🧬Resurrection Biology in the Digital Age: AI's Transformative Role in Reviving Extinct Species - Reviving extinct species with AI
34:03 Share

Welcome to a new special deep dive from the podcast AI Unraveled.

It's created by Etienne Newman, who's a senior engineer and a passionate soccer dad from Canada. Great to be here. And hey, if you're finding these deep dives into the world of AI fascinating, please take a moment to like and subscribe over on Apple Podcasts. It really helps. Definitely. Also, quick heads up, check the show notes. We've got a referral link and a discount code in there if you want to sign up for Google Workspace. Oh, yeah. What's included? Well, that gets you access to some seriously popular

powerful AI tools, you know, like Gemini PRO, Notebook Limb, and of course, Teams integration too. Handy stuff. And one more thing, if you're prepping for those tough in-demand certifications, definitely explore Etienne's AI-powered JamGatek app. It's designed to help you really nail like over 50 different kinds of PBQs and simulation exams. Sounds like a lifesaver for certification prep. It really is. Okay, so today, we're diving into something that

Honestly, used to feel like pure science fiction. Totally. We're talking about de-extinction, bringing back species we thought were lost forever. And AI is, well, it's playing a massive role in making this less sci-fi and more potential science fact. Exactly. For years, you know, bringing back a woolly mammoth or a dodo, that was just movie stuff. Right. Pure Hollywood. Right.

But now, with all the breakthroughs in biotechnology and critically, the power of artificial intelligence, its ability to analyze, to predict this is becoming a real tangible scientific pursuit. Yeah. The sources we dug into for this deep dive, they really lay out the science that's making it possible. They spotlight some of the big projects, the flagship efforts. Like colossal biosciences. Exactly. Colossal is a big one. And the sources also get into the really tricky ethical questions.

And crucially, the ecological considerations. I mean, what happens when you actually do it? It's huge. So our mission today really is to unpack AI's role in all this. How is it actually making this happen? And what does it all mean for biodiversity, for conservation, for, well, for the future? Yeah, what are the implications? It's a big topic. Okay, let's dive in.

What was the turning point? What shifted de-extinction from just this cool idea to something labs are actually working on with serious funding? Well, it's really a convergence, isn't it? Several things came together. Advances in genetics, our ability to read and manipulate DNA got way better. Sure. Sequencing got faster, cheaper. Precisely. Yeah. But the real game changer, the catalyst, has been A.I.,

It's not just helping out. It's become fundamental. Fundamental how? Think about the sheer amount of data in a genome.

AI is essential for making sense of that, especially ancient damaged DNA. It's guiding the gene editing with incredible precision. Like with CRISPR. Exactly, like with CRISPR. And it's even helping model what might happen if we reintroduce a species. AI is needed for the analysis, the engineering, the prediction. It's woven through the whole process. It really taps into that, I don't know, that human fascination with lost worlds, doesn't it? Mammoth, dodos, pterosaurs.

They capture the imagination. You absolutely do. But you said AI is rewriting the rules of extinction. That sounds dramatic. What does it actually mean in practice? How is it different from older ideas? Well, think about the older methods. You had things like backbreeding. Like trying to breed cattle to look like aurochs. Exactly that. The heck cattle are a famous example. But you're limited to the genes that still exist in the living population, right? You can't bring back something that's truly gone. Right. You're just shuffling things.

The existing deck. Precisely. And then there was cloning, like Dolly the sheep. That works for recently extinct animals if you have perfectly preserved cells with intact nuclei. Which is almost never the case for ancient species. Almost never. Remember the Pyrenean ibex? They cloned one briefly, but it died almost immediately. The big hurdle was always the ancient DNA, fragmented, damaged, contaminated cells.

like trying to read a book that's been through a shredder and left in the rain. Okay, so that's where AI steps in. That's exactly where AI makes the difference. Its ability to handle massive, messy data sets is key.

AI algorithms can sift through that genetic noise, identify the real ancient fragments, piece them together. Like a super-powered jigsaw puzzle solver. Kind of, yeah. And it can even predict missing bits based on related species and guide the CRISPR editing needed to put those ancient traits back into a living relative's genome.

