AI Summary
Sriram Natarajan recounts how an AI-powered app created a convincing fake video of him speaking Hindi, which sparked his concern about the ethical risks of generative AI. He illustrates the dangers with real-world examples, from deepfake videos to biased training data, and offers practical ways for users to engage responsibly. The talk encourages a collective commitment to questioning and curating AI so it serves human potential rather than undermining it.
Chapters
The speaker, who never learned Hindi, used an app that transformed a video of him into a convincing fluent Hindi speaker. While amazed, he immediately recognized the potential for misuse in spreading false information.
Generative AI can mimic any writer's style, produce whole novels without attribution, doctor images with harmful intent, and create fake videos that blur the line between fiction and reality.
Students in Carmel, New York, created a fake video of a school principal saying racist, vulgar, and aggressive things. The video spread across social media and seriously damaged the school's reputation.
The speaker shares that his cousin, after practicing Kundalini Yoga learned only from online videos, suffered mental instability and isolation, stopped eating for days, and ultimately met a tragic end. This underscores how unguided online content can cause real harm.
Most generative AI systems are experimental and trained on publicly available data, licensed data, and human-curated content. Because public data contains biases, misinformation, and inaccuracies, AI systems learn those same flaws.
Users should seek AI systems that prioritize safety, provide verifiable answers, and show how they work. One practical approach is to ask the AI to explain its safety principles and policies instead of reading dense legal documents.
The speaker advises asking AI point-blank, 'When should I not trust you?' Knowing when to abstain from using the tool is as important as knowing how to use it.
Flagging concerning AI outputs is a vital part of improving the technology. Unlike social media, where flagging reports violations, flagging in AI systems contributes to maturing the tools and keeping them on the right track.
By evaluating AI-generated content, probing for sources, and giving feedback, users can turn generative AI into a 'superpower' that amplifies human potential rather than obscuring it.
Mentioned in this Video
Study Flashcards (8)
What did Sriram Natarajan discover an app could do with a video of him?
easy
Click to reveal answer
What did Sriram Natarajan discover an app could do with a video of him?
It turned a video of him with a Hindi script into a realistic video where he appeared to speak fluent Hindi.
00:03
What harmful fake video did students in Carmel, New York create?
easy
Click to reveal answer
What harmful fake video did students in Carmel, New York create?
They created a fake video of a school principal saying racist, vulgar, and aggressive things, which spread on social media.
01:56
What happened to the speaker's cousin after practicing online Kundalini Yoga?
medium
Click to reveal answer
What happened to the speaker's cousin after practicing online Kundalini Yoga?
He experienced mental instability, became isolated from family, stopped eating for days, and ended up having a tragic outcome.
02:51
What are most generative AI systems currently trained on?
medium
Click to reveal answer
What are most generative AI systems currently trained on?
Publicly available data, mixed with licensed data and content curated by human trainers.
03:38
What question should you ask an AI system to understand when not to use it?
easy
Click to reveal answer
What question should you ask an AI system to understand when not to use it?
When should I not trust you?
05:05
How does flagging outputs in AI systems differ from flagging on social media?
hard
Click to reveal answer
How does flagging outputs in AI systems differ from flagging on social media?
In AI systems, flagging helps mature the tool and keep it on the right track, rather than simply reporting rule violations.
05:36
What two questions should you ask an AI to ensure transparency?
medium
Click to reveal answer
What two questions should you ask an AI to ensure transparency?
Where did this info come from? and How did you arrive at this?
06:01
What can you ask an AI to understand its strengths, weaknesses, and trustworthiness?
easy
Click to reveal answer
What can you ask an AI to understand its strengths, weaknesses, and trustworthiness?
Ask it to explain its capabilities, limitations, and training data sources.
06:28
💡 Key Takeaways
Fake principal video
A concrete case where generative AI was used to create weaponized misinformation that damaged a real school's reputation.
01:56The personal cost of unguided online content
The speaker's cousin's tragic story shows that even non-AI online content can have devastating real-world consequences, amplifying the stakes of the talk.
