0 Members and 1 Guest are viewing this topic.
Welcome! In this post, we’ll be taking a character-by-character look at the source code of the BioNTech/Pfizer SARS-CoV-2 mRNA vaccine.Now, these words may be somewhat jarring - the vaccine is a liquid that gets injected in your arm. How can we talk about source code?This is a good question, so let’s start off with a small part of the very source code of the BioNTech/Pfizer vaccine, also known as BNT162b2, also known as Tozinameran also known as Comirnaty.First 500 characters of the BNT162b2 mRNA. Source: World Health OrganizationThe BNT162b mRNA vaccine has this digital code at its heart. It is 4284 characters long, so it would fit in a bunch of tweets. At the very beginning of the vaccine production process, someone uploaded this code to a DNA printer (yes), which then converted the bytes on disk to actual DNA molecules.A Codex DNA BioXp 3200 DNA printerOut of such a machine come tiny amounts of DNA, which after a lot of biological and chemical processing end up as RNA (more about which later) in the vaccine vial. A 30 microgram dose turns out to actually contain 30 micrograms of RNA. In addition, there is a clever lipid (fatty) packaging system that gets the mRNA into our cells.RNA is the volatile ‘working memory’ version of DNA. DNA is like the flash drive storage of biology. DNA is very durable, internally redundant and very reliable. But much like computers do not execute code directly from a flash drive, before something happens, code gets copied to a faster, more versatile yet far more fragile system.For computers, this is RAM, for biology it is RNA. The resemblance is striking. Unlike flash memory, RAM degrades very quickly unless lovingly tended to. The reason the Pfizer/BioNTech mRNA vaccine must be stored in the deepest of deep freezers is the same: RNA is a fragile flower.Each RNA character weighs on the order of 0.53·10⁻²¹ grams, meaning there are 6·10¹⁶ characters in a single 30 microgram vaccine dose. Expressed in bytes, this is around 25 petabytes, although it must be said this consists of around 2000 billion repetitions of the same 4284 characters. The actual informational content of the vaccine is just over a kilobyte. SARS-CoV-2 itself weighs in at around 7.5 kilobytes.
SummarisingWith this, we now know the exact mRNA contents of the BNT162b2 vaccine, and for most parts we understand why they are there:- The CAP to make sure the RNA looks like regular mRNA- A known successful and optimized 5’ untranslated region (UTR)- A codon optimized signal peptide to send the Spike protein to the right place (copied 100% from the original virus)- A codon optimized version of the original spike, with two ‘Proline’ substitutions to make sure the protein appears in the right form- A known successful and optimized 3’ untranslated region- A slightly mysterious poly-A tail with an unexplained ‘linker’ in thereThe codon optimization adds a lot of G and C to the mRNA. Meanwhile, using Ψ (1-methyl-3’-pseudouridylyl) instead of U helps evade our immune system, so the mRNA stays around long enough so we can actually help train the immune system.
In this video Elon Musk talks about Tesla Full Self driving software remotely at a Chinese AI conference. Elon predicts that Tesla will achieve level 5 autonomy soon and sooner than people can imagine. Elon also indirectly criticizes Waymo, a googles self-driving software company. Waymo depends on LiDAR and HD maps. Most of the time, they train their self-driving software and car in simulation.
Wherein Dr. Know-it-all discusses the work of Dr. Arthur Choi (UCLA) and others concerning the quest to understand how deep convolutional neural networks function. This new field, XAI, or explainable AI, uses decision trees, formal logic, and even tractable boolean circuits (simulated logic gates) to explain why machine learning using deep neural nets functions so well some of the time, but so poorly other times.
Artificial Intelligence has been getting a bad rap of late, with numerous opinion pieces and articles describing how it has struggled to live up to the hype. Arguments have centered around computational cost, lack of high-quality data, and the difficulty in getting past the high nineties in percent accuracy, all resulting in the continued need to have humans in the loop.
