Any of you that have been involved with AI or been utilizing know that deepfakes are an ongoing issue, so we’re bringing in someone who can speak in detail about this topic. Joris Mollinga is the co-founder of Duck Duck Goose, a platform that offers advanced solutions to protect against deep fake videos and images. Duck Duck Goose is on a mission to detect deep fakes and filter out what to believe from what they perceive. It aims to realize a digital environment in which we can trust what we see. The emergence of deep fakes means that everything we see digitally can be manipulated. To ensure everyone’s autonomy to determine what’s true, they’re developing insightful deep fake detection software. Tune in as Joris shares how they are exposing deep fakes and adding trust to this uncertain new AI world.
- Meet Joris Mollinga, an AI specialist and co-founder at DuckDuckGoose.
- Learn about his journey from aerospace engineering to artificial intelligence.
- Discover how DuckDuckGoose aims to create a trustworthy digital environment amidst the rise of manipulated content known as ‘deep fakes’.
- Understand their unique software that classifies images or videos as real or fake while providing insights on the classification process.
- Explore how distrust in AI is growing due to phenomena like deep fakes.
- Get insight into an ‘AI vs AI battle’ where generative AI creates/manipulates digital footage while systems like those developed by DuckDuckGoose identify whether it’s original or fabricated content.
- Uncover who benefits most from deep fake detection products for digital identity verification.
- Learn about potential applications across various industries including video conferencing and news media.
- Delve into how selfies/videos are used as input for deep fake classification.
- “Reality is never strange. We are the ones that are strange.” –Joris Mollinga
- “Duck Duck Goose aims to realize a digital environment in which we can trust what we see. The emergence of deep fakes means that everything we see digitally can be manipulated.” –Joris Mollinga
- “The first gentleman that told me about ChatGPT was in November 2022. The thing that stuck with me is he said, ‘In 90 days, everybody that touches the industry or is intact is going to be using and know about this. Ninety days after that, the whole world will know.'” –Ron Levy
- “I have a couple of resources I like to share. As a student, I relied heavily upon a book called Pattern Recognition and Machine Learning by Bishop, which was what we called the Bible of Machine Learning.”
Listen to the podcast here
AI vs. AI In The Fight Against Deep Fakes, Feat. Joris Mollinga Of DuckDuckGoose
You are about to embark on another stellar journey. It’s brought to you by the Edge of AI, empowering pioneers of Web3 tech and culture, responsible for other groundbreaking endeavors like the Outer Edge LA Innovation Festival and the Edge of NFT Podcast. Find out more at EdgeOf.xyz, EdgeOfNFT.com, and OuterEdge.live.
My name is Joris Mollinga, Cofounder at Duck Duck Goose, where we are on a mission to detect deep fakes and filter out what to believe from what we perceive. I’m on the Edge of AI, a fantastic place to learn about AI. You better believe it. Stay tuned.
AI passengers, jump on in. Here’s what’s to come on this journey. Find out how our guest is exposing deep fakes and adding trust to this uncertain new AI world. Why? If you think the reality is strange, you are wrong. It’s you that’s strange. From graduate trained AI expert, learn what book is known as the AI Bible. Obvious and more, take your seat.
I’m excited about our guest. It’s with Joris Mollinga of Duck Duck Goose. It is a platform that offers advanced solutions to protect against deep fake videos and images. Any of you that have been involved with AI or been utilizing know this is an ongoing issue. He is an AI specialist who majored in Artificial Intelligence before cofounding Duck Duck Goose. In his role as a Cofounder, he oversees the product and development teams, driving innovation and leveraging his experience and expertise to keep the company at the forefront of technology.
With a strong focus on ethical AI practices, Joris advocates for the responsible development and use of AI systems that benefit society. We know the industry is rife and full of unintended consequences, and it takes guys like Joris to help keep us between the lines. Beyond his professional accomplishments, he also finds fulfillment and mentoring aspiring AI enthusiasts and engaging in charitable initiatives.
Duck Duck Goose aims to realize a digital environment in which we can trust what we see. The emergence of deep fakes means that everything we see digitally can be manipulated. To ensure everyone’s autonomy to determine what’s true, they’re developing insightful deep fake detection software.
