AI Is Dangerous, but Not for the Reasons You Think | Sasha Luccioni | TED

Summary

The speaker, an AI researcher, addresses the growing concerns around the societal impacts of AI models. She highlights the urgent need to measure and mitigate AI's environmental costs, tackle issues like data privacy and consent, and address inherent biases that can perpetuate stereotypes and discrimination. By creating transparent tools and disclosing information, the speaker advocates for a future where AI models are more trustworthy, sustainable, and less likely to cause harm. The overarching message is that instead of obsessing over hypothetical existential risks, we should focus on addressing AI's tangible impacts on people and the planet right now.

Takeaways

• 🌍 AI models contribute to climate change and environmental issues due to their high energy consumption and carbon emissions during training and deployment.
• 📘 AI models use artwork, books, and other creative works for training without consent from the creators, raising copyright concerns.
• ⚖️ AI models can encode biases and stereotypes, leading to discrimination against certain communities when deployed in real-world scenarios.
• 🔍 Transparency and tools are needed to understand and measure the impacts of AI models, such as their environmental footprint, use of copyrighted data, and biases.
• 🛡️ Initiatives like CodeCarbon, Have I Been Trained?, and Stable Bias Explorer aim to provide tools for measuring and mitigating AI's impacts.
• 💻 Companies and developers should prioritize choosing more sustainable, unbiased, and ethically-trained AI models based on measurable impacts.
• 📋 Legislation and governance mechanisms are required to regulate AI's deployment in society and protect against harmful impacts.
• 🙋‍♀️ Users should have access to information about AI models' impacts to make informed choices about which models to trust and use.
• 🌱 Rather than focusing solely on hypothetical future risks, immediate action is needed to address AI's current, tangible impacts on society and the environment.
• 🤝 Collective effort from researchers, companies, policymakers, and users is required to steer AI's development in an ethical and responsible direction.

Q & A

What was the strangest email the AI researcher received?

The AI researcher received an email from a random stranger saying that their work in AI is going to end humanity.

What are some of the negative headlines surrounding AI that the researcher mentioned?

The researcher mentioned headlines about a chatbot advising someone to divorce their wife, and an AI meal planner app proposing a recipe containing chlorine gas.

What are some of the current tangible impacts of AI that the researcher discussed?

The researcher discussed the environmental impact of training AI models, the use of copyrighted art and books without consent for training data, and the potential for AI models to discriminate against certain communities.

What is the Bloom model, and what did the researcher's study find about its environmental impact?

Bloom is the first open large language model like ChatGPT, created with an emphasis on ethics, transparency, and consent. The researcher's study found that training Bloom used as much energy as 30 homes in a year and emitted 25 tons of carbon dioxide.

What tool did the researcher help create to measure the environmental impact of AI training?

The researcher helped create CodeCarbon, a tool that runs in parallel to AI training code and estimates the energy consumption and carbon emissions.

What is the purpose of the "Have I Been Trained?" tool developed by Spawning.ai?

The "Have I Been Trained?" tool allows users to search massive datasets to see if their images or text have been used for training AI models without their consent.

What did the researcher find when searching for images of herself in the LAION-5B dataset?

The researcher found some images of herself from events she had spoken at, but also many images of other women named Sasha, including bikini models.

What tool did the researcher create to explore bias in image generation models?

The researcher created the Stable Bias Explorer, which allows users to explore the bias of image generation models through the lens of professions.

What did the researcher's tool find regarding the representation of gender and race in professions generated by AI models?

The tool found significant over-representation of whiteness and masculinity across 150 professions, even when compared to real-world labor statistics.

What was the researcher's response to the email claiming that their work would destroy humanity?

The researcher responded that focusing on AI's future existential risks is a distraction from its current, tangible impacts, and that the work should be focused on reducing these impacts now.

Outlines

00:00

🤖 AI Researcher's Bizarre Email: Humanity's Demise?

The speaker, an AI researcher, received an unusual email claiming that their work in AI would lead to the end of humanity. They acknowledge the growing prominence of AI in the news, with both positive (medical breakthroughs) and negative (biased chatbots, dangerous recipe suggestions) headlines. Despite the doomsday scenarios discussed, the researcher admits uncertainty about AI's long-term impacts but highlights pressing current issues like environmental costs, consent violations, and bias. They emphasize the need for transparency, disclosure, and tools to better understand AI's impacts.

