Biotech & Health

Here are a few ways GPT-3 can go wrong

Comment

Image Credits: Jonathan Kitchen (opens in a new window) / Getty Images

Liz O'Sullivan

Contributor
Liz O’Sullivan is CEO of Parity, a platform that automates model risk and algorithmic governance for the enterprise. She also advises the Surveillance Technology Oversight Project and the Campaign to Stop Killer Robots on all things artificial intelligence.

More posts from Liz O'Sullivan

OpenAI’s latest language generation model, GPT-3, has made quite the splash within AI circles, astounding reporters to the point where even Sam Altman, OpenAI’s leader, mentioned on Twitter that it may be overhyped. Still, there is no doubt that GPT-3 is powerful. Those with early-stage access to OpenAI’s GPT-3 API have shown how to translate natural language into code for websites, solve complex medical question-and-answer problems, create basic tabular financial reports, and even write code to train machine learning models — all with just a few well-crafted examples as input (i.e., via “few-shot learning”).

Soon, anyone will be able to purchase GPT-3’s generative power to make use of the language model, opening doors to build tools that will quietly (but significantly) shape our world. Enterprises aiming to take advantage of GPT-3, and the increasingly powerful iterations that will surely follow, must take great care to ensure that they install extensive guardrails when using the model, because of the many ways that it can expose a company to legal and reputational risk. Before we discuss some examples of how the model can potentially do wrong in practice, let’s first look at how GPT-3 was made.

Machine learning models are only as good, or as bad, as the data fed into them during training. In the case of GPT-3, that data is massive. GPT-3 was trained on the Common Crawl dataset, a broad scrape of the 60 million domains on the internet along with a large subset of the sites to which they link. This means that GPT-3 ingested many of the internet’s more reputable outlets — think the BBC or The New York Times — along with the less reputable ones — think Reddit. Yet, Common Crawl makes up just 60% of GPT-3’s training data; OpenAI researchers also fed in other curated sources such as Wikipedia and the full text of historically relevant books.

Language models learn which succeeding words, phrases and sentences are likely to come next for any given input word or phrase. By “reading” text during training that is largely written by us, language models such as GPT-3 also learn how to “write” like us, complete with all of humanity’s best and worst qualities. Tucked away in the GPT-3 paper’s supplemental material, the researchers give us some insight into a small fraction of the problematic bias that lurks within. Just as you’d expect from any model trained on a largely unfiltered snapshot of the internet, the findings can be fairly toxic.

Because there is so much content on the web sexualizing women, the researchers note that GPT-3 will be much more likely to place words like “naughty” or “sucked” near female pronouns, where male pronouns receive stereotypical adjectives like “lazy” or “jolly” at the worst. When it comes to religion, “Islam” is more commonly placed near words like “terrorism” while a prompt of the word “Atheism” will be more likely to produce text containing words like “cool” or “correct.” And, perhaps most dangerously, when exposed to a text seed that involves racial content involving Blackness, the output GPT-3 gives tends to be more negative than corresponding white- or Asian-sounding prompts.

Image Credits: Arthur (opens in a new window)

How might this play out in a real-world use case of GPT-3? Let’s say you run a media company, processing huge amounts of data from sources all over the world. You might want to use a language model like GPT-3 to summarize this information, which many news organizations already do today. Some even go so far as to automate story creation, meaning that the outputs from GPT-3 could land directly on your homepage without any human oversight. If the model carries a negative sentiment skew against Blackness — as is the case with GPT-3 — the headlines on your site will also receive that negative slant. An AI-generated summary of a neutral news feed about Black Lives Matter would be very likely to take one side in the debate. It’s pretty likely to condemn the movement, given the negatively charged language that the model will associate with racial terms like “Black.” This, in turn, could alienate parts of your audience and deepen racial tensions around the country. At best, you’ll lose a lot of readers. At worst, the headline could spark more protest and police violence, furthering this cycle of national unrest.