This AI driven sophistication is attracting serious money. Companies like Colossal. And they're getting results. You mentioned dire wolves, woolly mice. Early results, yes. And they're definitely debated. But they're using these AI powered techniques to create animals with some traits of extinct species like mice engineered with mammoth genes to study cold adaptation. It's happening. Okay, so the AI helps piece together the ancient blueprint. You used the term functional de-extinction earlier.

It sounds like the goal isn't necessarily a 100% perfect replica. That's a really crucial point. Creating an exact genetic copy, like a twin separated by millennia, is probably impossible because that ancient DNA is just too degraded. Too many missing pieces. Exactly. So the focus now is mostly on functional de-extinction.

The idea is to create an organism that, OK, maybe its genome isn't identical, but it looks, acts, and most importantly, functions in the ecosystem like the extinct species did. So it can do the same job. Like if the mammoth helped maintain the grasslands, the goal is an elephant hybrid that can do that, too. Precisely. It's about creating an ecological proxy, a stand in.

They do this by carefully editing the genome of the closest living relative using Kritker, guided by AI comparing the ancient and modern genomes. The result is a hybrid.

engineered to carry the catch rates and fill that ecological niche. Okay, that makes more sense. It's pragmatic. So let's recap the oldie methods quickly. Back breeding, you said, was limited because you only work with existing genes. Right. Can't recreate lost genes. Only amplify traits that are still around, maybe hidden. And cloning SCNT somatic cell nuclear transful needed intact cells, which you just don't get from ancient remains. Correct. You need a preserved nucleus with undamaged DNA.

Works for Dolly, maybe for something that died very recently and was frozen perfectly, but not for a mammoth or a thylacine. Got it.

So that brings us to genome editing. CRISPR-Cas9. This is the cutting edge, the main tool now. Absolutely. This is what companies like Colossal are banking on. First, you sequence the ancient DNA, whatever fragments you can get. AI helps piece that together. Right, the digital archaeology bit. Then you compare that ancient sequence to the genome of the closest living relative Asian elephant for the mammoth, say.

AI helps pinpoint the key genetic differences, the ones that likely coded for the mammoth's unique traits, like shaggy fur or cold-resistant blood. Okay, so you find the mammoth bits in the code. Exactly. Then CRISPR-Cas9 comes in. Think of it like molecular scissors guided to very specific spots in the living relative's DNA. So you cut the elephant DNA. You cut the elephant DNA at precise locations. Then you can either let the cell repair itself, maybe deleting a small bit, or you can provide a new DNA template.

the mammoth bit, for the cell to stitch in during the repair process. Wow. So you're literally editing the elephant genome to include mammoth genes. That's the essence of it. And AI is crucial throughout this. Managing the huge data sets, identifying the best gene targets, predicting if an edit will actually work, predicting potential side effects. Especially if multiple genes are involved for one trait, like cold tolerance. Exactly. Many traits are polygenic, controlled by lots of genes interacting.

AI helps model those complex interactions and decide which edits are most likely to give you the desired outcome. Without AI, the sheer complexity would be overwhelming. Okay, so CRISPR makes the edits, AI provides the map and predicts the outcome. But the result, as you said, is a hybrid, an engineered proxy.

Colossal talks about engineering resilience, though. What's that about? It sounds like more than just making it look like a mammoth. It is. Their vision is pretty pragmatic, actually. They're not just aiming for a physical lookalike. They want to create animals that can actually survive and maybe even thrive in today's world, which is different from the world the original species lived in. So future-proofing the de-extinct animal? Sort of, yeah. Three main parts to it, as they describe it. First,

De-extinct the core genes, the ones that really define the extinct species key traits. The mammoth fur, the dodo beak, whatever. Right. Second, engineer enhancements. Maybe add genes for resistance to modern diseases or tolerance to slightly different climates than their ancestors faced. Adaptations for the 21st century. Exactly. And third, engineer resilience may be traits against poaching or better ability to cope if resources are scarce.

It acknowledges that just dropping an ancient animal into today's world might not work. It needs to be equipped for current challenges. That makes a lot of ecological sense. The mammoth steppe is gone. The Arctic is changing. So they're creating a mammoth-like elephant designed for today. That's the idea. A functional proxy with modern resilience built in. You mentioned their dire wolf project involved, what, 20 edits and 14 genes? That sounds like a lot, but also maybe not that many compared to the whole genome.

