Section 5 of “Reinventing Societal Infrastructure with Technology” which will be released end of January. I will be posting a new section daily. Please share your feedback as this is a work in progress.
Fundamental reinvention has never been more possible than it is today. There are a range of new recent technological axes of development that give me hope. Driver technology tools that are plausible and visible today are:
AI & large scale data capability
Additive manufacturing / 3D printing
Biotechnology (ohmic measurement, CRISPR, gene synthesis, precision control of genes, pathways…)
Computational design, computational modelling/simulation & computer networking
Social connectivity & networking; distributed access
Software eating the world
Other new still fermenting ideas I have surely missed or underestimated
AI will, inevitably, change the structure of our society. This is not a statement about exciting future technical developments, but merely an observation about what is ALREADY possible now, although has not yet achieved widespread adoption. The rate of change of new AI capability, the building block for changing businesses and human activity, is very rapidly expanding.
Fundamentally, we are now able to achieve human-like (and occasionally super-human) performance on tasks that were, just a few years ago, regarded as completely out of reach of machines. Probably the greatest example here is computer vision. It was stagnant for decades, but has made so much progress in the last years we can now have computers classify images and videos with super human performance, provided we have enough training data in the domain, be it face recognition or reading MRI images. The same is becoming true in recognizing human speech and even generating voice, or reading someone’s mood or mental health status from their voice in superhuman ways. Just thinking about machines with the capability to understand vision and recognizing voice by itself, will fundamentally change how we think of work in general and what our interface with machines will look like in the future.
That being said, there is likely an even greater progress possible near term. Currently, the best performing AI systems require huge amounts of data to train to human-like performance. However, work is underway to reduce this burden in various domains; often all is needed are humans being to feed just a few examples to guide the neural nets. This will enable us to apply AI to domains where little data exists or the data is hard to get for structural or legal reasons, greatly widening the applicability of AI in all business and societal processes.
Model-building of the world as humans do is another dimension of innovation; an AI may be able to predict how much force would knock over a glass of water. AIs are also learning fast from the world of simulations and games. I suspect the lego blocks of types of intelligence will expand from a few (names like CNN’s, RNN’s, GAN’s, to more recent additions like probabilistic programming, Bayes nets (redone), graph models, …. to all kinds of new capability hard to predict or name today) to many different types of intelligences capable of being combined to do unusual things; much like lego blocks can enable building very complex structures compared to the original red and yellow blocks.
It is not only the ability of AI to “judge” and recognize images and audio that is a driver here. We are now on the cusp of having AI generate images in a domain, at high resolution, and high quality that mimic art or any desired input distribution. The same now goes for music, where we can create AI that can mimic and improvise play in any style. “Creativity” used to be one of the standard answers to the question of what humans are uniquely good at that machines can never do. Nonetheless, it looks increasingly doubtful that even that claim is irrefutable. One day, we might see a short film, generated solely by an AI, or a top ten hit that never had a human composer.
One last comment that should make us quite hopeful about the accessibility of these technologies is that most of the fundamental breakthroughs have been out in the open, published, and discussed publicly. Yes, AI talent is hard to come by today. At the same time, it is also one of the most popular areas of study these days, and sooner or later this challenge is going to be overcome. Coupled with high-quality frameworks for AI research and deployment now being freely available as open source, rapid progress on both research and applications is at hand.
In a talk to the National Bureau of Economic Research on “Economic Implications of AI,” I looked at the top twenty employment categories in the US and concluded most jobs in most of these categories would be eliminated or change substantially for humans. Technology will reallocate where and how people spend time and resources. We will have great abundance, growing productivity, and GDP but with increasing income disparity. Further, changes will be slow, almost imperceptible in terms of employment the first five or ten years, and take decades before going exponential in actual number of jobs impacted. But by the time the first 5% of jobs are impacted, the future will be inevitable.
The renewed interest in robotics is, to a large extent, similar to the renewed interest in AI. As a society, for a long time we had robots that were amazingly durable, amazingly precise, but fundamentally were just acts of good mechanical engineering and careful motor control. This was enough to solve manufacturing tasks in very structured environments where all parts have defined positions and the manufacturing line does not change rapidly. A core example here is the chassis production of cars, which has little human involvement today; but come to sorting eggs by size and grade, and only human assembly line workers could do that.
The new part in robotics deals with robots that can make decisions in a largely unstructured environment. Probably the most discussed example of this today are self-driving cars that have to make decisions in the real world and not in a defined, pre-planned environment. But there are other, equally broad implications on the horizon. A company struggling with automation due to dealing with soft materials and rapidly changing product mixes right now faces large costs of automation. However, the next generation robots might change this.