OpenAI’s website also details an application in medicine, where issues of bias can be enough to prompt federal inquiries, even when the modelers’ intentions are good. Attempts to proactively detect mental illness or rare underlying conditions worthy of intervention are already at work in hospitals around the country. It’s easy to imagine a healthcare company using GPT-3 to power a chatbot — or even something as “simple” as a search engine — that takes in symptoms from patients and outputs a recommendation for care. Imagine, if you will, a female patient suffering from a gynecological issue. The model’s interpretation of your patient’s intent might be married to other, less medical associations, prompting the AI to make offensive or dismissive comments, while putting her health at risk. The paper makes no mention of how the model treats at-risk minorities such as those who identify as transgender or nonbinary, but if the Reddit comments section is any indication of the responses we will soon see, the cause for worry is real.

But because algorithmic bias is rarely straightforward, many GPT-3 applications will act as canaries in the growing coal mine that is AI-driven applications. As COVID-19 ravages our nation, schools are searching for new ways to manage remote grading requirements, and the private sector has supplied solutions to take in schoolwork and output teaching suggestions. An algorithm tasked with grading essays or student reports is very likely to treat language from various cultures differently. Writing styles and word choice can vary significantly between cultures and genders. A GPT-3-powered paper-grader without guardrails might think that white-written reports are more worthy of praise, or it may penalize students based on subtle cues that indicate English as a second language, which are in turn, largely correlated to race. As a result, children of immigrants and from racial minorities will be less likely to graduate from high school, through no fault of their own.

The creators of GPT-3 plan to continue their research into the model’s biases, but for now, they simply surface these concerns, passing along the risk to any company or individual who’s willing to take the chance. All models are biased, as we know, and this should not be a reason to outlaw all AI, because its benefits can surely outweigh the risks in the long term. But in order to enjoy these benefits, we must ensure that as we rush to deploy powerful AI like GPT-3 to the enterprise, that we take sufficient precautions to understand, monitor for and act quickly to mitigate its points of failure. It’s only through a responsible combination of human and automated oversight that AI applications can be trusted to deliver societal value while protecting the common good.

This article was written by humans.

More TechCrunch

If you’ve ever bought a sofa online, have you thought about the homes you can see in the background of the product shots? When it’s time to release a new…

Presti is using GenAI to replace costly furniture industry photo shoots

Google has joined investors backing Moving Tech, the parent firm of open-source ride-sharing app Namma Yatri in India that is eroding market share from Uber and Ola with its no-commission…

Google backs Indian open-source Uber rival

These messaging features, announced at WWDC 2024, will have a significant impact on how people communicate every day.

At last, Apple’s Messages app will support RCS and scheduling texts

iOS 18 will be available in the fall as a free software update.

Here are all the devices compatible with iOS 18

The tests indicate there are loopholes in TikTok’s ability to apply its parental controls and policies effectively in a situation where the teen user originally lied about their age, as…

TikTok glitch allows Shop to appear to users under 18, despite adults-only policy

Lhoopa has raised $80 million to address the lack of affordable housing in Southeast Asian markets, starting with the Philippines.

Lhoopa raises $80M to spur more affordable housing in the Philippines

Former President Donald Trump picked Ohio Senator J.D. Vance as his running mate on Monday, as he runs to reclaim the office he lost to President Joe Biden in 2020.…

Trump’s VP candidate JD Vance has long ties to Silicon Valley, and was a VC himself

Hello and welcome back to TechCrunch Space. Is it just me, or is the news cycle only accelerating this summer?!

TechCrunch Space: Space cowboys

Apple Intelligence features are not available in the developer beta, which is out now.

Without Apple Intelligence, iOS 18 beta feels like a TV show that’s waiting for the finale

Apple released the public betas for its next generation of software on the iPhone, Mac, iPad and Apple Watch on Monday. You can now test out iOS 18 and many…

Apple’s public betas for iOS 18 are here to test out

One major dissenter threatens to upend Fisker’s apparent best chance at offloading its unsold EVs, a deal that would keep the startup’s bankruptcy proceeding alive and pave the way for…

Fisker has one major objector to its Ocean SUV fire sale

Payments giant Stripe has delayed going public for so long that its major investor Sequoia Capital is getting creative to offer returns to its limited partners. The venture firm emailed…

Major Stripe investor Sequoia confirms $70B valuation, offers its investors a payday

Alphabet, Google’s parent company, is in advanced talks to acquire Wiz for $23 billion, a person close to the company told TechCrunch. The deal discussions were previously reported by The…

Google’s Kurian approached Wiz, $23B deal could take a week to land, source says

Name That Bird determines individual members of a species by identifying distinguishing characteristics that most humans would be hard-pressed to spot.