It is complex, definitely. But you're right. It highlights the challenge. Those 20 edits are targeting specific traits identified through comparing ancient dire wolf DNA with gray wolf DNA. But traits like cold adaptation and mammoths that must involve way more genes interacting, right? Fur, fat, blood chemistry. How do they even begin to untangle that? That's the crux of the problem for complex traits. They're polygenic, as we said, meaning many genes contribute.

And then there's epistasis, where genes influence each other's effects. It's not just adding up individual gene functions, it's a complex network. Like an orchestra, not just solo instruments. Good analogy. Recreating that requires multiplex gene editing, making many precise changes across the genome simultaneously.

And that is incredibly difficult. More edits mean more chances for things to go wrong for unintended interactions. - So predicting the outcome gets exponentially harder. - Exactly. And that's where again, AI and sophisticated computational modeling become absolutely indispensable. Trying to predict the phenotypic outcome of dozens, maybe hundreds of edits without AI, almost impossible. - So AI is really the computational architect here, making sense of the blueprint and guiding the construction. Let's dig into that more.

Starting with reconstructing those ancient blueprints, paleogenomics. You said the DNA is a mess. How does AI clean it up? Right. Ancient DNA is usually terrible. Tiny fragments, chemical damage like de-emination, loads of contamination from bacteria, fungi, even researchers. A needle in a haystack, basically. A shredded, rusty needle in a haystack full of other shredded needles. AI, specifically machine learning, is brilliant at pattern recognition.

It learns to spot the telltale signs of actual ancient DNA characteristic damage patterns, typical fragment lengths, and distinguish it from contamination. OK, so it filters out the junk. It filters the junk. Then AI powered algorithms assemble the remaining fragments, like putting together that shredded document, trying to figure out which tiny piece goes next, dealing with repetitive sequences, bridging gaps. Stuff that would take humans forever. Or be impossible. AI can also help correct errors caused by postmortem damage.

There are specialized tools, like an assembler called Carpidime, designed specifically for this damage-aware assembly, getting longer, more accurate sequences. And Colossal is using this for the mammoth and thylacine. Recordedly, yes. Combining AI-driven assembly with newer long-read sequencing tech and even RNA analysis, which tells you which genes were actually switched on to get these remarkably complete genomes, like that near-complete thylacine genome they announced. Okay.

Danielle Pletka: Okay, so AI pieces together the shattered history. You also mentioned generative AI potentially filling in gaps. How does that work? Is AI writing mammoth DNA? Marc Thiessen: This is where it gets really cutting edge and maybe a bit spooky. Generative models like one called EVO2 that some researchers developed are trained on, well, basically T-O-N-S of genetic data from living things, trillions of base pairs. Danielle Pletka: So it learns the rules of genomes. Marc Thiessen: Exactly. It learns the patterns, the structures, how genes are organized, how they relate to function,

Because it understands these deep rules, it can potentially predict what should go in the missing sections of an ancient genome. Based on the surrounding sequence and what it knows about related species? Yes, and its general knowledge of how genomes work. It could infer missing segments, even predict their function, and help check if the overall reconstruction makes sense.

Other AI types, like Jans, might also theoretically learn the statistical patterns of genomes and generate plausible fill-ins. Wow. The key thing is, getting that ancient blueprint right is step one. If the AI reconstruction is flawed, everything downstream identifying gene targets, the editing itself could be based on bad info. Leading to unexpected results. Or just failure. So AI, especially generative AI, is shifting things from just reading the damaged code to potentially completing it.

based on biological principles. Which, as you said, raises huge questions. If AI is writing parts of the code, is it still a mammoth? That's the philosophical minefield, isn't it? What does authenticity even mean when you have AI-generated sequences in there? It definitely blurs the lines. Okay, mine's slightly blown. So...

Let's say AI helps assemble or even complete the blueprint. What about figuring out what those ancient genes actually did? Functional genomics. Right. You need to know which genes gave the mammoth its fur or the dire wolf its build.

Again, AI's predictive power is key. Models like EVO2 can predict the functional impact of specific genetic variations found in the ancient DNA. They can even infer protein shape and function just from the sequence. So it looks at a mammoth gene and says, this probably coded for a protein involved in fat metabolism. Something like that, yeah. And then there's comparative genomics powered by AI. You line up the ancient genome with its living relative mammoth versus elephant, and the AI highlights the difference.