02:51Flawed training data
Explains the root cause of AI bias: systems learn from public data that already contains societal biases and misinformation.
03:38When should I not trust you?
A simple, memorable question any user can ask to gauge an AI's limitations and abstain when necessary.
05:05AI as a superpower
A closing quote that reframes responsible AI use as a way to amplify, not obscure, human potential.
06:28Full Transcript
[00:03] growing up in India I missed out on learning Hindi the language that connects more than half a billion people and when I finally tried speaking
[00:17] people and when I finally tried speaking it well let's just say my first attempts were like a funny dance routine a mix of confusion tongue twisters and EXP
[00:30] Impressions that probably left native speakers scratching their speakers scratching their heads but then my journey took a Twist I stumbled upon an app that magically turned a video of me and a
[00:45] magically turned a video of me and a popular Hindi script into a flu and fake popular Hindi script into a flu and fake video suddenly there I was a fluent video suddenly there I was a fluent Hindi speaker on screen like magic
[01:00] I was blown away by the result it's a testament to how far come yet right then and there a bunch of red
[01:12] flags popped up in my mind this was about the flip side how easily could this be misused to
[01:24] spread false information and that concern seems valid when we look at how generative artificial intelligence is increasingly becoming an artistic Creator it can now mimic any writer's
[01:40] Creator it can now mimic any writer's Unique Style churning out entire novels without attribution it can doctor images of people with harmful intent and it can create fake videos
[01:56] that blur the lines between fiction and reality school students in Carmel New York created a fake video of a principal from
[02:11] created a fake video of a principal from a nearby School saying racist vulgar and aggressive things the video storm through social media damaging the school's reputation as someone who's deeply
[02:26] involved in the world of AI it hit me hard to see the technology I'm so passionate about being used for deception and manipulation it's like a dark sequel to the social media
[02:39] Saga where misinformation can actually harm real harm real people and I've seen this
[02:51] ago my cousin started practicing Kundalini Yoga by just watching online videos and without any guidance it led him down a path of mental
[03:05] mental instability and isolation from family he even went without food for days days ultimately leading to his tragic
[03:22] creating harmful or misguided content with just few text promps with the power of social media to spread it we have a recipe for
[03:38] disaster the reality today is that most generative AI systems remain in research or experimental modes and they are trained on publicly available data mixed with some licensed data and content curated by human
[03:52] trainers what's deeply concerning is that public information often contains societal biases misinformation and inaccuracies baked right in and these systems learn from it with all its flaws and
[04:11] thing these systems acknowledge the possibility of generating inaccurate or offensive information showcasing how easily it can information showcasing how easily it can be harnessed for malicious
[04:24] intent so when we engage with these systems it is our Collective responsibility to be curious we should go for AI systems that
[04:36] make safety a priority give us answers that we can validate and let us see how they work all while staying
[04:50] by having conversations with them ask an AI system to explain its safety principles and policies and and see how it responds instead of slogging through complex terms and services and privacy
[05:05] policy documents we now have a chance to get insights through its summarization get insights through its summarization capabilities also ask it point blank when should I not trust you knowing when to abstain from using
[05:22] these tools is just as vital and prepares you to make informed decisions raises concerns simply hit that flag
[05:36] button unlike social media where flagging content is used to report rule violations in AI systems your voice plays a vital role in maturing these
[05:49] plays a vital role in maturing these tools and helps them keep on the right content take a moment to ask questions like where did this info come
[06:01] like where did this info come from or how did you arrive at this just like with a friend insisting on transparency helps ensure accuracy and avoids spreading false
[06:16] information so by carefully evaluating generative AI systems the content they produce and by providing feedback we can harness AI as our
[06:28] feedback we can harness AI as our superpower generative AI system have a conversation with it ask it to explain its
[06:40] capabilities limitations and training data sources allowing you to better understand its strengths weaknesses and
[06:52] trustworthiness that's your immediate impact and it can truly shape the impact and it can truly shape the experience not just for you but for
[07:05] Channel toward a future where ai's potential amplifies rather than obscures the obscures the Brilliance of human potential