The software is not only able to classify a certain image or video as real or fake but also provide insight into why that content has been classified as manipulated. Their software is available as a forensic analysis tool but also via their website for free so that everyone can see the difference. Let’s start with a little background, Joris. You completed your graduate work in AI before cofounding Duck Duck Goose. What’s it been like behind the scenes in AI academia and industry over the past few years as the developments have unfolded? How did you end up becoming a cofounder?
Thanks for having me on the show. It’s been an exciting journey for me in both AI academia and the AI industry in the past years. As an AI specialist, I had the opportunity to immerse myself in cutting-edge research, exploring the potential and challenges of artificial intelligence. It’s inspiring how AI is transforming the way we interact with technology and systems around us.
My background is not in Computer Science or AI. I’m originally an aerospace engineer. As a kid, I always wanted to become a pilot. That didn’t work out for various reasons but I ended up doing aerospace engineering instead, which was interesting. What I found most interesting during this aerospace degree was a few elective courses I had on artificial intelligence.
During those elective courses, some deep learning 101, basic tinkering with deep neural networks, and solving some easy or basic regression or classification problems got me inspired to pursue a career in AI. That’s when I decided, once I completed my bachelor’s, to pursue a Master’s degree in AI at the University of Amsterdam.
I graduated at the beginning of 2020 before COVID hit and founded Duck Duck Goose shortly thereafter. At Duck Duck Goose, we aim to realize a digital environment where we can still believe what we perceive. You set this yourself. At the moment, this is threatened by the availability of synthetic media. Synthetic media is a broad term. Within that scope, we focus on deep fakes. At Duck Duck Goose, we provide deep fake detection products and services.
Before we go deeper into that, I want to get a little foundation of Duck Duck Goose and yourself. We know your background. You’ve told us you went to the University of Amsterdam and you have your Master’s degree from there, which is fantastic. You live in the Netherlands.
I live in Amsterdam.
That’s also a great thing about our industry but over the last couple of decades, we’ve been able to have companies dispersed all over the world. It’s seamless as we’re sitting here. I’m in the Los Angeles area, you are in Amsterdam, and here we are. That’s how you operate your business. Give us a quick snapshot of Duck Duck Goose. In other words, are you operating in a physical facility? How many people do you have? You don’t have to go too deep but how did you guys get together? Everybody is going to wonder about the name.
The question about the name, we get asked a lot, as you can imagine. I met my two founders, Mark and Parya, at the beginning of 2020. There was a mutual connection who laid the introduction. Duck Duck Goose is a spinoff of a university project. I joined the company at the moment that it was spun off from the university. We started with the three of us. The team consists of eleven team members. We’ve grown steadily in the past years. We have an office in the Netherlands where our team is dedicated to providing robust and reliable deep fake detection products and services.
The University of Amsterdam has a Master’s class in AI. I’m going to suggest there can’t be many of those out there. Is it unique or one of the first movers?
No, there are not too many of them out there. They have been around for quite a long time. Especially within Europe, the Amsterdam machine learning lab, with a couple of well-known professors, is quite highly regarded.
There’s a term called genetic algorithms. Can you hit that for me a little bit?
The genetic algorithm is an optimization technique that is inspired by the process of nature or the principle of natural selection. It aims to solve complex optimization problems by mimicking the principles of evolution, like mutation or survival of the fittest. Over time, this algorithm or process converges towards better solutions, enabling genetic algorithms to efficiently explore large solution spaces and come up with optimal or near-optimal solutions. In these elective courses, I took Machine Learning and AI. There was some tinkering with genetic algorithms. I found that fascinating. That inspired me to pursue this career in AI.
For the audience, this is going to culminate in an understanding of what Duck Duck Goose is. It is amazing. We’ve all seen the wow moments of, “This is a deep fake. Can you believe it?” We know that that’s out there. As AI is snowballing every single day and finding its way into everything we touch and do, the distrust is going to grow more. Duck Duck Goose is a solution to that problem. It’s amazing to me.
Within that, as we move forward, there are going to be some screen shares. I do realize some of the audience will be reading. We’ll try and describe what we’re looking at to the best we can. Why don’t you talk about how you help detect deep fake threats? How did you decide to pursue this value proposition?