Those differences are your prime suspects for the unique traits. That's how Colossal picked targets for the dire wolf edits. Presumably, yes. Comparing ancient dire wolf sequences to modern gray wolves to find the key variations.

But then you want to know why the AI flagged a specific gene, right? You don't want it to be a black box. Definitely not. You need to trust its reasoning. Which is where explainable AI or XAI comes in. Techniques like LIME or ACHE try to show which parts of the input data, which bits of the DNA sequence led the AI to make its prediction about function. It provides insights,

help scientists validate the AI's hypotheses. So the AI shows its work. That's crucial. You also mentioned molecular de-extinction, bringing back molecules, not whole animals. That sounds less...

Ethically fraught, maybe? Potentially, yes. And it's yielding some really interesting results already. The idea is to use AI to scan the predictive proteins of extinct animals looking for molecules with useful properties. Like medicines. Exactly. Researchers used an AI model called APEX to find potential antimicrobial peptides molecules that fight bacteria in the predictive proteins of mammoths, ancient elephants, even giant sloths.

Seriously. Finding antibiotics in mammoths. Well, molecules that act like antibiotics. They found things they've named Mammothusin, Alephicin, Myelodonin 2, and early tests show they can inhibit modern, sometimes drug-resistant bacteria. That is amazing. Talk about a tangible benefit. It really is. It offers a potential justification for all this complex work, something with maybe more immediate applications and fewer ethical hurdles than bringing back whole animals.

AI turns paleogenomics from just describing ancient code to potentially resurrecting functional molecules from it. So AI gets the blueprint, predicts function, even finds useful molecules. What about the actual engineering, making the CRISPR edits? How does AI help there? It enhances the whole process, precision, efficiency, predictability.

A big worry with CRISPR is off-target edits cutting the DNA in the wrong place. Which could cause problems. Big problems. AI tools like Deep CRISPR analyze the target DNA sequence and the guide RNA that directs the CRISPR enzyme, and they predict likely off-target sites with pretty good accuracy. This lets scientists design better, more specific guide RNAs. Minimizing collateral damage. Right. And other tools, like azimuth, predict the on-target efficiency.

how likely is this guide RNA to actually make the cut where you want it? So you pick the guides most likely to work well and safely. Exactly. When you're doing multiplex editing, making lots of changes like in the mammoth or dire wolf projects,

This AI optimization is absolutely vital. You need efficiency and you really need to avoid off-target mutations piling up. And AI also helps predict the combined effect of all those edits, dealing with that epistasis problem. Increasingly, yes. Colossal mentions using AI modeling of 3D protein folding to see how an edit might change a protein's shape and function.

And researchers are developing AI models to predict the overall phenotypic outcome of multiple edits, trying to understand those complex interactions. Like in the woolly mice experiments. Yeah, those experiments putting mammoth genes into mice to test their effects rely on AI for selecting targets and designing guides. But it's still hard. Those experiments also show that biological validation is crucial. AI makes predictions, but you still have to test it in a living system. It's not magic. The biology still has the final say. Definitely not magic.

But AI provides a massive leap in targeting and predicting, making the whole synthetic biology aspect much more rational and efficient. It's driving innovation that has spinoffs for agriculture medicine. Right. The spillover effects. OK, so AI helps the DNA, the function, the editing. But you still need to go from an edited cell to an actual animal. Gestation, development, AI's role there. This is still a major bottleneck, but AI is starting to help here, too.

Early development after cloning or using edited cells is notoriously inefficient. Lots of embryos fail. Wasted effort and ethical concerns. Both. So AI-assisted embryo viability assessment is a growing field. Deep learning models analyze images or other data from developing embryos, looking for subtle signs of health and potential that humans might miss. Like picking the winners early on. Essentially, yes.

Selecting the embryos most likely to successfully implant and develop. This is already used in human IVF. And the potential for de-extinction projects, where you might have limited embryos or surrogates like Asian elephants for mammoths, is huge. Improving the odds. Exactly. Then, looking bigger picture, ecological integration. How will this engineered animal fit into today's world?

AI is crucial for building complex ecosystem simulations. Modeling the ripple effects. Yeah, trying to predict how the reintroduced species might interact with existing plants, animals, the climate. Colossal is apparently doing this for their dire wolf and thylacine projects. There are also tools like genomic offset models. Genomic offset? What's that? It uses genomic data from the animal and environmental data from potential release sites to predict how well that population might adapt or maladapt to the new environment.

