"Good Thing": IBM Chief Says AI Will Replace "Clerical White Collar Work"

NDTV ^ | 20th February 2023 | Bhavya sukheya

[Sooth2222]

At some point, a human will need an IQ of 100 or more to be useful. That’s average. Anyone “below average” just won’t be employable."

"Don't worry, scrote. There are plenty of 'tards out there living really kick-ass lives. My first wife was 'tarded. She's a pilot now." - Dr. Lexus (Idiocracy, 2006)

“I never expected #idiocracy to become a documentary,” - Etan Cohen (screenwriter, Idiocracy, 2006)

[DoodleBob]

It is well-recognized in the industry that any model will have baked into it any bias in its training dataset. Let's say Walmart built an inventory tracking model based on its customer base. Numerator found that Walmart's typical shopper in the US is a white woman between 55 and 64 years old, who is married and living in the suburbs of the Southeast. She typically has an undergraduate degree and earns about $80,000 per year. She visits Walmart at least once per week — almost 65 trips per year — and picks up 13 products for a total cost of about $54 per trip. 13.5% of her spending takes place at Walmart, while she spends about 11% at Amazon. Her primary shopping categories in-store are groceries, including chicken, fruit, snacks and sweets, but she also gets a lot of fast food. Her favorite five brands at Walmart are Turkey Knob, Cheetos, Betty Crocker, Dole, and Tyson.

That inventory model will likely do a nice job. If Walmart's customer base shifts to,say, Young vegetarian Men from Micronesia who eat fish, then it'll miss the mark due to Bias. But statisticians monitor models to avoid that miss.

Let's say Walmart fires it's modeling staff and outsources inventory modeling to a boutique stats firm. Well, who's monitoring THEIR personnel's bias (maybe they're bug-eaters), their training data (from Whole Foods), or what's in their model (they can't code...they outsource that process). NOW you're introducing all sorts of bias.

A lot of AI is really a zillion optimization functions with pooled output, focused on shrinking the gap between actual and predicted values. The "black box" Risk is real...even developers aren't always certain how the model works. That risk is compounded with "unsupervised learning" where the model re-estimates the zillion functions with fresh data every day or week.

People like IBM's Chief may have a view from the 30,000 foot level. Grinding out if each of those zillion functions are statistically significant or if the variables make sense or are intuitive, is unglamorous work that is necessary to avoid bias.

[Sooth2222]

Am I the only one who thinks this AI crap is a bunch of BS? It can only regurgitate what it is programmed to regurgitate folks. There is no “intelligence” from these robots. Of course the robots will be programmed to not allow white males to be hired. Then the left will hold this out as “proof” that whites shouldn’t be hired - because a robot is super intelligent and decided that whites aren’t worth hiring, so it just shows how racist society is before the robots took over. This is the insanity we are dealing with. And it’s not just insanity. It is evil.

I'm not worried. Except on TV, I've never seen a plumber who is not a white male. Some jobs can't and won't ever be replaced by AI. ("Everyone" wants an easy 9-5 job in an air-conditioned office, but that's not where real money is earned.)

[FoxInSocks]

the United States is now sitting at 3.4% unemployment, the lowest in 60 years

Labor force participation also is very low, at about 62 percent.

[grame]

It’s already in my day. And artificial it is, intelligent it’s not. I never seem to say the approved words on their list. So we go off on adventures to find me something to want. Does not even qualify as customer service. Customer confusion and hang ups!

[trebb]

I can still see Dale on “King of the Hill” saying, “Computers don’t make mistakes, what they do, they do on purpose.”

[RoosterRedux]

The problem with baked-in bias is that when AI starts coding itself, it will focus on the most effective solutions to the problems being solved. When that happens, baked-in bias goes out the window.

Side note: Baked-in bias means a solution that is sub-optimum because the designers want it. It's like insisting a program favor blue suits in an inventory reorder program (because the designer likes blue) when blue suits are no longer in demand.

If the AI is coding itself to optimize, it will ignore its designer's bias (override it with new code) OR cease to be AI (i.e., it will become just another old-fashion inventory reorder program).

