Hybrid Vigor: The Intersection of Genetics and Artificial Intelligence
Why I’m Studying Some Of Our Most Dangerous Superpowers
The worlds of genetics and machine learning are both in the midst of their own Cambrian Explosions. If you have been living under a rock, you might not have noticed. From gene drives to “The Singularity”, these two fields have made waves in science and technology headlines for good reasons; these fields give humanity a small taste of what it’s like to play god. Both of these fields have seen remarkable advances in the 2010’s and the pace of research is continuing to pick up speed as we approach the 2020’s.
With genetic engineering, we can reprogram the basic building blocks of life. As we become more and more adept at manipulating genes and genomes, we may eventually have the power to create entirely new forms of life that have never existed before. On the other side, AI researchers are working towards an “artificial general intelligence”, a computer than can think and reason about the world in general, on its own. While current machine learning systems are narrow AI’s, the speed of advances has outpaced even the most optimistic predictions. Scientists and researchers have even discovered that a popular kind of machine learning algorithm — neural networks — have an unanticipated degree of generality.
I personally harbor both excitement and trepidation about humanity’s ever growing technological prowess. Machine learning has already shown great promise in automated cancer screening, but the same technology is deployed to massively automate surveillance in ways that ought to frighten us. Genetic engineering displays a similar dichotomy. The promise of eliminating genetic diseases like Huntington’s disease is counterbalanced by fears of eugenics. Clearly, futurists and sci-fi writers have plenty of fodder for the next wave of entertainment, whether they will be utopian or dystopian is left as an exercise for other writers.
One thing that is clear — the people of the world need to have a richer dialog about the state of these fields; and that dialog must be fueled by accurate and timely information regarding these challenging and complex subjects. Unfortunately, our media diets (in part thanks to machine learning) are clogged up with the banal and the outrageous. Science writing doesn’t get top billing, there is precious little of it that is accessible to the average citizen; and unfortunately some of the most accessible writing is full of errors and downright lies. But, with so many quandaries that rely on science entering the public domain, we can no longer afford to have the general public (and especially the voting public) remain uninformed about these topics.
So, I have embarked on a quest to do my part to change this state of affairs. I am going back to grad school to study the intersection of genetics and machine learning, and along the way I am going to be writing, recording, and creating other materials to spread the knowledge that I hope to gain. Starting with this article: a description of what I think is so exciting — and so important — about the intersection of these two disciplines.
Changing The World; Can We? Should We? And if so, How?
We have entered a new era of advancement in genetic engineering. This new frontier began with the discovery of a breakthrough technique for using RNA-programmable CRISPR-Cas9 to enact site specific genetic editing. In English — a reliable tactic for making precise changes to an individual’s DNA. In the words of Jennifer Doudna & Emmanuelle Charpentier, two of the geneticists who pioneered the use of this gene editing technique:
[CRISPR-Cas9] will enable large-scale screening for drug targets and other phenotypes and will facilitate the generation of engineered animal models that will benefit pharmacological studies and the understanding of human diseases. CRISPR-Cas9 applications in plants and fungi also promise to change the pace and course of agricultural research. — The New Frontier of Genome Engineering, http://science.sciencemag.org/content/346/6213/1258096
While the theoretical applications are about as big as you can imagine, the technique has already been applied in plenty of real-life situations; and as you might imagine there is a lot of controversy surrounding the use of genetic editing in general. The controversy is as broad as the application, and genetic editing has been the subject of legal, political, ethical, and existential conundrums.
When Monsanto first released their bio-engineered crops there were legal issues to sort out. For example, in Bowman v. Monsanto the supreme court found that farmers could not store and grow patented seeds, meaning competitors could not create a secondary market for the patented seeds. Later, the so called “terminator seeds” were developed. Plants grown from terminator seeds produce seeds that are not viable, making it impossible for anyone to grow Monsanto crops for seeds, then harvest and resell the patented seeds. For Monsanto this represented a kind of insurance policy against intellectual property theft.
It’s worth noting that seeds created through the process of traditional plant breeding can also be patented; genetic engineering accelerates the process of producing a patentable seed with a given trait but is not a requirement for earning a patent. Indeed, humans have been deliberately “genetically modifying” plants since Mendel’s peas, and breeding species without specific knowledge of the genetic component for much longer than that; for example dog breeding began almost 15,000 years ago.
While there is an ongoing worldwide moratorium on the use of terminator seeds, the terminator gene concept has found another application combating Zika virus in the Florida Keys. Biotech firm Oxitec has engineered a mosquito with a terminator gene — a custom built genetic mutation which makes the mosquito’s children frail and sickly. Florida voters, motivated by a desire to eliminate Zika virus, have already voted to approve the release of this mosquito.