AI & ML are simply tools for building complex (and sometimes non-linear) models that consider large amounts of information. They are most potent in applications where their pattern finding power significantly exceeds human capability. If we adjust our attitude and expectations, we can leverage their power to bring about all sorts of tangible outcomes for humanity.With this type of re-calibration, our mission should be to use AI to help human decision makers, rather than replace them. Machine learning is now being used to build weather and climate impact models that help infrastructure managers respond with accuracy and allocate their resources efficiently. While these models do not perfectly match the ground truth, they are much more accurate and precise than simple heuristics, and can save millions of dollars through more efficient capital allocation.
When advertisers create a Facebook ad, they target the people they want to view it by selecting from an expansive list of interests. “You can select people who are interested in football, and they live in Cote d’Azur, and they were at this college, and they also like drinking,” Goga says. But the explanations Facebook provides typically mention only one interest, and the most general one at that. Mislove assumes that’s because Facebook doesn’t want to appear creepy; the company declined to comment for this article, so it’s hard to be sure.Google and Twitter ads include similar explanations. All three platforms are probably hoping to allay users’ suspicions about the mysterious advertising algorithms they use with this gesture toward transparency, while keeping any unsettling practices obscured. Or maybe they genuinely want to give users a modicum of control over the ads they see—the explanation pop-ups offer a chance for users to alter their list of interests. In any case, these features are probably the most widely deployed example of algorithms being used to explain other algorithms. In this case, what’s being revealed is why the algorithm chose a particular ad to show you.The world around us is increasingly choreographed by such algorithms. They decide what advertisements, news, and movie recommendations you see. They also help to make far more weighty decisions, determining who gets loans, jobs, or parole. And in the not-too-distant future, they may decide what medical treatment you’ll receive or how your car will navigate the streets. People want explanations for those decisions. Transparency allows developers to debug their software, end users to trust it, and regulators to make sure it’s safe and fair.The problem is that these automated systems are becoming so frighteningly complex that it’s often very difficult to figure out why they make certain decisions. So researchers have developed algorithms for understanding these decision-making automatons, forming the new subfield of explainable AI.
In 2017, the Defense Advanced Research Projects Agency launched a US $75 million XAI project. Since then, new laws have sprung up requiring such transparency, most notably Europe’s General Data Protection Regulation, which stipulates that when organizations use personal data for “automated decision-making, including profiling,” they must disclose “meaningful information about the logic involved.” One motivation for such rules is a concern that black-box systems may be hiding evidence of illegal, or perhaps just unsavory, discriminatory practices.
As a result, XAI systems are much in demand. And better policing of decision-making algorithms would certainly be a good thing. But even if explanations are widely required, some researchers worry that systems for automated decision-making may appear to be fair when they really aren’t fair at all.For example, a system that judges loan applications might tell you that it based its decision on your income and age, when in fact it was your race that mattered most. Such bias might arise because it reflects correlations in the data that was used to train the AI, but it must be excluded from decision-making algorithms lest they act to perpetuate unfair practices of the past.The challenge is how to root out such unfair forms of discrimination. While it’s easy to exclude information about an applicant’s race or gender or religion, that’s often not enough. Research has shown, for example, that job applicants with names that are common among African Americans receive fewer callbacks, even when they possess the same qualifications as someone else.A computerized résumé-screening tool might well exhibit the same kind of racial bias, even if applicants were never presented with checkboxes for race. The system may still be racially biased; it just won’t “admit” to how it really works, and will instead provide an explanation that’s more palatable.Regardless of whether the algorithm explicitly uses protected characteristics such as race, explanations can be specifically engineered to hide problematic forms of discrimination. Some AI researchers describe this kind of duplicity as a form of “fairwashing”: presenting a possibly unfair algorithm as being fair. Whether deceptive systems of this kind are common or rare is unclear. They could be out there already but well hidden, or maybe the incentive for using them just isn’t great enough. No one really knows. What’s apparent, though, is that the application of more and more sophisticated forms of AI is going to make it increasingly hard to identify such threats.
