Skip to content

Google IO 2023 AlphaFold Fake Image Check and Universal Translator With Dubbing And Lip Synching

Jump to bottom
FurkanGozukara edited this page Oct 24, 2025 · 1 revision

Google I/O 2023: AlphaFold, Fake Image Check & Universal Translator With Dubbing And Lip Synching

Full tutorial link > https://www.youtube.com/watch?v=rrBS9rcIcWk

Google I/O 2023: AlphaFold, Fake Image Check & Universal Translator With Dubbing And Lip Synching

image Hits Patreon BuyMeACoffee Furkan Gözükara Medium Codio Furkan Gözükara Medium

YouTube Channel Furkan Gözükara LinkedIn Udemy Twitter Follow Furkan Gözükara

🔥Hot off the stage from #GoogleIO, hear James, the leader of a new area at Google called Technology and Society, discuss the fascinating interplay between #AI and society.

In this video, James addresses the need for #responsibleAI and how Google intends to tackle it. From the brilliant successes of Google's DeepMind's #AlphaFold, to the complex challenges of misinformation and deepfakes, he discusses Google's commitment to addressing the benefits and risks of AI with their principles established in 2018.

He shares exciting developments in evaluating online information and introduces innovative tools like 'About This Image' in Google Search, Google Lens, and watermarking in generative models. He also presents an impressive example of Universal Translate, an experimental AI video dubbing service, and discusses its potential benefits and inherent risks.

Moreover, he talks about Google's initiatives in mitigating toxic outputs of AI models, including automated adversarial testing and the use of Perspective API.

Watch to find out how Google plans to work with researchers, social scientists, industry experts, governments, creators, publishers, and everyday people to build a responsible AI ecosystem.

Stay tuned till the end for a handoff to Samir, who will share exciting developments in Android. Enjoy the watch! 🚀

Source Google. Full Event (10 May 2023) ⤵️

https://www.youtube.com/watch?v=cNfINi5CNbY

Our Discord server ⤵️

https://bit.ly/SECoursesDiscord

If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on 🥰 ⤵️

https://www.patreon.com/SECourses

Technology & Science: News, Tips, Tutorials, Tricks, Best Applications, Guides, Reviews ⤵️

https://www.youtube.com/playlist?list=PL_pbwdIyffsnkay6X91BWb9rrfLATUMr3

Playlist of StableDiffusion Tutorials, Automatic1111 and Google Colab Guides, DreamBooth, Textual Inversion / Embedding, LoRA, AI Upscaling, Pix2Pix, Img2Img ⤵️

https://www.youtube.com/playlist?list=PL_pbwdIyffsmclLl0O144nQRnezKlNdx3

#Google #GoogleIO #TechnologyandSociety #DeepMind #AI #ResponsibleAI #AIprinciples #Deepfakes #Misinformation #UniversalTranslate #AdversarialTesting #PerspectiveAPI #Android #JamesAtGoogleIO #SamirAtGoogleIO

Video Transcription

  • 00:00:09 Hi everyone. I'm James. 

  • 00:00:12 In addition to research, I lead a new area  at Google called technology and society. 

  • 00:00:20 Growing up in Zimbabwe, I could not have imagined  all the amazing and groundbreaking innovations 

  • 00:00:26 that have been presented on this stage today. While I feel it's important to celebrate the  

  • 00:00:33 incredible progress in AI and the immense potential that it has for people and society  

  • 00:00:38 everywhere, we must also acknowledge that it's an emerging technology that is still being  

  • 00:00:44 developed and there's still so much more to do. 

  • 00:00:49 Earlier you heard Sundar say that our approach  to AI must be both bold and responsible. 

  • 00:00:56 While there's a natural tension between the  two, we believe it's not only possible but 

  • 00:01:03 in fact critical to embrace  that tension productively. 

  • 00:01:08 The only way to be truly bold in the long  term is to be responsible from the start. 

  • 00:01:15 Our field defining research is helping scientists  make bold advances in many scientific fields, 

  • 00:01:22 including medical breakthroughs. Take for example Google DeepMind's AlphaFold,  

  • 00:01:27 which can accurately predict the 3D shapes of 200 million proteins. 

  • 00:01:33 That's nearly all the catalogue  proteins known to science. 

  • 00:01:37 AlphaFold gave us the equivalent of nearly  400 million years of progress in just weeks. 

  • 00:01:52 So far more than one million researchers around  the world have used AlphaFold's predictions, 

  • 00:01:58 including Feng Xiang's pioneering lab at  the Broad Institute of MIT and Harvard. 

  • 00:02:08 In fact in March this year, Zhang and his  colleagues at MIT announced that they'd used 

  • 00:02:15 AlphaFold to develop a novel molecular syringe  which could deliver drugs to help improve 

  • 00:02:21 the effectiveness of treatments  for diseases like cancer. 

  • 00:02:32 And while it's exhilarating to see such bold  and beneficial breakthroughs, AI also has 

  • 00:02:39 the potential to worsen existing societal  challenges like unfair bias, as well as pose 

  • 00:02:46 new challenges as it becomes more  advanced and new uses emerge. 

  • 00:02:51 That's why we believe it's imperative  to take a responsible approach to AI. 

  • 00:02:57 This work centers around our AI principles  that we first established in 2018. 

  • 00:03:03 These principles guide product development  and they help us assess every AI application. 