At Duck Duck Goose, we leverage the power of AI for deep fake detection. It’s an AI versus AI battle because it’s generative AI creating or manipulating digital footage and our AI being able to distinguish real footage from deep fake footage. We’ve trained systems to tell the difference between a deep fake image and a normal image by training our software on data sets of real and deep fake images. These data sets are a combination of proprietary data sets that we develop ourselves and publicly available open-source datasets.At Duck Duck Goose, we leverage the power of AI for deep fake detection. It's an AI versus AI battle. Click To Tweet
The AI is training itself once you get it going. It keeps reaching out and getting more data. I assume that’s the way it’s working.
We still need to feed the data for AI to keep improving itself. You are correct in the sense that it is becoming better over time. That is a requirement because deep fake technology is also evolving at a rapid pace and generative AI technology is exploding. A couple of years ago, no one heard of the term narrative AI but it’s exploding at the moment. We need to keep up-to-date with the latest developments in generative AI.
What type of clients are you chasing? What is your optimal client? Is it institutional clients? Is it for the masses individually? Give me an idea of, at least on the roadmap, where these starting points are.
We focus our deep fake detection products and services on digital identity verification. Any process where an individual has to authenticate or identify his or herself with a selfie or a selfie video, these systems can be spoofed by presenting this liveness or these digital identity verification systems with a deep fake. This penetration testing is also something that we do at Duck Duck Goose to get a foot in the door to kickstart the conversation with potential clients. We sell to digital identity verification, which is mainly well-established large companies providing liveness and digital identity verification check.
However, this is only our short-term focus. As a startup, you need to start somewhere. We chose digital identity verification. As our product and software can be applied to any digital footage, images, and video, which is everything because we spend how much time of our day on our smartphone looking at videos and things on social media, digital identity verification is not the end game. There are tons of different industries we’d like to apply our solution to, for example, video conferencing or news and media, to name a few.
Is it facial recognition or retina? Is it anything specific? Where are you starting on that?
It’s more than facial recognition. What we take as input is a selfie of someone trying to identify himself or herself in some onboarding process for a bank account, a cryptocurrency broker account, a dating app, or a shared scooter. We take a selfie or a selfie video as input. From there, we do a deep fake classification. We receive the inputs. A couple of seconds later, we give our users a result that consists of two things.
First of all, a binary classification, whether we think it’s authentic or deep fake, which is expressed as a probability score. Second, we show what regions of the image are deemed suspicious by our AI and what pixels contribute most to the decision of our software. This serves as a tool to guide the user in substantiating his or her decision, whether rejecting or accepting this selfie.
I would suggest there are a lot of people looking into or are somewhat involved that may not know the difference between machine learning AI and generative AI. Can you hit that a little bit for us to educate? It’s rare for us to have the level of expertise you own to give us complete clarity.
Generative AI is a subset of artificial intelligence that focuses on creating new content rather than analyzing or recognizing patterns in existing data. It uses machine learning models, particularly generative models, to generate new data that resembles the training data it was trained upon. This capability makes it incredibly powerful and versatile. Common applications of generative AI are models that can produce images of people or text-to-image models where you give it a prompt and it creates half a dozen images resembling the text prompt but also audio or voice that one could generate with generative AI models.
I’ll say about generative AI that we’re most of us are familiar with. It was invented in November of 2022. That’s technically not accurate. AI has been developing for many years or decades. It’s gotten a snowball-rolling effect. When I say it was invented in November 2022, that’s when the masses were able to touch it and use it. I commonly joke that the internet was invented in 2004. What I mean there is that’s when people got high speed. Before that, it was interesting and a chore but it was at that moment that it came to the masses. It’s open AI and it’s November 2022. Do you want to speak to that a little bit or tell me if I’m off base?
You’re referring to the release of ChatGPT in November 2022. That was the first application of generative AI, which went mainstream and was out of the blue. It is very easy for the mainstream to use before then. The technology was not new. Large language models, the technology behind ChatGPT, or the type of neural networks behind ChatGPT have existed already for a couple of years. They were not new but this mainstream adoption on November 2022 shook the world quite a bit.
The first gentleman that told me about ChatGPT was in November 2022. The thing that stuck with me is he said, “In 90 days, everybody that touches the industry or is intact is going to be using and know about this. Ninety days after that, the whole world will know.” His timeframe was about dead on. With that snowball movement, you can’t even imagine what it’s going to be 6 months or 1 year from now, let alone further out, which is why Duck Duck Goose is critically important. You got to be able to tell real from fake. That’s what you guys do. Let’s circle back. You never did tell us where the name came from.