No company would want to be perceived as perpetuating antiquated thinking or deep-rooted societal injustices. So a company might hesitate to share exactly how its decision-making algorithm works to avoid being accused of unjust discrimination. Companies might also hesitate to provide explanations for decisions rendered because that information would make it easier for outsiders to reverse engineer their proprietary systems. Cynthia Rudin, a computer scientist at Duke University, in Durham, N.C., who studies interpretable machine learning, says that the “explanations for credit scores are ridiculously unsatisfactory.” She believes that credit-rating agencies obscure their rationales intentionally. “They’re not going to tell you exactly how they compute that thing. That’s their secret sauce, right?”And there’s another reason to be cagey. Once people have reverse engineered your decision-making system, they can more easily game it. Indeed, a huge industry called “search engine optimization” has been built around doing just that: altering Web pages superficially so that they rise to the top of search rankings.
MIT researchers have developed a type of neural network that learns on the job, not just during its training phase. These flexible algorithms, dubbed "liquid" networks, change their underlying equations to continuously adapt to new data inputs. The advance could aid decision making based on data streams that change over time, including those involved in medical diagnosis and autonomous driving.
Hasani designed a neural network that can adapt to the variability of real-world systems. Neural networks are algorithms that recognize patterns by analyzing a set of "training" examples. They're often said to mimic the processing pathways of the brain—Hasani drew inspiration directly from the microscopic nematode, C. elegans. "It only has 302 neurons in its nervous system," he says, "yet it can generate unexpectedly complex dynamics."Hasani coded his neural network with careful attention to how C. elegans neurons activate and communicate with each other via electrical impulses. In the equations he used to structure his neural network, he allowed the parameters to change over time based on the results of a nested set of differential equations.
Risky behaviors such as smoking, alcohol and drug use, speeding, or frequently changing sexual partners result in enormous health and economic consequences and lead to associated costs of an estimated 600 billion dollars a year in the US alone. In order to define measures that could reduce these costs, a better understanding of the basis and mechanisms of risk-taking is needed.
Specific characteristics were found in several areas of the brain: In the hypothalamus, where the release of hormones (such as orexin, oxytocin and dopamine) controls the vegetative functions of the body; in the hippocampus, which is essential for storing memories; in the dorsolateral prefrontal cortex, which plays an important role in self-control and cognitive deliberation; in the amygdala, which controls, among other things, the emotional reaction to danger; and in the ventral striatum, which is activated when processing rewards.
The researchers were surprised by the measurable anatomical differences they discovered in the cerebellum, an area that is not usually included in studies of risk behaviors on the assumption that it is mainly involved in fine motor functions. In recent years, however, significant doubts have been raised about this hypothesis – doubts which are now backed by the current study.
“It appears that the cerebellum does after all play an important role in decision-making processes such as risk-taking behavior,” confirms Aydogan. “In the brains of more risk-tolerant individuals, we found less gray matter in these areas. How this gray matter affects behavior, however, still needs to be studied further.”
Joscha Bach (@Plinz) tweeted at 11:20 AM on Fri, Jan 29, 2021:In the long run, machine learning and a publicly accessible stock market cannot coexist(https://twitter.com/Plinz/status/1355007909281718274?s=03)
Joscha Bach (@Plinz) tweeted at 7:54 PM on Fri, Jan 29, 2021:The financial system is software executed by humans, full of holes and imperfections, and very hard to update and maintain. Using substantial computational resources to discover and exploit its imperfections will eventually nuke it into oblivion(https://twitter.com/Plinz/status/1355137134789681158?s=03)
Driverless robotaxis are now available for public rides in ChinaAutoX is the first in the country to offer rides without safety drivers.
