30 years of human experience codified in it.

[pedrocr]

The AI taxonomy includes the term "Expert Systems" for these kinds of things. On the one hand it's definitely not of the new wave of ML AI so hyping those things as innovative is off. On the other hand we should definitely give more attention to that kind of setup and understand how to build/maintain/test it properly. Otherwise often it ends up being ran by a few hundred Excel sheets and a few severely underpaid people and that's a disaster waiting to happen. The AS400->VB6->NewShiny path actually sounds like a success case given the messes that are out there.

[ethbr0]

If I had a nickel for every time I've seen "business rules engine" turned into "AI" in the last few years...

But I guess if we complain that half of our colleagues and the media don't understand ML, why should we expect management to?

When the command from C-level is "We need some AI projects to tell our shareholders about," we shouldn't be surprised when middle management suddenly has successful AI projects in their slide decks.

[np_tedious]

Isn't your organization itself a machine the learned these rules over time? Maybe the marketing checks out

[knodi123]

We did the same thing when I worked for a resume search/sort/share site. Built a big ML tool that could look at job listings and resumes and pick who was best for each job. Our training data set was millions of resumes, hundreds of thousands of jobs, and in most of those jobs, we could say which resumes got shortlisted and which resumes got hired.

In the end, it gave basically the same results as keyword searching. But we marketed the shit out of it.

[thecopy]

If it worked, why was it crap?

[NumberCruncher]

The "the right tool for the right job" applies for ML topics too.

If the job involves "looking smart and innovative" for whatever reasons, people tend to err on the side of overly complex solutions.

On the other hand if the advice "let's just go with an SQL query built on a stack of egregiously oversimplifying assumptions" comes from someone, who doesn't know how SQL and linear regression / logistic regression with binning/bucketing / simple decision trees work, I would ask for a second opinion. Because a huge part of the retail banking, non-life insurance and marketing business is running on this simple stack. Obviously profitable.

If the same advice comes from someone, who knows when to use deep learning instead of XGBoost and why, I would go with his/her advice. And I would try to keep him happy and on my team.

[arketyp]

Furthermore in the article, yes.

[arnaudsm]

This isn't ironic, I've actually done that multiple times in a large company. No one noticed, everyone went home happy.

[bostonpete]

Well, the article does conclude with that exact tweet...

[marcosdumay]

My workplace has got all kinds of attention for building a blockchain based data collection system that encompasses an entire sector of the economy. It's "almost done", so we are right now starting a simple set of REST services that write into a badly normalized transactional database just in case it stays "almost done" for too long.

[q-base]

That quote is seriously brilliant! Thanks for sharing.

[DonHopkins]

Just don't build one solution to your problem with regular expressions: then you have two problems.

[smichel17]

TFA ends with that quote.