Duck Duck Goose is a children’s game where players sit in a circle facing inwards. There’s one child who is it. That child walks around, tapping on the shoulder of the children and sitting in the circle, saying, “Duck.” Until finally choosing a goose and the goose becomes a designated chaser. The children run and the goose has to chase one of the children and then the game restarts. The metaphor here is that the goose is hidden within the ducks like deep fake images and videos are hidden within the authentic image and video material. Hence the name Duck Duck Goose.
Let’s do a screen share and take a look at what you guys do. Show us a demo. I’ll be interrupting a little bit to describe to the audience what we’re looking at.
Feel free to interrupt me, Ron. I started my screen share. You should be able to see the selfie or photograph of this man. This man does not exist. This is a face morph. It is a combination of two faces of myself and my cofounder Mark. Using AI technology, AI has taken the average of our two faces and created a new identity. We call him Moris, which is Mark and Joris, which never exists and never will exist.
If you were looking at the screen, what you would see is a headshot. It is a closeup of someone. If you weren’t on this show and saw this, you would think it’s someone and it’s very real.
This is what you would take as input. This would be a selfie we would receive in a digital identity verification process. If we run this image through our software, we are returned with two things. The first thing our software returns is the probability of this being a deep fake. In this case, it’s classified as highly likely to be a deep fake, over 98%. That’s the first part of the service we provide.
The second part we provide is this heat map, which indicates which pixels contribute most to the decision of our AI. This heat map with red, green, and bluish colors highlights the intensity of the pixels that contribute to the decision of AI. In this case, you can see the entire face is highlighted, which is correct because, in this face morph, everything is manipulated or altered with deep fake technology and the labeling of the entire face.
If you’re looking at the screen, what you would see is four images. The top left is the original image we saw in the last grade. The top right is that same image but cropped. It is a little tighter focus on the face. Below those are heat maps, much like you might see on Google Analytics if you’ve ever used that. The overlay shows blue, green, yellow, and red. The red is mainly around the nose area into the face.
The other heat map is similar. It’s named a heat map with the intensity of artifacts. It’s prominent showing these changes. I’m sure 98.1%, which is what it says at the top in regard to a probability of a deep fake, is a great example because you can see the color highlights. You immediately look at it and say, “If it’s 10% or 20%, they would be much more minimal.” With that, go ahead, Joris.
What I also want to point out is we give an estimate on what type of deep fake generation methods this could be. There are dozens, if not more, hundreds of different ways to create deep fakes, like face swapping, face morphing, and lip-syncing, to name a few. We also try to do a classification on that. In this example, we predict it’s around 22%, an entire face synthesis, 21% of face swap, or 21% of face morph. This can help the user substantiate his or her decision.
I assume this all happens quickly for using your tool. You did this. You get this near instantly.
For images, this is also almost real-time. We also apply this to video. It’s a bit heavier on video but can also be done near real-time.
For the public and the media, how are you more vigilant against the spread of wrong information through deep fakes? What can they do to be more vigilant even before turning on your tool or knowing they should be turning on your tool?
We provide a retrospective analysis of images and video. For some use cases, that’s not enough. The damage may already be done even before the retrospective analysis that we provide comes into play. For many years, people have believed that images and video are the last stronghold in information supply. If you see it, it must be real and it cannot be faked. That’s what most people still believe.
We live in a time when this is no longer the case. Anything you see on any screen, tablet, smartphone, or social media may be faked or at least manipulated. There should be an awareness shift among people and the general public that this stronghold of information supplies that what you see must be real. It no longer applies in 2023.
You talk about a problem to be solved. It’s big in different ways. People say AI and think it’s an individual thing. It’s not. It’s many use cases and industries it touches. You’re starting with digital IDs but there are probably 100 that we can aim easily in other directions for you to go and uses, like maybe we have your tool on our computer. It’s automatically checking things as we go. That’s one possibility. Do you imagine years down the road how we will be using it? Will it be an app on our mobile device?
At the moment, we provide a product which is similar to what you described, Ron. We created a browser plugin, deep fake proof, which you install within Chrome or whatever browser you are using. It checks the footage images you are watching if you’re not by accident looking at deep fake material and maybe being misled by deep fake material.
It checks whatever you are watching and notifies you if you think a deep fake was found. In a couple of years, we see Duck Duck Goose as the go-to company for digital media authentication. Our primary focus is on images and video but we want to expand that to voice in the short-term and other modalities of media and data in the future.