After lots of tests, it’s now possible to hail a truly driverless robotaxi in China. AutoX has become the first in the country to offer public rides in autonomous vehicles without safety drivers. You’ll need to sign up for a pilot program in Shenzhen and use membership credits, but after that you can hop in a modified Chrysler Pacifica to travel across town without seeing another human being.
Fully driverless robotaxis are still very rare anywhere in the world, and it’ll take a combination of refined technology and updated regulation before they’re relatively commonplace. This is an important step in that direction, though. They might get a boost in the current climate, though. The COVID-19 pandemic has added risk to conventional ride hailing for both drivers and passengers, and removing drivers could make this one of the safest travel options for people without cars of their own.
In the blog post where it declared the GPT-3 API, OpenAI stated three key reasons for not open-sourcing the deep learning model. The first was, obviously, to cover the costs of their ongoing research. Second, but equally important, is running GPT-3 requires vast compute resources that many companies don’t have. Third (which I won’t get into in this post) is to prevent misuse and harmful applications.Based on this information, we know that to make GPT-3 profitable, OpenAI will need to break even on the costs of research and development, and also find a business model that turns in profits on the expenses of running the model.
In general, machine learning algorithms can perform a single, narrowly defined task. This is especially true for natural language processing, which is much more complicated than other fields of artificial intelligence. To repurpose a machine learning model for a new task, you must retrain it from scratch or fine-tune it with new examples, a process known as transfer learning.But contrary to other machine learning models, GPT-3 is capable of zero-shot learning, which means it can perform many new tasks without the need for new training. For many other tasks, it can perform one-shot learning: Give it one example and it will be able to expand to other similar tasks. Theoretically, this makes it ideal as a general-purpose AI technology that can support many new applications.
For all the talk about how artificial intelligence technology is transforming entire industries, the reality is that most businesses struggle to obtain real value from AI. 65% of organizations that have invested in AI in recent years haven’t yet seen any tangible gains from those investments, according to a 2019 survey conducted by MIT Sloan Management Review and the Boston Consulting Group. And a quarter of businesses implementing AI projects see at least 50% of those projects fail, with “lack of skilled staff” and “unrealistic expectations” among the top reasons for failure, per research from IDC.
Encouragingly, AI is already being leveraged to simplify other tech-related tasks, like writing and reviewing code (which itself is built by AI). The next phase of the deep learning revolution will involve similar complementary tools. Over the next five years, expect to see such capabilities slowly become available commercially to the public.
One of the latest collaborations between artificial intelligence and humans is further evidence of how machines and humans can create better results when working together. Artificial intelligence (AI) is now on the job to combat the spread of misinformation on the internet and social platforms thanks to the efforts of start-ups such as Logically. While AI is able to analyze the enormous amounts of info generated daily on a scale that's impossible for humans, ultimately, humans need to be part of the process of fact-checking to ensure credibility. As Lyric Jain, founder and CEO of Logically, said, toxic news travels faster than the truth. Our world desperately needs a way to discern truth from fiction in our news and public, political and economic discussions, and artificial intelligence will help us do that.
The Fake News “Infodemic”People are inundated with info every single day. Each minute, there are 98,000 tweets, 160 million emails sent, and 600 videos uploaded to YouTube. Politicians. Marketers. News outlets. Plus, there are countless individuals spewing their opinions since self-publishing is so easy. People crave a way to sort through all the information to find valuable nuggets they can use in their own life. They want facts, and companies are starting to respond often by using machine learning and AI tools.
As the pursuit of fighting fake news becomes more sophisticated, technology leaders will continue to work to find even better ways to sort out fact from fiction also well as refine the AI tools that can help fight disinformation. Deep learning can help automate some of the steps in fake news detection, according to a team of researchers at DarwinAI and Canada's University of Waterloo. They are segmenting fact-checking into various sub-tasks, including stance detection where the system is given a claim on a news story plus other stories on the same subject to determine if those other stories support or refute the claim in the original piece.