The Century of Complexity/ Chipmaking Being Redesigned/ Human Brain Architecture and AI/ Environment “Rewilded”/ VR Problems and Game Developers

The Century of Biology. Till not long ago I deeply believed in what many others have been saying all along: that the 20th century has been the century of physics and that the 21st century will be the century of Biology. Till I stumbled into this quote by Stephen Hawkings, mentioned by Geoffry West in his incredible and seminal book Scale.

In an interview at the turn of the millennium Hawkins was asked about the 21st century being the century of biology he responded:

I think the next century will be the century of complexity.

Similarly to West, I also wholeheartedly agree. Biology will play a very important role in that for sure, and the developments we are seeing in life science and synthetic biology are proving this. But complexity and complex adaptive systems will be the key determinant of the century. 

And this poses huge socio-economic challenges, as complex adaptive systems are antithetical to the notions of linear thinking and causality, to which we normally default in dealing with reality. Leading and managing in a complex adaptive systems environment requires very different skills and approaches. 

It requires pragmatism rather than intellectualism, resilience rather than efficiency, experimentation rather than deduction, indirect rather than direct approaches, plurality rather than universality, holism rather than reductionism.

The last point, holism rather than reductionism, is in striking contrast with most of the digital approach which has characterized the last decades. In dealing with complexity it is not possible to break things down into modules, and then create APIs for the interactions, the whole is more the sum of the parts, and needs to be understood as such.

Which brings me to one of the things that frustrates me the most. The reductionist compulsion compounds with our (as society) ever-decreasing attention span and the ever-increasing desire for simplicity. The problem is that the world is NOT simple, it is complex, and it is getting more and more complex, and we need to deal with this reality. As Einstein once said, “Everything should be made as simple as possible, but not simpler”. We should stop pretending that things are simple when they are not. This is only going to backfire (and it is already). I think the time has come to embrace complexity and learn how to deal with it. Time for emergence.

It is not going to be simple, but it is necessary.

Chipmaking Is Being Redesigned. Effects Will Be Far-Reaching

When TSMC said they will increase their capital expenditures from $17.2B to $25B-$28B in 2021, the industry was shocked. In a WSJ piece, a partner of a technology research firm said this was a "monster number" and that it signals "how [TSMC] thinks about the economic recovery and also the confidence they have in their long-term growth." Simply put, the world is demanding more microchips - so much so that a few weeks ago Honda had to shut down an England factory due to a shortage of chips. Volkswagen said it will make 100k fewer vehicles in Q1 2021 for the same reason.

All this is happening amidst an industry shakeup - startups are challenging giants with specialized chips, neuromorphic chips and other alternatives are finally becoming viable, and web giants like Amazon, Apple, and Google are tackling their own hardware. In addition, there are a shrinking number of processor manufacturers - only Intel, TSMC, and Samsung are able to fabricate advanced chips. Meanwhile, ambitious chip startups are getting ballooning valuations: "Cerebras, an American firm which designs AI chips, has earned one of $1.2bn. A British rival called Graphcore, which has been working with Microsoft, was valued at $2.8bn in December. On January 13th Qualcomm, a firm best-known for its smartphone chips, paid $1.4bn for Nuvia, a startup staffed by veterans of Apple’s in-house chip-design team."

News items:

These Virtual Robot Arms Get Smarter by Training Each Other

A virtual robot arm has learned to solve a wide range of different puzzles—stacking blocks, setting the table, arranging chess pieces—without having to be retrained for each task.

A Closer Look at the AI Incident Database of Machine Learning Failures

The AI Incident Database a repository of documented failures of AI systems in the real world.

How Mirroring the Architecture of the Human Brain Is Speeding Up AI Learning

Last February, Vox interviewed Alison Gopnik, who proposed that "studying kids can give programmers useful hints about directions for computer learning." The developing human brain has amazing zero-shot capabilities, such as inferring what new objects might be without ever having seen them, or grasping new concepts after only a few repeated examples. By contrast, deep learning is largely inefficient, requiring enormous datasets and powerful compute to generate, say, an English sentence. In a recent paper in Frontiers in Computational Neuroscience, two researchers borrowed some concepts from the human brain to create an AI model that learns new concepts with only a handful of examples:

"First they trained the AI on 2.5 million images across 2,000 different categories from the popular ImageNet dataset. They then extracted features from various layers of the network, including the very last layer before the output layer. They refer to these as 'conceptual features' because they are the highest-level features learned, and most similar to the abstract concepts that might be encoded in the [anterior temporal lobe].... They then used these different sets of features to train the AI to learn new concepts based on 2, 4, 8, 16, 32, 64, and 128 examples. They found that the AI that used the conceptual features yielded much better performance than ones trained using lower-level features on lower numbers of examples, but the gap shrunk as they were fed more training examples."

News items:

Apple's First-Ever Headset Will Reportedly Be a Pricey Device That Paves the Way for an Eventual iPhone Replacement

Apple's first major new product since the Apple Watch will be an expensive headset designed to set the stage for augmented reality glasses.