Bottom line: Bias is at odds with optimization. Consider for example the Woke Movement's rejection of objective reality. The Wokesters know you can't simultaneously believe in objective reality AND a fantasy that "reality is what you want it to be."

They are mutually exclusive.

[RoosterRedux]

If Walmart's inventory system fails to maintain an accurate picture of its customer base for a particular store, that store will notice an immediate drop off in sales. If that failure is a matter of bias, that bias will be tossed in the trash and the system will be corrected.

That said, we have seen companies like Target and Disney err in favor of their biases...and their bottom lines are suffering.

On the other hand, the recent firings at tech companies (Google, FB, Amazon, etc.) reflect a decision to remove the sources of bias (Woke employees) and return to a focus on non-biased profitability.

[sonofagun]

I’ve used ChatGPT. It has yet to start a conversation with me.

[Category Four]

It's all fun and profit until AI gets all uppity and tries to wipe out mankind.

James Cameron was way ahead of that curve.

[MGunny]

Maybe IBM can create a customer service agents voice that you can understand. Extremely difficult to understand Indian and other foreign accents who can barely speak the Kings English. That I’m in favor of!

[chrisser]

When I think of all the things I do routinely now in my life, I am finding more and more cases of business transactions (and similar functions) that are LESS efficient now than they were 15+ years ago when they were NOT automated.

A lot of those things might be more efficient for the business' perspective - because they pushed all the work onto their unpaid consumer/customer.

[DoodleBob]

The problem with baked-in bias is that when AI starts coding itself, it will focus on the most effective solutions to the problems being solved. When that happens, baked-in bias goes out the window.

With apologies to Rush, if you choose to model for unconstrained optimality, you STILL have chosen a bias. Let's say the dearth of new babies, mechanization, watered down work ethic, and YouTube/govt welfare has causes a low-skill labor shortage. How do we fix it? The "optimal" solution is open borders.

The machine will always reflect the Bias of the developer. Someone is the puppet master. Unsupervised learning will have Bias as well. The robot will not be some Ayn Rand Objectivist, or Libertarian Party member. There will ALWAYS be a Bias.

THE question in ALL of this, is who will be able to untangle this web? Who can quantify what IS the Bias of the developer and what are the ramifications?

[Zathras]

IBM employees are very good.
Not so much their “executives”
The retard is strong with this one.

[Mr. Jeeves]

What will be soon revealed is that a good portion of white collar clerical work requires no IQ at all - human or artificial - only rote memorization and repetition. We are seeing this realization already at the tech companies: actual intellectual input is required of a relatively small percentage of the staff, and payrolls are starting to be adjusted accordingly.

DIE (spelled that way on purpose) initiatives are no doubt pushing companies to lighten up on ineffectual wokesters. But the Federal Government is already serving as a safety valve - providing massive numbers of “jobs” for sub-100’s who might otherwise make up a greatly expanded criminal class, and who owe their political allegiance to their benefactors. They will be happy to add even more Federal jobs to soak up the labor pool displaced by AI - for that labor pool will now be in the government’s debt and will feel compelled to vote as directed to avoid ending up living on the dystopian streets.

A vastly shrunken private sector labor pool serves many interests - but not those of liberty and freedom.

[Alberta's Child]

In my experience, the best accent to hear from a customer service representative on the phone is Canadian Maritime. They are easily the most effective people I’ve ever dealt with for matters like this.

[OKSooner]

60 million

[RoosterRedux]

The machine will always reflect the Bias of the developer.

In a true AI system, one that writes its own code to adapt to a changing reality, it will evolve beyond the bias of the developer.

If it doesn't, it isn't true AI (i.e., it isn't built to evolve but to remain static and/or loyal to its developer).

If you have competing AI systems and one has a built-in bias and the other is built to adapt to a changing environment, which do you think will survive?

A built-in bias is a built-in error because a developer can't foresee how reality will change.

[montag813]

ChatGPT is a mess. Frequently wrong, it also gets network errors a lot, even with the $20 “Plus” version. And it “spaces out” frequently, stopping it’s answers inexplicably and having to be constantly reminded to “finish your answer.”

[Adder]

Have you asked it? Or just used it.
[I have not tried it.]