One builds on the other. The more you do with digital identity, the more you get your media and videos going. You’re not starting from scratch each time you are building on what you’ve already done, which is amazing. When you think about videos, it’s one of the most dangerous but there are dangers in every direction.
You talk about politics and affecting the masses, citizens, and cultures. With AI, I could produce a video with somebody I don’t like, maybe someone in the government doing something untoward, and get it out on all the socials. It’ll go viral in a moment and be treated as real even though it’s not. We know that can happen and it does happen, quite honestly.
The ability for individuals to know right away, if that’s the case, you’ve got a mission that’s incredibly important and powerful. That’s more of a statement than a question but why don’t you dovetail from that into? What is going on with generative AI that maybe the average reader isn’t tracking? Share for the final on this segment what’s on your roadmap. Instead of me projecting out, you tell me what the most important points are you’re looking forward to.
When it comes to generative AI, it is here to stay. The applications are virtually endless, from design, gaming content, other types of content creations and voice, and all niche applications. Within these niches, generative AI can be a great tool and optimize many things. This is a niche focus so it will become a feature in many products. A great example of this is Adobe Firefly, where within the latest Adobe software, with a bit of prompting, one does the Photoshopping instead of how it used to be done.
The next big step in narrative AI is good multi-model generative models. These are models that combine different types of modalities of data as input, for example, sound and video or text and video, and output multiple different modalities of data. It is a video with sound, for example. This is relatively in its infancy but give it a couple of months, maybe one and a half years, and the multimodal generative AI will kick off.
On our roadmap at Duck Duck Goose, our primary focus at the moment is to grow in digital identity verification and grow within that vertical. We’ll be releasing products for synthetic speech detection because voice biometrics is becoming more popular or mainstream as an identification method, which can also be spoofed by using voice narrative AI models. For us, the next several months are all about growing in digital identity verification and releasing new features and products.
Most of us heard a long time ago that if you ever answer a call and you don’t know who it is, most of us don’t even answer anymore. If you ever answer, they’re going to ask you a question. Hopefully, you will say yes. As soon as you say yes, they utilize that and use it against you. We’ve all learned that a long time ago. Taking all those fears away would be fantastic.
This voice is booming. It’ll become a big thing. I pick up my phone if someone calls me and I don’t recognize the number. You are founding a startup while still looking for a product solution. You’ll pick up the phone. You never know who might call.
Thank you for that. That was segment one. We are going to head to segment two. It’s time for AI Wants To Know. AI is curious and so are we. These are ten quick questions designed to uncover the intriguing mysteries that AI longs to comprehend but can’t quite grasp. Think of it as a snack break in our journey. Keep the answers quick but the safety belt sign is also off. If it feels right, we can occasionally roam about the cabin, exploring more of who you are and what makes you tick. What’s the first thing you ever remember being proud of?
When I was eight, we relocated to India with our family. My parents, both of them had worked there. I hardly spoke any English, let alone Hindi. After one month of being there, my brother and I spoke fluent English and Hindi. I’m quite proud of that, picking that up quickly.
As you should be, that’s amazing.
I lost the Hindi. Once we got back, there was no opportunity to practice. That’s a shame.
Stay in India for a month. It’ll all come right back. I know that from experience with the different languages. What do you need help with that you wish you did not?
One area where I sometimes can do with a bit of help is seeing the greater picture. Synthesizing and comprehending the broader context can be challenging. My two cofounders, Mark and Parya, do a great job at helping me with that. In that respect, it is a balanced team.
If we all had the same strengths, we would need partners. That sounds like a perfect partnership. What do others often look to you for help with?
My colleagues and team members often come to me for critiques or reviews of their work, which I’m happy to give. Friends often come for relationship advice and renovating houses. I’m quite proficient in doing things inside the house and enjoy doing that.
We’re going to head for question number four. What do you treasure most about your human abilities?
As long as I can remember, I have been curious and have the drive to keep learning. That’s something I treasure about my human abilities.
If you look back on your whole life, what is the most consistent thing about you?
I have a passion for things that move. It started with cars. It became buses, trains, planes, and boats. You name it. I’m fascinated to learn about vehicles, how they work, and why they’re so efficient. That’s a consistent thing throughout my life.
If you look back throughout your whole life, what has changed the most?