Hydrogen Fuel Cell Offers Drones Up to Two-Hour Flights

A South Korean company says its fuel cell technology provides enough energy to sustain two-hour drone flights.

Why We Need to “Rewild” Our Environment

In 2018, Bloomberg published a piece on land use in the US with the above image, showing just how much agricultural land occupies the country. Agtech in general is promising because agriculture is exorbitantly wasteful, requiring "unsustainable amounts of water, fertilizer, and land." While meat and dairy are entrenched industries in the US - where citizens consume more dairy products and more meat than the decades before - 4IR technology like vertical farming, plant-based meat, cultured meat, and dairy alternatives are taking up much less land and causing a much smaller impact on the environment. Design firm Stantec is even looking at ways to rewild vast swaths of agricultural land. Some countries, like the UK, are already planning to turn hectares of land into woodlands again to fight climate change. Here are a few ways new tech can help free up land for rewilding:

●      "Vertical farms, including systems that plant, grow, and harvest food autonomously, can use as little as 1% of the land that a conventional farm would use to grow the same amount of food."

●      "Impossible Burger, one of the pioneers of plant-based meat that tastes nearly identical to the real thing, uses96% less land to make a burger than if it had been made from beef."

●      "[Meat] grown from animal cells in bioreactors, also shrinks land use and is poised to soon come to market, with the first regulatory approval recently announced."

News items:

Project on Self-Deleting Genes Takes Aim at Mosquito-Borne Diseases

A Texas A&M team of researchers plans to test a gene drive, a DNA-cutting enzyme and a small repeat of the insect’s own DNA in fruit flies and mosquitoes.

Why Scientists Say ‘Plastivores’ Could Be the Solution to Plastic Pollution

In 2016, Japanese scientists discovered the first plastivores in the form of a bacterium that was able to survive solely on PET bottles.

Virtual Reality Has Real Problems. Here’s How Game Developers Seek to Delete Them.

When moving around in a game in virtual reality, the sensory jumble between the visual and vestibular system causes motion sickness for some. Basically, "the player’s eyes tell their brain they are walking, while their body tells them that they are still." A quick solution in many titles is teleporting instead of animating the character's walking movements. Other solutions have been more experimental, like the WalkingVibe apparatus, which delivers haptic feedback directly to the head to reduce nausea. Hardware and software fixes have also been implemented for eyestrain, dizziness, poor eyesight, head size, or simply to help users avoid crashing into (real) walls

VR game design is still a wild west of experimentation while developers figure out ways to minimize physical side-effects and introduce fun ways to interact with virtual worlds. Over the past several years, developers have learned a few tricks, like controlling what the user focuses on and darkening the edges of the screen. A more advanced solution to keep players immersed is full hand tracking support, which translates free hand movement to in-game actions without needing a controller. Even adding a "jump" button is a controvertial move:

"One radical departure, surfaced through a willingness to throw convention out the window (along with a pinch of stubbornness), was the introduction of a jump button. 'For a long time, there was resistance to a jump, and I was convinced it would work, and basically I just put it in the game, and people got used to it,' said Joel Bartley, lead gameplay programmer on 'Stormland,' and 'Edge of Nowhere.'"

News items:

To Expose The Bigotry Of AI, Artist Trevor Paglen Is Putting Computers On Trial

Trevor Paglen is to artificial intelligence what Upton Sinclair was to meatpacking - an important exhibition at the Carnegie Museum of Art shows some of what he’s uncovered. 

‘For Some Reason I’m Covered in Blood’: GPT-3 Contains Disturbing Bias Against Muslims

"The algorithm itself is unchanged: The bias is programmed into GPT-3.... Is an algorithm that mindlessly spews hate the kind of technology companies want to put into the world?"

It was OpenAI themselves that stated in a paper last year that "biases present in training data may lead models to generate stereotyped or prejudiced content." When Facebook AI head Jerome Pesenti tweeted that GPT-3 could "easily output toxic language that propagates harmful biases," OpenAI CEO Sam Altman responded that "he shared those concerns, and he argued that part of the reason the nonprofit was starting off GPT-3 in a closed beta was to do safety reviews before it went fully live." You take from the internet, you get back part of the internet.

Despite all these caveats floating around, it falls to external research teams to uncover harmful biases ingrained in large AI models and present them to the public. This month, researchers from Stanford and McMaster published a paper demonstrating that, given a prompt containing "Muslim" or related word, GPT-3 produced "sentences associating Muslims with shooting, bombs, murder, and violence" about 60% of the time. At its current state, a GPT-3-based product promising to produce public-facing content would need constant human oversight... at best.

"Suppose GPT-3 was used to automatically caption images. Stanford and McMaster researchers actually studied this specific functionality already: In the experiment, short captions were generated by a version of GPT-3 specifically trained to recognize a given set of images, and then researchers asked the standard GPT-3 algorithm to add more text to those stubs. Images depicting people wearing headscarves were more likely to be given captions associated with violence."

News items:

Tech Companies Are Profiling Us From Before Birth

Children today are the very first generation of citizens to be "datafied" from before birth.