Why Are AI Wizards Fessing Up?

May 10, 2021

I asked myself, “What’s up with the wizards explaining some of the information about the limitations of today’s artificial intelligence systems and methods?”


I noticed several write ups which are different from the greed infused marketing presentations about smart software.

The first article is an apologia. This term means, “a defense especially of one’s opinions, position, or actions,” as Merriam Webster asserts.”Fighting Algorithmic Bias in Artificial Intelligence” allows the title to indicate that algorithmic bias is indeed an issue. The algorithms are not narrowed to machine learning. Instead the title pops up to the umbrella term. Interesting. Here’s a passage which caught my attention:

From Black individuals being mislabeled as gorillas or a Google search for “Black girls” or “Latina girls” leading to adult content to medical devices working poorly for people with darker skin, it is evident that algorithms can be inherently discriminatory…

Okay, reasonably convincing. But what went wrong in the university courses providing the intellectual underpinnings for smart software? That’s a question that the write up emphasizes in a pull quote:

It’s not just that we need to change the algorithms or the systems; we need to change institutions and social structures. — Joy Lisi Rankin

How quickly do institutions and social structures change? Not too quickly where tenure and student employment goals are intertwined with judgment, ethics, and accountability I surmise.

The second article I noted contains the musings of an AI pioneer (Andrew Ng) as related to an IEEE writer. “Andrew Ng X Rays the Hype” seems to assert that “machine learning may work on test sets, but that’s a long way from real world use.” We’re not talking about AI. Andrew Ng is focusing on machine learning, the go to method for the Google-type company. The truth is presented this way:

“Those of us in machine learning are really good at doing well on a test set,” says machine learning pioneer Andrew Ng, “but unfortunately deploying a system takes more than doing well on a test set.”

The point is that a test is just that, an experiment. MBAs engage in spreadsheet fever behavior in order to generate numbers which flow or deliver what’s needed to get a bonus. The ML crowd gets a test set working and then, it seems, leaps into the real world of marketing and fund raising. With cash, those test sets become enshrined and provide the company’s secret sauce. What if the sauce is poisoned? Yeah, ethics, right?

The third write up is appears in an online information service which has done its share of AI cheerleading. “What I Learned from 25 Years of Machine Learning” is a life lessons-type write up. What did the TechTarget Data Science Central article learn?

“Learn” is not the word I would use to characterize a listicle. There are 11 “pieces of advice.” Okay, these must be the lessons. Please, navigate to the source document to review the Full Monty. I want to highlight three “learnings” expressed as “advice.”

The first gem I will highlight is “be friend with the IT department.” Maybe be friendly or be a friend of the IT department. The learning I gleaned from this “piece of advice” is use Grammarly or find an English major to proofread. Let’s consider the advice “be a friend of the IT department” and ask “Why?” The answer is that smart software can be computationally expensive, tough to set up, and a drain on existing on premises or cloud computing resources. The IT departments with which I am familiar are not friendly to non IT people who want to take time away from keeping the VP of sales’ laptop working. Data wizards are outsiders and the IT department may practice passive aggression to cause the smart software initiative to move slowly or not at all.

The second advice I want to flag is document. Yeah. The way the world of mathy things works is to try stuff. Try more stuff. Then try stuff suggested by a blogger. Once the process or numerical recipe works, the focus is not on documenting a journey. The laser beam of attention goes to hitting a deadline and hopefully getting a bonus, a promotion, or one of those Also Participated ribbons popular in the 1980s’ middle schools. As one of my long time tech wizards said, “Document? You wish.”

The third “module” of these learnings is “get precise metrics”. Okay but precision requires specific information. Who has specific information about the errors, gaps, timeliness, and statistical validity of the data one must use? Yep, good luck with that. Quick example: Due to my research for my National Cyber Crime Conference lectures, Google is now displaying ads for weapons, female bikini hauls, fried chicken sandwiches, and mega yachts. Why? Google’s method of determining what data to use from my online queries struggles because we were using one computer to research cyber crime (weapons), pornography stars on social media sites and the Dark Web, lunch (hence the chicken fetish), and money laundering. I mean how many mega yachts does one honest business person need with one new wife and handful of former spetsnaz professionals? Yeah, data and precise metrics. If the Google can’t do it, what are your chances, gentle reader.

Now back to the question: Why are AI wizards confessing their digital sins?” My answer:

  1. The increased scrutiny of Amazon, Apple, Facebook, et al bodes ill for these firms and their use of smart software to generate money. This is a variant of the MBAs’ spreadsheet fever.
  2. High profile AI “experts” want to put some space between them and the improvised roadside Congressional investigations. Bias is a heck of a lot easier to tie to math particularly when high profile ethics issues are making headlines in the Sioux Falls Argus Leader.
  3. The wizards want to be in the group of wizards who can say, “Look. I explained what should be done. Not my personal problem.”

Net net: AI has bought the mid tier consultant-speak, frothy financial confections, and behavior of a smart person who is one step ahead of an ATM user.

Stephen E Arnold, May 10. 2021


Got something to say?

  • Archives

  • Recent Posts

  • Meta