Me being able to handle losing. I was a terrible loser as a kid and can’t handle that much better. As a startup founder, you have more lows than you have highs. You need to be able to handle those. I wasn’t able to do that several years ago.As a startup founder, you have more lows than you have highs. You need to be able to handle those. Click To Tweet
It goes back to an earlier question. When you were saying you focus on the granular, especially as an entrepreneur, you got to pull back and look at the whole picture sometimes. What do you find strangest about reality?
Reality is never strange. We are the ones that strange. The reason it seems this way is that our perception of reality is influenced by our consciousness or subconsciousness, which has inherent tendencies and characteristics which shape the way we interact with reality. Our experiences and logical intuitions over time contribute to our expectations of how things should be or how we view the things around us. Reality is there. It’s not strange. We are, or our perception of it is.
We’ve all had those moments of life where you feel alive. Usually, it’s to an event or something happens. You get that powerful, alive feeling. What’s an example of that that you’ve had?
That was when I went flying. I have a private pilot license. You can find me up in the skies above the Netherlands during the weekend. That makes me feel alive.
How about traits? Do you have a single trait you would put as your most unique one?
I found this the most difficult question. I have a few peculiar threads but we’ll leave it at that. The readers who know me a bit better might be laughing.
We’ll get your partners on the show and ask them. This one is a sideways question. If you weren’t human, what would you be?
I would love to be a house cat. I would like nothing more than to spend my days grooming, sleeping, eating, and silently judging humankind.
I appreciate all that. It gives us a little more insight into your personality and we’ll take that all day long.
Segment three is AI Leaders and Influencers. It allows you to highlight some of the leading individuals, projects, and organizations that influence you or that people might want to follow. You are in it. It’s not that different from crypto and blockchain several years ago when nobody knew who to listen to and we listened to everybody. What you heard wasn’t always accurate or good. You are so deep in this. Maybe you can give some insights as to different AI leaders or influences.
There are many leaders or influencers to mention but to name a few, Andrew Ng is a prominent figure within the AI community, Fei-Fei Li, the former Director of the Stanford AI Lab and Founder of AI4All, and Timnit Gebru, who is a computer scientist and prominent advocate of ethical AI, which is relevant. I find it interesting. Those are relevant to deep fake detection. Mustafa Suleyman, who is the cofounder of DeepMind. He has done some great work there. There are many names for a single list.
For those of us that aren’t as immersed as yourself, that was a great list. It gives us a great target to go toward and learn even more. Those are the people but when you talk about resources, websites, apps, books, and podcasts, do you have anything to share with us about the resources you find the most valuable and what you spend your time on?
I have a couple of resources I like to share. As a student, I relied heavily upon a book called Pattern Recognition and Machine Learning by Bishop, which was what we called the Bible of Machine Learning. I still look through it every now and then. It offers a mathematical foundation for modern machine learning, which any machine learning or deep learning engineer needs to understand.
That’s for the technical listeners. I follow Yannic Kilcher YouTube channel. He goes into academic research papers or bills. He explains the state of the art in machine learning through academic research papers, discusses these papers, critiques them, and explains how they work or how the code implementation works.
I follow a newsletter called The Sequence, which covers a bit of everything. It keeps me up-to-date on the investment landscape in which AI startup is raising money. Which features are being released within machine learning or deep learning libraries? What code updates are being released? What else is going on in the community? The newsletter helps me with that.
Reading up can be a challenge because there are so many people in the world working on deep fake detection alone. It’s hard to keep up with what’s going on out there. I use Zeta Alpha, which is a platform for searching academic research papers. It’s a Google Scholar on steroids. I find it incredibly useful for staying up-to-date within my domain of deep defect detection.
Amazing bunch of resources mentioned. Thank you for that. I appreciate it. It’s going to save someone tens if not hundreds of hours to go into the wrong places. We’re going to move to segment five, which is AI Tips. Tell us some of the cool ways that you use AI that we might not have explored or ways that everyone may not have realized are available. You might want to give some examples that illustrate techniques used to seed AI or get the best results. Share something you have insights into because of your specialty that you have.
Explainable AI is here to stay. Instead of being a feature on top of classification or regression models, I encourage AI practitioners to design for explainability. It can no longer be seen as an adult. It will also be mandated in Europe. There is a regulation coming up that makes this explainability mandatory. For certain use cases, we cannot do without this explainability for biometrics or things using the court.