  • 00:03:10 They prompt questions like, will it be socially  beneficial or could it lead to harm in any 

  • 00:03:17 way? One area  

  • 00:03:20 that is top of mind for us is misinformation. Generative AI makes it easier than ever to create  

  • 00:03:28 new content, but it also raises additional questions about its trustworthiness. 

  • 00:03:35 That's why we're developing and providing people  with tools to evaluate online information. 

  • 00:03:40 For example, have you come across a photo  on a website or one shared by a friend with 

  • 00:03:48 very little context, like this one at the  moon landing, and found yourself wondering, 

  • 00:03:53 is this reliable? I have and I'm sure many of you have as well. 

  • 00:03:59 In the coming months, we're adding two  new ways for people to evaluate images. 

  • 00:04:05 First, with our About This  Image tool in Google Search,  

  • 00:04:09 you'll be able to see important information 

  • 00:04:12 such as when and where similar images may  have first appeared, where else the image 

  • 00:04:19 has been seen online, including news, fact  checking, and social sites, all this providing 

  • 00:04:25 you with helpful context to  determine if it's reliable. 

  • 00:04:30 Later this year, you'll also be able to use  it if you search for an image or screenshot 

  • 00:04:34 using Google Lens, or when  you're on websites in Chrome. 

  • 00:04:41 As we begin to roll out the generative image  capabilities, like Sundar mentioned, we will 

  • 00:04:47 ensure that every one of our AI generated  images has metadata and markup in the original 

  • 00:04:54 file to give you context if you come  across it outside of our platforms. 

  • 00:05:00 Not only that, creators and publishers will  be able to add similar metadata, so you'll 

  • 00:05:06 be able to see a label in images in Google  Search marking them as AI generated. 

  • 00:05:21 As we apply our AI principles, we also start  to see potential tensions when it comes to 

  • 00:05:28 being bold and responsible. Here's an example. 

  • 00:05:32 Universal Translate is an experimental AI video  dubbing service that helps experts translate 

  • 00:05:39 a speaker's voice while also  matching their lip movements. 

  • 00:05:44 Let me show you how it works with an online  college course created in partnership with 

  • 00:05:49 Arizona State University. What many college students don't realize  

  • 00:05:54 is that knowing when to ask for help and then following through on using helpful resources is  

  • 00:05:59 actually a hallmark of becoming a productive adult. 

  • 00:06:19 We use next generation translation models to  translate what the speaker is saying, models 

  • 00:06:26 to replicate the style and the tone, and  then match the speaker's lip movements. 

  • 00:06:31 Then we bring it all together. 

  • 00:06:34 This is an enormous step forward for learning  comprehension, and we're seeing promising 

  • 00:06:39 results with course completion rates. But there's an inherent tension here. 

  • 00:06:45 You can see how this can be incredibly beneficial,  but some of the same underlying technology 

  • 00:06:52 could be misused by bad  actors to create deep fakes. 

  • 00:06:57 So we've built this service with  guardrails to help prevent misuse,  

  • 00:07:01 and we make it accessible only to authorized partners. 

  • 00:07:12 And as Sundar mentioned, soon we'll be integrating  new innovations in watermarking into our latest 

  • 00:07:19 generative models to also help with  the challenge of misinformation. 

  • 00:07:25 Our AI principles also help  guide us on what not to do. 

  • 00:07:31 For instance, years ago, we were the first  major company to decide not to make a general 

  • 00:07:37 purpose facial recognition  API commercially available. 

  • 00:07:42 We felt there weren't  adequate safeguards in place. 

  • 00:07:46 Another way we live up to our AI principles  is with innovations to tackle challenges as 

  • 00:07:52 they emerge, like reducing the risk of problematic  outputs that may be generated by our models. 

  • 00:07:59 We are one of the first in the industry to  develop and launch automated adversarial testing 

  • 00:08:06 using large language model technology. We do this for queries like this  

  • 00:08:12 to help uncover and reduce  inaccurate outputs, like the one 

  • 00:08:16 on the left, and make them  better, like the one on the right. 

  • 00:08:21 We're doing this at a scale that's never been  done before at Google, significantly improving 

  • 00:08:26 the speed, quality, and coverage of  testing, allowing safety experts to  

  • 00:08:32 focus on the most difficult cases. 

  • 00:08:36 And we're sharing these innovations with others. For example, our Perspective API, originally  

  • 00:08:42 created to help publishers mitigate toxicity, is now being used in large language models. 

  • 00:08:49 Scientific researchers have used our Perspective  API to create an industry evaluation standard. 

  • 00:08:56 And today, all significant large language models,  including those from OpenAI and Anthropic, 

  • 00:09:02 incorporate this standard to evaluate  toxicity generated by their own models. 

  • 00:09:18 Building AI responsibly must be a collective  effort involving researchers, social scientists, 

  • 00:09:26 industry experts, governments,  and everyday people. 

  • 00:09:31 As well as creators and publishers, everyone  benefits from a vibrant content ecosystem 

  • 00:09:38 today and in the future. 

  • 00:09:41 That's why we're getting feedback and we'll  be working with the web community on ways 

  • 00:09:46 to give publishers choice and  control over their web content. 

  • 00:09:52 It's such an exciting time. There's so much we can accomplish  

  • 00:09:57 and so much we must get right together. We look forward to working with all of you. 

  • 00:10:04 And now I'll hand it off to Samir, who will  speak to you about all the exciting developments 

  • 00:10:09 we're bringing to Android. Thank you.

Clone this wiki locally