The main driver for this is two-fold. One is straightforward, it is simply the availability of very low-cost, high-resolution sensors, in particular camera systems and 3D acquisition technologies. The other boils down to the fact we can now interpret vision and 3D data by learning from examples, instead of having to hand-code the rules. Reinforcement learning, learning from simulation, and understanding how to reduce the training samples required are the core elements of modern robotics.
Those robots are a very different breed from the old, and the trade-off space will be vastly different. Formerly, we got precision from adding tighter motor control or heavier arms.. This new class of robots have cheaper, lighter arms and still get the precision back by relying on visual servoing. In essence, it means the vision system is able to correct the robotic arm as it gets close to the object we wish to manipulate. A robot arm capable of doing human tasks should not weigh any more than a human arm does and then scale sub-linearly from there. This makes this next generation of robots cheaper, able to handle very flexible tasks, and deal with environments that have been thought as impossible in robotics before.
From a societal and economic perspective this enables a completely new way of thinking about production lines. Proximity to the end-customer, thus, becomes more important than the availability of cheap labor for menial tasks in unstructured environments or the need for scale. This is especially true when combined with new technologies like 3D-printing. Custom, personalized, and local may become economically better in areas like producing jeans, sofas and beds or many types of fresh food.
Additive Manufacturing/3D printing
Additive manufacturing, in essence printing objects instead of manufacturing them traditionally, has already made inroads in multiple areas. It consists of a family of technologies that can manufacture polymer parts to high-density metal parts. Even composites are being 3D-printed. The current beachheads for those technologies have been largely in design and prototyping environments. This means shortening the design cycle as we can almost instantly have a prototype part; it has actually already become a standard feature for many industries.
This is, however, changing. We are seeing more and more production parts made by additive manufacturing. Using these techniques to create performance critical parts that are not manufacturable with traditional methods is already becoming commonplace. Examples here include turbine parts, rocket engine components, and implants. This acts as the key catalyst to move the industry from using the technology for prototyping to a manufacturing regime. We climb down the cost curve as an ever greater number of parts that used to be hard to customize, not buildable at all, or consisted of multiple assemblies, can now be built with these machines.
We are now tackling some of the fundamental limitations of the technology, such as cost per part, materials we can use, removal or avoidance of necessary support structures to make this family of technologies even more widely applicable. These technologies, in turn, are also changing conventional wisdom like benefits of scale, locations, and schedules for manufacturing, supply chains, spare parts, or maintenance. Do we need to make shoes in China for US consumption or can they be 3D-printed locally and customized to each foot? Do we need to stock every spare part for a Boeing 727 in every airport in the world? Should it take six months to get a sofa manufactured in China only to see it does not fit in your small studio apartment?
This has significant consequences for the way we think about complexity in our design. If complexity becomes in essence free, that is not tied to manufacturing steps, our possible design space explodes. In particular, if manufacturing complexity is not the bottleneck anymore, designing structures will be the new bottleneck. Instead of designing by hand, we will likely create them by specifying the input loads and tasks. Optimization software will process the data to create structures looking a lot more organically than now and producible ONLY by additive manufacturing.
Computational Design / Learning Physics
In design of objects we have long used computer tools from EDA tools for the electronics industry to CAD tools for physical objects and from simulations for verification of performance of the designed objects.
We are now moving into a regime where the actual act of designing a structure is now becoming part of the duty of our tools and humans act more as a trainer, judge and specifier of external conditions. And even those roles may change to just specifying goals given constraints and preferences. Let’s take designing a structure for an aircraft. It has to fulfill various structural loads and remain as light as possible at the same time. We are now capable to let the algorithms decide the design to minimize weight and specify the external loads. This has been possible before via optimization procedures, but unlike before, we can “learn” from past successful designs and guide the search space.
In no place is this more apparent than in the design of drug targets. Instead of running quantum simulations to understand the binding of molecules to targets, a slow and costly procedure, we can now learn from past binding data to automatically come up with novel designs that might be good candidates for a new drug. A general principle behind it being that even though we often understand the underlying physics of what we are trying to design, the exploration process is too costly to run by brute force exploration of the design space. Learning from past successful designs, be they molecules designed for a target, physical objects, or different layouts on a circuit board, allows us to meaningfully change the performance of these objects. We may soon see a new range of computationally-designed materials beyond copper, steel, and aluminium alloys for everything, from medical devices to body organs to your car and sofa.