As a start, offering the same level of explainability that a human could is a great starting point. That’s considered a task of defect detection. We need to substantiate our classification to our customers. By mimicking how we would do this ourselves and letting the AI learn how to do that, it offers a good starting point for explainable AI. What I’ve learned is that in industry, you don’t necessarily need the latest and the newest to get the job done. You don’t need 70 billion parameters or large language models. Sometimes a simple or random forest classifies works as well.
Most of the time, there is no need to over-engineer things. Engineers always like to over-engineer things. I like to over-engineer things. It’s the fun part of the job but often simple methods are as great. Lastly, there will be more of a focus on bias and transparency of training data in the future, especially with the big tech companies. While the products they release, they’re quite closed about what data they use to train these LLMs. This issue will be increasingly relevant in the future.
There have been threats of some substantial lawsuits. I don’t know whether they’ve been turned into lawsuits yet but in regard to where did a product do its training and if it did it on someone else’s product, it’s an issue. All that is coming to light. You mentioned some laws being written in Europe. That’s an interesting direction. AI is amazing but it can also do some amazingly bad things. We know that. It can be used improperly or for bad things by bad people. There have to be regulation and governance around it.
The reality is for yourself, and I know for me, coming from the industry, the world moves fast, and any regulative body moves slow. By nature, it’s supposed to get it right. We have this situation where they want to be moving slowly and getting right while the industry is moving fast. By the time the regulators catch up, the industry’s already in another universe. I’ll call it an issue. You can call it a problem. I’m going to ask you to speak to that a little bit and see what you think.
Part of the answer to that is Duck Duck Goose and there are other products that have the same mission. You explained that explainable AI is a big thing. You asked for transparency in it. It shows us how you came up with this and what you utilized. Those things being designed are important and great. That is for Duck Duck Goose. I haven’t heard much about it. I don’t know if you want to speak about some of the regulations that are coming out, some that should come out, or any advice to the regulators, whether it’s the EU or anywhere else. That subject, in general, if you could speak to that, that would be great.
Regulation is always lagging behind technological developments that have been for centuries. With the potential of abusing generative AI technology, there is an urgency to do things here. That gives opportunities for companies like my company that offer retrospective analysis. I was very glad to read that seven big tech firms signed into a White House AI commitment, among which open AI and meta base saying that they’ll cooperate with and think along with what the legislation is coming up. I’m glad to read that the big tech is taking responsibility when it comes to open-sourcing, being transparent, or trying to mitigate bias as much as possible.Regulation is always lagging behind technological developments. With the potential of abusing generative AI technology, there is an urgency to do things here. Click To Tweet
That was big news and we’ll watch it as it goes. A big pat on the back for the mission that you guys have. It’s critically important. The fact that you’re so advanced in it, you as a company, is amazing. Where can readers go to learn more about you, the projects you’re working on, and Duck Duck Goose in general?
You can find more on the Duck Duck Goose website, DuckDuckGoose.ai. If this show resonated with you, you would like to follow up on this conversation, or learn more about the deep fake products and services that we provide in Duck Duck Goose, don’t hesitate to reach out. We would love to connect.
Joris, you brought it to yet another level. It’s fantastic. You sharing all that you had to share on this show is going to help many other people. I can’t thank you enough for it but there are probably some people behind you that have helped you. Take a moment. Anybody you’d like to give a shout-out to or mention, this is your moment.
Thanks for the opportunity, Ron. I’d like to give a shout-out to all the advisors that we have at Duck Duck Goose: Joe, Ian, Hans, Yun, and especially Jin, who made the connection with you. Thank you so much. We would not be where we are at Duck Duck Goose without you.
It’s time for another safe landing at the outer edges of the AI universe. On behalf of our guests and the entire crew, I like to thank you for choosing to voyage with us. We wish you a safe and enjoyable continuation of your journey. When you come back aboard, make sure to bring a friend. The starship is always ready for more adventurers.
Head over to Spotify or iTunes. Rate us, and share your thoughts. Your support and feedback mean the world to us. Don’t forget to visit EdgeOfAI.xyz to learn more. Connect with us on all the major social platforms by searching for EdgeOf_ai. Join the exciting conversations happening online. Before we sign off, mark your calendars for our next voyage, where we’ll continue to unravel the mysteries and advancements of AI. Until then. Bye-bye.