Assessing the risk of failure or maybe even the risk of it becoming invasive? Both. It helps compare potential reintroduction sites and flag risks. And some researchers are exploring deep reinforcement learning, where an AI agent learns optimal strategies in a dynamic environment to model ecosystem responses and maybe even guide management decisions after reintroduction. Wow. So AI is involved right up to...

planning the release and management. It's really end to end. It's becoming that way. It's about improving the biological processes like embryo viability, but also about bringing a much needed focus on ecological responsibility and risk assessment before you potentially release anything. Okay, let's talk about the flagships actually doing this. Colossal Biosciences comes up constantly. Tell us more about their main projects. Mammoth, Dire Wolf, Phyllocene, Dodo. Right, Colossal is definitely the most high profile player.

Co-founded by Ben Lamb and George Church, significant funding, AIMS focused on that functional de-extinction using a systems approach heavily reliant on AI. So the mammoth, cold-resistant elephant hybrid. That's the goal. Their process involves getting ancient DNA, sequencing it with AI help, comparing it to the Asian elephant, AI again, identifying key mammoth genes for cold adaptation, using CRISPR to edit those into elephant cells. Okay.

AI optimizing the edits, predicting protein folding, then SCNT to create embryos and eventual gestation in a surrogate elephant or maybe an artificial womb someday. And the woolly mice are like a test bed for those mammoth genes. Exactly. A faster way to see if editing in a specific mammoth gene actually makes a difference to traits like hair growth or fat storage.

But yeah, huge challenges remain. Data management, SCNT efficiency, the long elephant gestation. And the ecological goal is restoring the mammoth step. That's the grand vision, yeah. Using these proxies as ecosystem engineers to combat Arctic warming by converting tundra back to grassland. A very ambitious goal. Okay, then the dire wolf, they claim to return. What was the science there and the controversy? The claim was based on sequencing ancient dire wolf DNA from fossils.

An AI's role? But the controversy is whether 20 edits makes it a dire wolf. Exactly.

Many scientists argue it's a genetically modified gray wolf, maybe a hybrid with some dire wolf traits, but not a true de-extinction. The genetic distance and the limited edits make it debatable. Right. And the phyllocene, the Tasmanian tiger. Another major target. The challenge there is the greater genetic distance to living relatives, the dizerids, like the numbat or Tasmanian devil.

So AI is even more critical for comparing genomes, figuring out the key thylacine genes, and modeling reintroduction. But they got a very complete genome. Yeah. Reportedly 99.9% complete using long read sequencing and RNA analysis assembled with AI tools.

The ecological aim is to restore an apex predator to Tasmania, potentially helping control invasive species. And the dodo seems symbolic. Highly symbolic, yeah. Represents human-caused extinction. Colossal is working on it, leveraging advances in avian gene editing and using AI for the genome work. The Nicobar pigeon is the closest relative. They also think it played a role in seed dispersal. And Colossal spins off the tech they develop. That's a key part of their model. Companies like FormBio,

bioinformatics software, came out of their work. De-extinction acts as an engine of innovation, driving development of tools they can then commercialize. Interesting model. Now, what about revive and restore? How do they differ? They're quite different.

Nonprofit, conservation focused, more about genetic rescue than headline grabbing de-extinction of long lost species. So helping currently endangered species. Often, yes, or species extinct more recently where the intervention might directly support existing conservation.

They act more as conveners, funders, coordinators, fostering collaboration. While they use genomics and computation, their specific AI use isn't detailed like colossals. What are their main projects? The passenger pigeon. The great comeback is a big one. Sequencing its genome, planning to edit the band tail pigeon. Big challenges there. Avian editing, identifying genes for complex flocking behavior, ecological integration of potentially billions of birds. In a very different scale. Huge scale.

