The 5 articles you read in AI hell – The Next Web

The devil went down to Silicon Valley; he was looking for a soul to steal. But he ended up taking a consulting gig with Palantir instead.

In the meantime, the algorithms in charge of punishing the wicked now. And these days the sign above hells gates reads Abandon Open Source, with an Amazon smile beneath the print.

Those condemned to an eternity of pain and suffering in the modern era are now forced to read the same five AI articles over and over.

Which kind of sounds like what its like to read tech news back here on Earth anyway. Dont believe me? Lets dive in.

No it wasnt. These articles usually involve a text generator such as OpenAIs GPT-3. The big idea is that the journalist will either pay for access or collaborate with OpenAI to get GPT-3 to generate text from various prompts.

The journalist will ask something silly like can AI ever truly think like a human? and then GPT-3 will use that prompt to generate a specific number of outputs.

Then, the journalists and editors go to work. Theyll pick the best responses, mix and match sentences that make the most sense, and then discard the rest.

This is the editorial equivalent of taking the collected works of Stephen King, copy/pasting a single sentence from each book into a word doc, and then claiming youve published an entirely new book from the master of horror.

In hell, you stand in a long line to read hyperbolic, made-up stories about AIs capabilities. And, as your ultimate punishment, you have to rewrite them for the next person in line.

I remember reading about an early funding round for an AI company called PredPol. It had raised several million dollars to develop an AI system capable of predicting crime before it happens.

Im sorry. Perhaps you didnt read that right. It says: predicting crime before it happens.

This is something thats impossible. And I dont mean technologically impossible, I mean not possible within the realms of classical or quantum physics.

You see crime isnt generated from hotspots like mobs spawning in an MMO every 5 minutes. A first year statistics or physics student understands that no amount of historical data can predict where new crimes will occur. Mostly because the past isnt literally prescient. But, also, its impossible to know how many crimes have actually been committed.Most crimes go unreported.

PredPol cant predict crime. It predicts arrests based on historical data. In other words: PredPol tells you where youve already arrested people and then says try there again. Simply put: it doesnt work because it cant work.

But it raised money and raised money until one day it grew into a full-grown company worth billions all for doing nothing.

In hell, you have to read funding stories about billion-dollar AI startups that dont actually do anything or solve any problems. And youre not allowed to skim.

Theres variations on this one Googles AI demonstrates a 72% reduction in racial bias, Amazons new algorithm is 87% better at spotting and removing Naziproducts from its store front and theyre all bunk.

Big techs favorite PR company is the mainstream media.

Facebook will, as a hypothetical example, say something like our new algorithms are 80% more efficient at finding and removing toxic content in real time, and thats when the telephone game starts.

Youll see half a dozen reputable news outlets printing headlines that basically say Facebooks new algorithms make it 80% less toxic. And thats simply not true.

If a chef were to tell you theyve adopted a new cooking technique that results in 80% less fecal matter being detected in the soup theyre about to serve, you probably wouldnt think that was a good thing.

Increasing the efficiency of an algorithm doesnt result in a unilateral increase in overall system efficiency. And, because statistical correlations are incredibly difficult to make when you dont have access to the actual data being discussed, the people writing up these stories are simply taking the big tech marketing teams word for it.

In hell, you have to read articles about big tech companies that only have quotes from people who work at those companies and statistics that cant possibly be verified.

Weve all read these stories. They cover the biggest issues in the world of AI as if theyre writing about the weather.

The story will be something like Clearview AI gets new government contracts, and the coverage will quote a politician, the CEO of Clearview, and someone representing law enforcement.

The gist of the piece will be Ethics aside, law enforcement agencies say these products are invaluable.

And then, way down towards the end of the article, youll see the obligatory studies have shown that facial recognition struggles to identify some faces. Experts warn against the use of such technologies until this bias can be solved.

In hell, every AI article you read starts with the sentence this doesnt work as well for Black people or women, but were just going to move past that like it isnt important.

My least favorite AI article is the ones that profess to tell me what non-experts think.

These are the articles with headlines like Study: 80% of people believe AI will be sentient within a decade and 75% of moms think Alexa is a danger to children.

These studies are typically conducted by consultancy companies that specialize in this sort of thing. And usually theyre not out conducting studies on the speculation that some journalist will find their workappealing. They get paid to do their research.

And by research, I mean: sourcing answers on Amazons Mechanical Turk or giving campus students a gift card to fill out a survey.

These studies are often bought and paid for ahead of time by an AI company as a marketing tool.

These pitches, in my inbox, usually look something like Hey Tristan, did you hear that 92% of CEOs dont know what Kubernetes is? Are you interested in this exclusive study and a conversation with Dr Knows Itall, founder of the Online School For Learning AI Good? They can speak to the challenges of hiring quality IT talent.

Can you spot the rubbish?

In hell, the algorithm tells you that you can read articles covering actual computer science research as soon as you finish reading all the vapid survey pieces on AI published in mainstream outlets.

But youre never done are you? Theres always another. What do soccer dads think about gendered voice assistants? What percentage of people think data is a character on Star Trek? Will driverless cars be a reality in 2022? Heres what Tesla owners think.

Yes, AI hell is a place filled with horrors beyond comprehension. And, just in case you havent figured it out yet, were already here. This article has been your orientation.

Now if youll just sign in to Google News, well get started (Apple News is currently not available in hell due to legal issues concerning the App Store).

Read more here:
The 5 articles you read in AI hell - The Next Web