Innovations in biotechnology might be grouped into three different levels: measurement, understanding/modeling, and modification. Our ability to measure biomolecules at continually higher resolution and in greater bandwidth is enabling steady improvements of our measurements of individual organisms (like humans), but also groups of organisms (from the microbiome of a human gut to the complex commensal relationships of organisms in a coral reef or in a patch of forest floor). This amount of data acquisition these days is extremely complex and high-dimensional. Currently, only AI is able to create accurate predictive models and, thus, an efficient form of understanding the data. This, however, requires considerable advances in data storage and analytics. The third element is in the increasingly advanced and precise toolkit being developed for editing biology down to a single molecule. Tools like the Cas family of CRISPR associated proteins are enabling very specific, rapid editing of DNA, the blueprint for most of what we consider living things.
George Church has likened studying the diversity and complexity of biology to an advanced alien civilization leaving all its technology in our backyard for us to analyze. Biology has been able to create the machinery to very efficiently convert wide ranges of energy from one form to another. It is able to harness that energy into vast abilities to transmute forms of matter. This alchemy of biology still produces the vast majority of materials of interest to humanity. We are developing a deep control of the machinery of biology, which is just as crucial as the initial domestication of plants and animals thousands of years ago. Synthetic biology will impact chemicals and materials, energy, and human and animal editing, which will have great economic and societal implications.
Food products and pharmaceuticals are largely the result of biochemical processes. Basic components of our environment, like the oxygen we require to breathe, are the result of biological processes. Changing these systems with tools like a shovel or hammer would be impossible. However, we are gaining the potential to have molecular level control of all living things, giving us powerful new ways of combating issues like food security or climate change.
If human history has been a push to control the world for human good, then we are at the start a major new type of development. This process started with the quest for gaining control over environmental exposure by development of fire, buildings, and clothing; extending to gaining control over supply of food and materials by domestication of plants and animals, efficient agriculture, creating mining and mineral extraction and the industrial revolution; and then recently gaining control over information and data through the development of language and literacy to modern methods of data transmission, storage, and analysis. For the first time, we are gaining control over modifying ourselves directly.
The tools for editing DNA are only the first step in modifying the physiology of an adult living being. In order to do this, we have to target the cells and tissues we want to address specifically. Nevertheless, technological development in that direction is ongoing. We are also developing the ability to modify cells that can be introduced into the body with new genetics and designed molecular biochemistry or to modify and edit embryos prior to their implantation. We are learning to genetically modify a pig embryo to produce human compatible organs for transplantation. We can change our body composition, but we may also decide to edit ourselves from the very start. In terms of editing human embryos directly, we as a society need to determine how we want to use this technology; there are clearly many opportunities for improving health and wellness.
One of the more promising avenues is to capture human knowledge in biomedicine, reconcile inconsistencies automatically, and be able to simulate at the molecular level every pathway and all omics in the body computationally. This could lead to true understanding of normal and deviations from normal, the usual definition of disease. Drugs and their effects on a particular person could be modeled and dosages calculated. The possibilities for human or animal biology management are exciting though impossible to specifically predict if we develop this capability.
Social networking has changed our access and rate in which we access communication and how we collaborate. It has spurred new ideas and influenced the way we think about democracy. It has democratized information in a way that enhances education and new ideas. Twitter, as an example, has changed how we get news. It has influenced the end of regimes, and depending on the point of view, has had a positive or negative effect on politics. Social networking is a powerful tool that has allowed people to have a voice and connect globally. Slack, has brought social networking to industries and enterprises. It enabled more voices to collaborate and be heard, and helped making processes and businesses more efficient.
Like anything great, social networks can have negative and positive implications on the world. They are undoubtedly a powerful tool to both spur innovative ideas and, influence them to get traction. The allow to aggregate opinion, to get feedback for product designs and product reviews, and to have new channels of democracy (and it’s subterfuge). Scientific social networks accelerate communication and collaboration and increase the rate of progress or discussion. There are many ways in which industrial progress is leveraging these social tools. This tool is speeding the pace of change and innovation across key areas like education, health, and government.
The Internet and cryptography, the blockchain was created, which most notably has given rise to a distributed ledger — cryptocurrencies like bitcoin. This will be critical axes of innovation that will enable new businesses and paradigms from taking place, whether it be smart contracts, rethinking workflows, food traceability, medical records, blockchain will be a new way to use technology to rethink complete industries, like the financial system. When Haiti was hit by the hurricane, many or most of the records were lost. Blockchain, as an example, could allow people, businesses, and governments to rethink how they are storing and using their data, such as documents, information, payments.
Old tools that still have impact…
The “older technologies will continue to be axes of innovation that continue to have impact and provide benefit include:.
Computing & cloud computing
Sensors & cameras
**This is a section from “Reinventing Societal Infrastructure with Technology”. To read the previous section, click here.