05:01

🌎 Addressing AI's Environmental Impact and Copyright Infringement

The speaker discusses the environmental toll of training large language models like Bloom and GPT-3, which can consume immense amounts of energy and emit significant carbon dioxide. They highlight the lack of measurement and disclosure from tech companies, and the need for tools like CodeCarbon to estimate emissions. Additionally, they mention the 'Have I Been Trained?' tool, which allows artists to check if their work was used without consent for training AI models. This tool provided evidence for a class-action lawsuit against AI companies for copyright infringement. The speaker also discusses new opt-in/opt-out mechanisms for data sets, emphasizing that artwork shouldn't be an 'all-you-can-eat buffet' for training AI.

10:03

⚖️ Tackling AI Bias and Charting a Responsible Path Forward

The speaker addresses the issue of bias in AI systems, citing examples like facial recognition systems failing to detect women of color and wrongful accusations due to biased models. They introduce the 'Stable Bias Explorer' tool, which explores biases in image generation models across professions, revealing significant representation of whiteness and masculinity. The speaker emphasizes the importance of making AI accessible and understandable to people from all walks of life. They argue that by creating tools to measure AI's impact, companies, legislators, and users can make informed choices and develop guardrails to protect society and the planet. The speaker concludes by stating that focusing on AI's future existential risks is a distraction from addressing its current tangible impacts, and that we can collectively decide the direction we want to go.

Keywords

💡AI (Artificial Intelligence)

AI refers to the development of computer systems capable of performing tasks that typically require human intelligence, such as visual perception, speech recognition, decision-making, and language translation. The video discusses the impacts and implications of AI systems, both positive (like drug discovery) and negative (like biased or unethical outputs). It highlights the importance of understanding and mitigating AI's current tangible impacts on society and the environment.

💡Environmental Impact

The video emphasizes the significant environmental costs associated with training and deploying large AI models, particularly in terms of energy consumption and carbon emissions. It cites examples such as the Bloom language model emitting 25 tons of carbon dioxide during training, equivalent to driving a car around the world five times. The speaker advocates for transparency in reporting these impacts and developing tools like CodeCarbon to measure and reduce AI's environmental footprint.

💡Copyright and Consent

The video raises concerns about AI models being trained on copyrighted data, such as artwork and literature, without the consent of the creators. It discusses the 'Have I Been Trained?' tool, which allows artists and authors to check if their work has been used in training datasets. It also mentions a class-action lawsuit filed by artists against AI companies for copyright infringement, and efforts to create opt-in/opt-out mechanisms for data usage.

💡Bias

Bias refers to the encoding of stereotypes, racism, and sexism in AI models, often due to biases present in the training data or the algorithms themselves. The video cites examples of facial recognition systems failing to detect women of color, and image generation models perpetuating gender and racial biases in professions. It introduces the Stable Bias Explorer tool, which allows users to explore and understand biases in AI models across different professions.

💡Transparency and Accessibility

The video emphasizes the importance of transparency in AI development, by disclosing and measuring the impacts of AI models on society and the environment. It also advocates for making AI more accessible to non-technical users, so they can understand how these systems work and when they fail. Tools like the Stable Bias Explorer aim to make AI more accessible and understandable to people from all walks of life.

💡Ethics and Governance

The video highlights the ethical considerations surrounding AI development and deployment, such as the need for consent, environmental sustainability, and mitigating biases. It suggests that by measuring AI's impacts, companies and legislators can make informed decisions about which models to use or how to regulate AI in society. The speaker advocates for creating 'guardrails' to protect society and the planet from the negative impacts of AI.

💡Large Language Models

Large language models (LLMs) are AI systems trained on vast amounts of text data, enabling them to generate human-like text and perform language tasks like translation, summarization, and question-answering. The video discusses the environmental impact of training LLMs like Bloom and GPT-3, as well as the potential for these models to perpetuate biases or infringe on copyrights if trained on biased or unconsented data.

💡Existential Risk

Existential risk refers to the potential for advanced AI systems to pose a threat to human existence or civilization. The video mentions the concerns raised by some about the 'singularity' and 'doomsday scenarios' involving AI. However, the speaker argues that focusing on these future existential risks is a distraction from addressing the current, tangible impacts of AI on society and the environment.

💡Sustainability

Sustainability refers to the ability to meet present needs without compromising the ability of future generations to meet their own needs. In the context of AI, the video emphasizes the importance of developing sustainable AI models that minimize environmental impacts, respect copyrights, and mitigate biases. It suggests that by measuring and addressing these issues, we can create more trustworthy and sustainable AI systems for the future.

💡Tools and Metrics

The video highlights the development of various tools and metrics to measure and understand the impacts of AI systems, such as CodeCarbon (for environmental impact), 'Have I Been Trained?' (for copyright infringement), and the Stable Bias Explorer (for bias detection). These tools aim to provide transparency, accountability, and enable informed decision-making around AI development and deployment.