The idea is, first you release a bunch of male mosquitoes (because males don’t bite) that have been genetically engineered with this mutation into a Zika infected area. Then those mosquitoes get busy breeding with all the natural female mosquitoes in the area. The progeny of the genetically engineered mosquitoes die before they reach adulthood so the population dies off, and there are dramatically fewer mosquitoes left to spread the virus. This tactic is especially exciting because mosquitoes are one of the worlds biggest killers, they can carry not just Zika but malaria, dengue, yellow fever, and more.
Despite voter approval, no self terminating mosquitoes have been released Florida yet. After Florida voters made their decision, sorting out how to regulate the mosquitoes became the FDA’s problem. However, the mosquitoes have been released in the Cayman Islands, and in Brazil with impressive results, ranging from a 61% to 81% reduction in mosquito population. Oxitec is now looking at engineering similar Moths for application in industrial agriculture.
Another proposal, also involving mosquitoes, uses something called a gene drive. A gene drive is a mechanism that can force offspring to inherit a particular gene at rates as high as 99%, much higher than the standard 50/50 chance. Using this mechanism, scientists could force engineered changes to propagate much more aggressively through the population.
In the Oxitec mosquitoes, the genetic engineering causes the bugs with the mutant genes to produce sickly offspring that die before they can reproduce, so the mutant bugs are removed from the population in the next generation. The mosquitoes can’t breed so they can’t pass on their inability to breed. A different kind of mosquito has been proposed, to mixed reception, by the non-profit Target Malaria. Using a gene drive, the proposed mosquito would drive a gene into the population that triggers the terminator gene about eleven generations down the road — for mosquitoes 11 generations is about a year.
Because 11 generations gives the gene a lot of time to spread through the mosquito population, this mutation has the capacity to essentially eradicate mosquitoes entirely, and while no one loves those little bastards, they still play an important role in the ecosystem. If our ability to tackle global warming is any indicator, we might not be able to elegantly suffer the consequences of such a dramatic change to the biosphere. Scientists generally agree that gene drives should not be used in any wild settings, though some might add, “at least not until we have a much better understanding of the implications.”
If you want the plot line for a thriller/sci-fi based on all this, just imagine what would happen if we used a gene drive to spread a terminator gene in the human population. (Oh wait, that’s already Children of Men).
Then, there are ethical issues. The prenatal detection of Down syndrome unearthed some serious ethical traps; especially centering on the potential termination of a pregnancy where Down syndrome is detected. With genetic editing, we have the same problems and more. For example, in the future we may be capable of eliminating autism from the human genome. There are so many scientific barriers that would have to be broken before we could do this, but assuming we could get there, should we let ourselves do that research? And if we do proceed in that direction, what message would we be sending our friends and loved ones with autism? Clearly, the future of gene editing must continue to be the subject of intense ethical inquiry.
On the other hand, there are some slam dunks too. The most promising candidates for genetic therapies right now are genetic diseases such as sickle cell anemia, Huntington’s disease, and Cystic Fibrosis. For me, it’s easy to say yes to eliminating these malicious killers from humanity’s genome; but techniques that take us closer to curing these diseases will inevitably bring with them the ability to make horrifically bad decisions. Technology has always been a double edged sword.
And What About AI?
Like genetic engineering, artificial intelligence (and especially machine learning) has been exploding lately. From self driving cars, to cancer screening, to playing Go; but also from China’s Orwellian and algorithm driven surveillance program, to racial profiling, to computers that invented their own encryption. As in genetics, so it is in AI: we have to be careful, thoughtful, and intentional about how we proceed with the use and application of these technologies.
Time recently published a wonderful breakdown of how the “data driven revolution” — touted as a way for humanity to become more objective — actually ends up reflecting and sometimes amplifying the biases of the world. Searches for “unprofessional hairstyle” disproportionately display the perfectly professional hairstyles of black women. A facial recognition AI infamously classified several black men as gorillas. Similar algorithms are used by banks to determine if someone qualifies for a mortgage — we should be suspicious of claims that being “data driven” de facto reduces bias. The AI’s we build reflect the data we put into them, and the data humanity generates reflects the biases we all have.
By now you’re probably also aware of Cambridge Analytica’s use of artificial intelligence to target voters, and spread polarizing content online. We’ve become increasingly reliant on algorithms to help us determine what news, entertainment, media, and information to consume. We are just starting to understand how Facebook, Twitter, and similar mediums have impacted our information diet (and as a result our thoughts and opinions). Just like in genetics, we need the public to be more informed and thoughtful about how these things work if we’re going to use them for the betterment of society.