They've also looked at the heath hen using prairie chickens. A lot of their work involves using cloning for genetic rescue boosting diversity in bottlenecked populations like the black-footed ferret. Where cloning adds back lost genes from stored samples. Exactly. Same idea for the Przelski's horse, enhancing the genetic health of the managed population using genes from long-dead individuals. They also have plans for the great auk, maybe using razorbills.

and broader programs too on coral resilience, bird conservation biotechnology, supporting wildlife genomics. So a different philosophy, colossal, high-tech, commercial focus, bold claims, revive and restore,

conservation, collaboration, maybe more cautious. That's a fair summary of the different vibes, yeah. And it's not just them, right? There are universities, zoos. Oh, absolutely. Academic labs like George Church's at Harvard, Beth Shapiro's at UC Santa Cruz, Tom Gilbert's in Copenhagen, Stanford, Berkeley. They do tons of foundational work. Conservation groups like San Diego Zoo Wildlife Alliance with their amazing Frozen Zoo are crewful partners. Providing expertise, samples. That's exactly. Zoos, international bodies like the IUCN,

then specialized biotech companies contributing tools, and consortia like Pleistocene Park in Siberia, the Earth Biogenome Project. It's a really diverse interdisciplinary network. Okay.

It all sounds incredibly advanced, but it can't be easy. Let's talk about the hurdles, the labyrinth of challenges, as our outline calls it. What are the big technical and scientific obstacles still standing in the way? Oh, they're significant. We've touched on some. Ancient DNA quality is still fundamental. Even with AI, if the starting material is too degraded, you just can't get a complete, accurate blueprint.

Some experts are skeptical true recreation is ever possible for really ancient stuff. The blueprint problem. Then gene editing itself. Even with AI optimization, achieving high efficiency and precision for multiplex editing, changing many genes at once is tough. Off-target effects are still a risk. The woolly mice work reportedly had low success rates for getting all intended edits.

Imagine scaling that to an elephant and predicting the final phenotype, the actual traits from those complex edits. Still very hard due to epistasis and how genes interact with the environment. Biology is complex. And then getting from cell to baby animal. Embryology, surrogacy. Huge hurdles there. SCNT is inefficient. Interspecies surrogacy, putting a mammoth embryo in an elephant is fraught with potential issues.

Immune rejection, placental problems, different gestation lengths. Yeah. It's risky for the surrogate too. Especially a valuable endangered elephant. Absolutely. And avian surrogacy for birds is also technically very challenging. Artificial wounds, ectogenesis get talked about. Growing a mammoth in a lab. That's the sci-fi vision. It could potentially bypass surrogacy issues, offer a controlled environment. They've had some success with lambs.

But scaling that to an elephant, we're nowhere near that yet. And the ethical implications are profound. Yeah, that opens up a whole other ethical dimension. What about the ecology? Assuming you overcome the technical hurdles and make the animal, what happens when you put it back into the wild? Another massive set of challenges. Is the original habitat even suitable anymore? Climate change, human development.

Ecosystems are different now. How will this engineered proxy function in a novel environment? Could it become invasive? That's a major risk. Could it outcompete native species? Disrupt food webs? AI modeling helps assess this, but ecosystems are complex and unpredictable. Then there's disease. The de-extinct animal catching modern diseases or carrying ancient ones. Both are concerns. It might have no immunity to current pathogens.

Or it might reintroduce ancient diseases that modern wildlife has no defense against. We just don't know. Then impacting existing species, trophic cascades. Right. You introduce a new predator or a large herbivore, it will have ripple effects. Maybe beneficial restoring balance, but maybe negative. And founder populations will likely be small, raising concerns about genetic diversity and long-term viability. Okay.

OK, so technical and ecological challenges galore. Then there are the ethical and societal dilemmas. These seem just as complex. If not more so. Animal welfare is a huge one. The processes, SCNT, gene editing, surrogacy can involve high failure rates, potential health problems in the resulting animals. Can these proxies truly thrive in modern environments or just survive in managed settings?

What's their subjective experience? We can't really know. Exactly. Then resource allocation. De-extinction is incredibly expensive. Should we spend billions trying to bring back mammoths when countless species are going extinct right now due to habitat loss or poaching? The moral hazard argument, does it distract from preventing current extinctions? That's a major concern for many conservationists.