Highlights

AI models can contribute to climate change, use art and books without consent, and discriminate against entire communities, but we need to start tracking their impacts.

Training the open-source Bloom language model used as much energy as 30 homes in a year and emitted 25 tons of carbon dioxide, while models like GPT-3 emit 20 times more carbon.

Switching to a larger language model emits 14 times more carbon for the same task.

Introduced CodeCarbon, a tool to estimate the energy consumed and carbon emitted during AI training, to make informed sustainable choices.

Spawning.ai's 'Have I Been Trained?' tool lets artists search datasets to see if their work was used without consent to train AI models.

Artists Karla Ortiz and others used evidence from 'Have I Been Trained?' to file a class action lawsuit against AI companies for copyright infringement.

Spawning.ai partnered with Hugging Face to create opt-in and opt-out mechanisms for artists' work in training datasets.

AI models can encode biases and stereotypes, as demonstrated by facial recognition systems performing worse for women of color.

Porcha Woodruff was wrongfully accused of carjacking due to an AI system's misidentification.

Introduced the Stable Bias Explorer tool to explore biases in image generation models across professions.

Found significant representation of whiteness and masculinity across professions in image generation models, even in contrast with real-world statistics.

Presented the Stable Bias Explorer at a UN event as an example of making AI accessible and understandable to non-technical audiences.

Creating tools to measure AI's impact can inform companies, legislators, and users in making ethical choices and developing regulations.

Focusing on AI's future existential risks is a distraction from addressing its current, tangible impacts.

We are building the road as we walk it, and can collectively decide the direction for AI's development and deployment.

Transcripts

So I've been an AI researcher for over a decade.

And a couple of months ago, I got the weirdest email of my career.

A random stranger wrote to me

saying that my work in AI is going to end humanity.

Now I get it, AI, it's so hot right now.

(Laughter)

It's in the headlines pretty much every day,

sometimes because of really cool things

like discovering new molecules for medicine

or that dope Pope in the white puffer coat.

But other times the headlines have been really dark,

like that chatbot telling that guy that he should divorce his wife

or that AI meal planner app proposing a crowd pleasing recipe

featuring chlorine gas.

And in the background,

we've heard a lot of talk about doomsday scenarios,

existential risk and the singularity,

with letters being written and events being organized

to make sure that doesn't happen.

Now I'm a researcher who studies AI's impacts on society,

and I don't know what's going to happen in 10 or 20 years,

and nobody really does.

But what I do know is that there's some pretty nasty things going on right now,

because AI doesn't exist in a vacuum.

It is part of society, and it has impacts on people and the planet.

AI models can contribute to climate change.

Their training data uses art and books created by artists

and authors without their consent.

And its deployment can discriminate against entire communities.

But we need to start tracking its impacts.

We need to start being transparent and disclosing them and creating tools

so that people understand AI better,

so that hopefully future generations of AI models

are going to be more trustworthy, sustainable,

maybe less likely to kill us, if that's what you're into.

But let's start with sustainability,

because that cloud that AI models live on is actually made out of metal, plastic,

and powered by vast amounts of energy.

And each time you query an AI model, it comes with a cost to the planet.

Last year, I was part of the BigScience initiative,

which brought together a thousand researchers

from all over the world to create Bloom,

the first open large language model, like ChatGPT,

but with an emphasis on ethics, transparency and consent.

And the study I led that looked at Bloom's environmental impacts

found that just training it used as much energy

as 30 homes in a whole year

and emitted 25 tons of carbon dioxide,

which is like driving your car five times around the planet

just so somebody can use this model to tell a knock-knock joke.

And this might not seem like a lot,

but other similar large language models,

like GPT-3,

emit 20 times more carbon.

But the thing is, tech companies aren't measuring this stuff.

They're not disclosing it.

And so this is probably only the tip of the iceberg,

even if it is a melting one.

And in recent years we've seen AI models balloon in size

because the current trend in AI is "bigger is better."

But please don't get me started on why that's the case.

In any case, we've seen large language models in particular

grow 2,000 times in size over the last five years.

And of course, their environmental costs are rising as well.

The most recent work I led, found that switching out a smaller,

more efficient model for a larger language model

emits 14 times more carbon for the same task.

Like telling that knock-knock joke.

And as we're putting in these models into cell phones and search engines

and smart fridges and speakers,

the environmental costs are really piling up quickly.

So instead of focusing on some future existential risks,

let's talk about current tangible impacts

and tools we can create to measure and mitigate these impacts.

I helped create CodeCarbon,

a tool that runs in parallel to AI training code

that estimates the amount of energy it consumes

and the amount of carbon it emits.

And using a tool like this can help us make informed choices,

like choosing one model over the other because it's more sustainable,

or deploying AI models on renewable energy,

which can drastically reduce their emissions.