In one way though, AI is quite different from genetics: changes in AI are adopted much more quickly — and in my opinion — less thoughtfully than changes in genetics. Everyone with the power to enact genetic editing pretty much understands the gravitas associated with doing so. But many in the technology world have acted executively and carelessly in deploying AI systems. Others have done so maliciously. Several gambling organizations have been caught using AI to target those susceptible to gambling addiction. Others, including the NSA and Google, are tracking massive amounts of information, and increasingly machine learning is used to understand that data. Whether you’re flagged as a good candidate for a particular advertisement, a good candidate for active surveillance, or a potential addict, the mechanisms are more algorithmic than ever.
I think it’s a safe bet that more people will know how to program in the future. I think it’s also a safe bet that continued hardware, software, and algorithmic breakthroughs will all combine to make this technology more powerful — and more accessible — very quickly. The computer’s ubiquity, and the connective power of the internet, has put the powers of AI at the collective fingertips of anybody with some money and some programming skills. All you really need to deploy the highest scale machine learning algorithms is a credit card and an Amazon Web Services account. Whatever becomes possible to do with computers, there is a good chance someone will try to actually do it.
All of these challenges — ethical, pragmatic, and legal — draw me to both AI and genetics. I want to be able to face these questions head on, armed with the knowledge and understanding of how they work, and what they can do. I also think that there is a lot of potential for synergy, and potential to change the world for the better at the intersection of these two disciplines.
Hybrid Vigor and the Cult of the Renaissance Person
I am fascinated by the way math, biology, and the other sciences play off of each other. I believe in the cult of the polymath and reject the saying, “Jack of all trades, master of none.” To me, the real magic happens at the confluence of multiple fields. To steal a phrase from the genetics world, I think that the intellectual children of AI and genetics can display hybrid vigor.
The incredible advances in Artificial Intelligence and Machine Learning are smashing previous barriers in tons of fields, across many disciplines. So called “Data Science” has enabled incredible progress in a remarkable number of different contexts. Machine learning powered computer vision has been applied to self driving cars, medical imaging, robotics control, agriculture, surveillance, and more.
Light carries many information rich signals; how many species rely heavily on vision to navigate the world? It turns out that neural networks, a class of algorithm that was inspired by the way human neurons work, is much better than its predecessors at interpreting the signals contained in light. In particular, convolutional neural nets are blazing trails in computer vision.
In game playing, AlphaGo made waves when it bested Lee Sedol 4 games to 1. The creators of AlphaGo combined reinforcement learning and neural networks to defeat a world class Go player (which is an understatement about the legendary Lee Sedol). Genetic problems have already been modeled as games for scientific pursuits. Fold It is a game that harnesses the power of “The Meat Cloud” to solve genetic puzzles. Perhaps an AI could “fold it” as well.
The successful applications of AI techniques are too many to enumerate, but if computers can automatically detect meaningful patterns in light, and meaningful patterns in Go, they can detect meaningful patterns in genetic data too. And, if we’re open to it, perhaps the process of discovery in AI can teach us something more about consciousness, behavior, and thought in animals (for example, humans).
Over the next year, I’m going to be teaching myself the fundamentals of genetics as I work towards my graduate school goal: knowing more about genetics than most AI researchers, and knowing more about AI than most genetics researchers. Doing so satisfies my own desire to be more multidisciplinary; I have spent 10 years working towards mastery in programming and it’s time to branch out. But I also have a deeper purpose than simply learning more things: I want to share this information too.
Some People Just Want to Watch The World Learn
Like AI and genetics, education is in the process of a technology fueled revolution. There are more ways than ever for ambitious learners to teach themselves; from the stalwarts like Wikipedia, to newer entrants like Khan Academy, DuoLingo, Coursera, Udemy, Udacity. There is also a wealth of amazing content written by experts on their own blogs, youtube channels, or platforms like Medium and Quora.
Although I’m embarking on a journey through a traditional educational institution, I am also a believer in new and innovative ways to spread and share information. Some of the most exciting ideas are still locked up behind the doors of academia, and hard to find anywhere else. I want to go to graduate school to get some of that information. But I want to be a part of liberating and spreading that information too — and “new media” will have to be a part of that.
While education is becoming more accessible, it’s also becoming harder and harder to separate the wheat from the chaff in online venues. Misinformation and bad information is everywhere; people are shouting “fake news” and “propaganda” but we can barely hear it over the roar of endless in-your-face advertising pumped up to 11. Because of this, I want to do my part to share high quality information and educational content with the world.
As I begin to prepare for graduate school, and hopefully throughout the course of my degree(s), I plan to write down, record, and teach as much as I can. I hope you’ll enjoy following along as I start a new chapter in my own life, and I hope we’ll all learn something in the process.
This article was produced by Teb’s Lab, visit www.tebs-lab.com/about to learn more.