And the playing God or hubris argument. Are we overestimating our ability to control nature? Is it right to interfere in this way? What does natural even mean for these engineered proxies? Big questions. And public perception. People seem fascinated, but... Fascination is high, yes. But ensuring accurate communication is vital. Avoiding hype. AI could even be used to analyze public sentiment, but genuine public dialogue and engagement are needed.

and governance. How do you regulate this? The technology is moving so fast, regulations struggle to keep up. Issues around intellectual property, safety protocols, international coordination. And the dual use potential of AI and synthetic biology is a serious concern. These powerful tools could potentially be misused. AI adds another layer of complexity to the ethics and governance too, right? Definitely. The

The dual-use risk applies to the AI models themselves. Democratization of AI could make powerful bio-design tools more accessible. Reliance on complex black-box AI models raises issues of transparency and accountability.

Algorithmic bias could creep in. It complicates an already complex ethical landscape. Wow. Okay. It's a minefield. But amidst all these challenges, you mentioned spillover innovations. De-extinction research is pushing boundaries in other areas. Absolutely. The sheer difficulty is forcing innovation. We're seeing big advances in basic genomics and paleogenomics, better sequencing, handling degraded DNA, better AI bioinformatics tools, creating high-quality reference genomes. Stuff useful for all biology.

Totally. Huge progress in synthetic biology and gene editing, multiplex editing techniques, AI for optimizing CRISPR, modeling outcomes.

These tools are valuable for agriculture, creating disease models, basic research, and reproductive tech. Improvements in assisted reproductive technologies, the AI for assessing embryos we talked about, research into artificial wombs, could eventually have applications in human neonatology or livestock breeding. And direct benefits for living endangered species. Yes, genetic rescue, using cloning or gene editing, informed by this work, to boost diversity in bottleneck populations.

Engineering resilience, like Colossal's work on elephant viruses or toxin resistance, could help protect existing species from threats.

AI for monitoring populations too. And the molecular de-extinction, finding new drugs. That's a big one. AI mining ancient genomes for things like those antimicrobial peptides, mammothusin, etc. Huge potential for discovering novel therapeutics. Plus the commercial spinoffs like FormBio. Right, stimulating economic activity, creating companies and jobs around these new platform technologies. De-extinction is acting as this extreme R&D driver. So the journey itself is valuable, regardless of whether we see mammoths roaming soon.

Looking to the future then, with AI getting ever more powerful, what's next? Where is this heading? We'll likely see even more sophisticated AI. Advanced generative models might allow for designing whole gene networks or regulatory regions de novo from scratch to engineer-specific adaptations, or even design synthetic molecules to control gene activity.

genome scale generation might become feasible. AI is a true biological designer. Potentially. Enhanced explainable AI will give us deeper insights into how these complex models work. AI and systems biology will help reconstruct entire networks, metabolic pathways, gene interactions of extinct species by integrating different data types. A more holistic understanding. Right. AI will likely get better at optimizing reproductive tech, maybe guiding artificial womb development.

And ecological modeling will become more sophisticated, simulating adaptation, evolution, hybridization potential, even novel behaviors in more detail. So the long-term vision of restoring whole ecosystems like the mammoth step relies heavily on AI. Absolutely indispensable for planning, modeling, and managing something that complex. But the core challenges remain.

and AI might amplify some ethical dilemmas. The limits of DNA preservation won't disappear. The risk of hype undermining current conservation is real. So it's a co-evolution. AI pushes de-extinction. De-extinction pushes AI. That's a great way to put it. They accelerate each other.

which forces us to confront those profound questions about authenticity, what defines a species, when the designer is increasingly artificial intelligence itself. It's a truly fascinating and incredibly complex field, isn't it? Full of potential, but also riddled with challenges we're only just starting to grasp. Indeed.

AI has fundamentally changed the game for de-extinction. It's moved it from fantasy towards, well, a technologically sophisticated, if still incredibly difficult, pursuit. Its fingerprints are everywhere. Genomics, gene editing, embryology, ecology. Everywhere. But the technical, ecological, ethical hurdles are immense. It's a long road. The extreme R&D aspect is generating valuable spinoffs, no doubt. But as AI gets more powerful in this space, the need for technology

Careful thought, ethical debate and responsible governance just grew stronger. Absolutely critical. We need to keep pace with the technology. So as we wrap up this deep dive, maybe the thought to leave everyone with is this. We as humans equipped with AI are becoming editors of life's code. What does that capability truly mean for our responsibility to the life that's here now and to the echoes of life lost to time? Something to ponder. A profound question indeed. Thanks for joining us for this special deep dive.