But let's talk about other things

because there's other impacts of AI apart from sustainability.

For example, it's been really hard for artists and authors

to prove that their life's work has been used for training AI models

without their consent.

And if you want to sue someone, you tend to need proof, right?

So Spawning.ai, an organization that was founded by artists,

created this really cool tool called “Have I Been Trained?”

And it lets you search these massive data sets

to see what they have on you.

Now, I admit it, I was curious.

I searched LAION-5B,

which is this huge data set of images and text,

to see if any images of me were in there.

Now those two first images,

that's me from events I've spoken at.

But the rest of the images, none of those are me.

They're probably of other women named Sasha

who put photographs of themselves up on the internet.

And this can probably explain why,

when I query an image generation model

to generate a photograph of a woman named Sasha,

more often than not I get images of bikini models.

Sometimes they have two arms,

sometimes they have three arms,

but they rarely have any clothes on.

And while it can be interesting for people like you and me

to search these data sets,

for artists like Karla Ortiz,

this provides crucial evidence that her life's work, her artwork,

was used for training AI models without her consent,

and she and two artists used this as evidence

to file a class action lawsuit against AI companies

for copyright infringement.

And most recently --

(Applause)

And most recently Spawning.ai partnered up with Hugging Face,

the company where I work at,

to create opt-in and opt-out mechanisms for creating these data sets.

Because artwork created by humans shouldn’t be an all-you-can-eat buffet

for training AI language models.

(Applause)

The very last thing I want to talk about is bias.

You probably hear about this a lot.

Formally speaking, it's when AI models encode patterns and beliefs

that can represent stereotypes or racism and sexism.

One of my heroes, Dr. Joy Buolamwini, experienced this firsthand

when she realized that AI systems wouldn't even detect her face

unless she was wearing a white-colored mask.

Digging deeper, she found that common facial recognition systems

were vastly worse for women of color compared to white men.

And when biased models like this are deployed in law enforcement settings,

this can result in false accusations, even wrongful imprisonment,

which we've seen happen to multiple people in recent months.

For example, Porcha Woodruff was wrongfully accused of carjacking

at eight months pregnant

because an AI system wrongfully identified her.

But sadly, these systems are black boxes,

and even their creators can't say exactly why they work the way they do.

And for example, for image generation systems,

if they're used in contexts like generating a forensic sketch

based on a description of a perpetrator,

they take all those biases and they spit them back out

for terms like dangerous criminal, terrorists or gang member,

which of course is super dangerous

when these tools are deployed in society.

And so in order to understand these tools better,

I created this tool called the Stable Bias Explorer,

which lets you explore the bias of image generation models

through the lens of professions.

So try to picture a scientist in your mind.

Don't look at me.

What do you see?

A lot of the same thing, right?

Men in glasses and lab coats.

And none of them look like me.

And the thing is,

is that we looked at all these different image generation models

and found a lot of the same thing:

significant representation of whiteness and masculinity

across all 150 professions that we looked at,

even if compared to the real world,

the US Labor Bureau of Statistics.

These models show lawyers as men,

and CEOs as men, almost 100 percent of the time,

even though we all know not all of them are white and male.

And sadly, my tool hasn't been used to write legislation yet.

But I recently presented it at a UN event about gender bias

as an example of how we can make tools for people from all walks of life,

even those who don't know how to code,

to engage with and better understand AI because we use professions,

but you can use any terms that are of interest to you.

And as these models are being deployed,

are being woven into the very fabric of our societies,

our cell phones, our social media feeds,

even our justice systems and our economies have AI in them.

And it's really important that AI stays accessible

so that we know both how it works and when it doesn't work.

And there's no single solution for really complex things like bias

or copyright or climate change.

But by creating tools to measure AI's impact,

we can start getting an idea of how bad they are

and start addressing them as we go.

Start creating guardrails to protect society and the planet.

And once we have this information,

companies can use it in order to say,

OK, we're going to choose this model because it's more sustainable,

this model because it respects copyright.

Legislators who really need information to write laws,

can use these tools to develop new regulation mechanisms

or governance for AI as it gets deployed into society.

And users like you and me can use this information

to choose AI models that we can trust,

not to misrepresent us and not to misuse our data.

But what did I reply to that email

that said that my work is going to destroy humanity?

I said that focusing on AI's future existential risks

is a distraction from its current,

very tangible impacts

and the work we should be doing right now, or even yesterday,

for reducing these impacts.

Because yes, AI is moving quickly, but it's not a done deal.

We're building the road as we walk it,

and we can collectively decide what direction we want to go in together.

Thank you.

(Applause)