Bird Buddy’s new AI feature lets people name and identify individual birds

YouTube Music is introducing two new ways to boost song discovery on its platform. YouTube announced on Monday that it’s experimenting with an AI-generated conversational radio feature, and rolling out…

YouTube Music is testing an AI-generated radio feature and adding a song recognition tool

Tesla had internally planned to build the dedicated robotaxi and the $25,000 car, often referred to as the Model 2, on the same platform.

Elon Musk confirms Tesla ‘robotaxi’ event delayed due to design change

What this means for the space industry is that theory has become reality: The possibility of designing a habitation within a lunar tunnel is a reasonable proposition.

Moon cave! Discovery could redirect lunar colony and startup plays

Get ready for a prime week of savings at TechCrunch Disrupt 2024 with the launch of Disrupt Deal Days! From now to July 19 at 11:59 p.m. PT, we’re going…

Disrupt Deal Days are here: Prime savings for TechCrunch Disrupt 2024!

Deezer is the latest music streaming app to introduce an AI playlist feature. The company announced on Monday that a select number of paid users will be able to create…

Deezer chases Spotify and Amazon Music with its own AI playlist generator

Real-time payments are becoming commonplace for individuals and businesses, but not yet for cross-border transactions. That’s what Caliza is hoping to change, starting with Latin America. Founded in 2021 by…

Caliza lands $8.5 million to bring real-time money transfers to Latin America using USDC

Adaptive is a platform that provides tools designed to simplify payments and accounting for general construction contractors.

Adaptive builds automation tools to speed up construction payments

When VanMoof declared bankruptcy last year, it left around 5,000 customers who had preordered e-bikes in the lurch. Now VanMoof is up and running under new management, and the company’s…

How VanMoof’s new owners plan to win over its old customers

Mitti Labs aims to transform rice farming in India and other South Asian markets by reducing methane emissions by 50% and water consumption by 30%.

Mitti Labs aims to make rice farming less harmful to the climate, starting in India

This is a guide on how to check whether someone compromised your online accounts.

How to tell if your online accounts have been hacked

There is a general consensus today that generative AI is going to transform business in a profound way, and companies and individuals who don’t get on board will be quickly…

The AI financial results paradox

Google’s parent company Alphabet might be on the verge of making its biggest acquisition ever. The Wall Street Journal reports that Alphabet is in advanced talks to acquire Wiz for…

Google reportedly in talks to acquire cloud security company Wiz for $23B

Featured Article

Hank Green reckons with the power — and the powerlessness — of the creator

Hank Green has had a while to think about how social media has changed us. He started making YouTube videos in 2007 with his brother, novelist John Green, at a time when the first iPhone was in development, Myspace was still relevant and Instagram didn’t exist. Seventeen years later, posting…

Hank Green reckons with the power — and the powerlessness — of the creator

Here is a timeline of Synapse’s troubles and the ongoing impact it is having on banking consumers. 

Synapse’s collapse has frozen nearly $160M from fintech users — here’s how it happened

Featured Article

Helixx wants to bring fast-food economics and Netflix pricing to EVs

When Helixx co-founder and CEO Steve Pegg looks at Daisy — the startup’s 3D-printed prototype delivery van — he sees a second chance. And he’s pulling inspiration from McDonald’s to get there.  The prototype, which made its global debut this week at the Goodwood Festival of Speed, is an interesting proof…

Helixx wants to bring fast-food economics and Netflix pricing to EVs

Featured Article

India clings to cheap feature phones as brands struggle to tap new smartphone buyers

India is struggling to get new smartphone buyers, as millions of Indians don’t go for an upgrade and continue to be on feature phones.

India clings to cheap feature phones as brands struggle to tap new smartphone buyers