The age of AI 2.0 has begun
This shift that we’re seeing with AI 2.0 began with a small team at Google and their 2017 paper “Attention is All You Need”.
Search engines were the first to be transformed by AI 2.0.
AGI is not required to make colossal impacts on the global economy.
We cannot raise each new generation in fear of the future. There is unbelievable promise ahead and we must not lose sight of the ultimate goal.
The Real Story of AI
People often rely on three sources to learn about AI: science fiction, news and influential people.
How can we learn to stop worrying and embrace the future with imagination
According to Amara’s law, “We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.
The Golden Elephant
The Golden Elephant introduces basic of AI and deep learning, offering a sense of its strengths and weaknesses.
Deep learning is a recent AI breakthrough. Among the many subfields of AI, machine learning is the field that has produced the most successful applications, and within machine learning, the biggest advance is deep learning – so much so that the terms AI, machine learning, and deep learning are sometimes used interchangeably.
Inspired by the tangled webs of neurons in our brain, deep learning constructs software layers of artificial neural networks with input and output layers.
The first academic paper describing deep learning dates all the way back to 1967.
Deep learning requires much more data than humans, but once trained on big data, it will outperform humans by far for a given task.
People have a unique ability to draw on experience, abstract concepts and common sense to make decisions. In order for deep learning to function well, the following are required: massive amounts of relevant data, a narrow domain, and a concrete objective function to optimize.
Every powerful technology is a double-edged sword. The core issue of deep learning is simplicity of the objective function, and the danger from single-mindedly optimizing a single-objective function, which can lead to harmful externalities. Two solutions are to teach AI to have more complex objective functions and (as suggested by Stuart Russell) to ensure that objective functions are always beneficial to humans by keeping humans in loop.
Another two issues are fairness and explainability and justification.
Gods Behind the Masks
Truth and morning become light with time. African proverb
Gods Behind the Masks tells a tale of visual deception. This story describes a future in which people can no longer rely on their naked eyes to tell real videos from fake ones.
Among our six senses sight is the most important.
Computer vision includes the following capabilities in increasing complexity:
- Image capturing and processing.
- Object detection and image segmentation.
- Object recognition.
- Object tracking.
- Gesture and movement recognition.
- Scene understanding.
Computer vision can be used in real time in areas ranging from transportation to security.
Our visual cortex uses many neurons corresponding to many restricted subregions (known as receptive fields) within what our eyes see at any given time. These receptive fields identify basic features, such as shapes, lines, colors, or angles. The neocortex store information hierarchically and processes these ‘receptive-fields’ outputs into more complex scene understanding.
Convolutional neural networks (CNN) are specific and improved deep learning architecture designed for computer vision, with different variants for images and videos.
Our future is one where everything digital can be forged, including online video, recorded speech, security camera footage, and courtroom evidence video.
Deeefakes are built on a technology called generative adversarial networks (GAN). A GAN is a pair of adversarial deep learning neural networks. The first network, the forger network, tries to generate something that looks real. The other network, the detective network, compares the forger’s synthesized dog picture with genuine dog pictures, and determines if the forger’s output is real or fake.
Until we have this longer-term solution based on blockchain or equivalent technology, we hope there will be continuously improved technology and tools for detecting deepflakes.
Biometrics is the field of study of using a person’s physical characteristics to verify his or her identity.
AI will take over routine tasks or recognizing and verifying people.
Twin Sparrows
Twin Sparrows explore the future of AI education. It introduces the idea of personal AI companion.
Price is what you pay. Value is what you got. Warren Buffett.
NLP or natural language processing is the ability to for machines to process and understand human languages. It is subbranch of AI.
Self-supervising learning means AI supervises itself and no human labeling is required. This approach is called sequence transduction.
In 2017, researchers at Google invented transformers, a new sequence transduction model that, when trained on huge quantities of text, can exhibit selective memory and attention mechanisms that can selectively remember anything important and relevant in the past.
After Google’s transformers work, a more well-known extension called GPT-3 was released in 2020 by OpenAI.
GPT-3 is weak in causal reasoning, abstract thinking, explanatory statements, common sense, and (intentional) creativity.
Central to human intelligence are the abilities to reason, plan, and create.
I consider the obsession with AGI to be a narcissistic, human tendency to view ourselves as the gold standard.
Teaching consists of lectures, exercises, examinations, and tutoring. All four components require a lot of the teacher’s time.
In this vision of an AI-infused school, there will be plenty for human teachers to do. Teachers will play two important roles: they will be human mentors and connectors for the students. They will also direct and program AI teacher and companion in ways that will best address the students’ needs.
Contactless Love
What has become clear now is that AI will reshape healthcare, from spending the discovery of vaccines and drugs to accelerating the integration of technologies like AI diagnostics into existing care.
Existing healthcare databases and processes will be digitized, including patient records, drug efficacy, medical instruments, wearable devices, clinical trials, quality-of-care surveillance, infectious-disease-spread data, as well as supplies of drugs and vaccines.
Wearable devices continuously monitor heart rate, blood pressure, blood sugar, and an increasing number of vital statistics that can provide warning signs. This tracking will yield huge databases that can help AI correlate these statistics for more-accurate monitoring, early detection, medical treatment, and maintenance.
DNA sequencing produces vital digital information. Digital polymerase-chain reaction (dPCR) can accurately detect pathogens and gene mutations. CRISPR is a breakthrough technology for gene editing. Drug and vaccine discovery are going digital and starting to be integrated with AI. Early AI projects like IBM Watson failed because they still didn’t have enough real patient-treatment-and-outcome data.
Drug discovery is the process of finding treatment molecule in the following four steps:
- Use mRNA sequence to derive the pathogen’s protein sequence.
- Find the 3D structure of the pathogen’s protein sequence (protein folding).
- identify the target on that 3D structure.
- Generate likely treatment molecules and then select the best preclinical candidates from them.
DeepMind developed AlphaFold2 for step 2.
Scientists can also work with AI symbiotically to invent new compounds.
Precision medicine is a term that refers to tailoring an individualized treatment for a given patient, rather than producing blockbuster, one-size-fits-all-type drugs.
Diagnostic and even surgeries can become AI and automation domains. But general human-level sight, touch, manipulation, movement, and coordination are too difficult to perfect in twenty years.
Most expensive technologies hit maturity when industries can see high value in their applications.
Digitization of workflow makes it easier than ever to recognize, out-source, or automate work.
My Haunting Idol
XR is a term encompassing three types of technologies. VR, AR and MR. Virtual reality renders a fully synthesized virtual environment in which the user is immersed. Augmented reality is based on the world that the user is physically in, capturing it through a camera, and then superimposing another level on top of it. Mixed reality mixes virtual and real worlds into a hybrid world.
In the past few decades, significant improvements have been made on networking, resolution, refresh rate, and latency.
Beyond XR glasses, I believe XR contact lenses may be the first XR technology to achieve the milestone of mass acceptance.
Scent emitters, taste simulators, and haptic gloves that stimulate touch are all emerging to cover our six-senses – actually five, since we don’t expect an ESP simulator.
Training will be a major XR application area.
Content creation in an XR environment is similar to creating a complex 3D game.
XR will need to overcome issues of nausea, which is mostly caused by the latency of the experience.
The most natural way to see a virtual environment would be with the naked eye, as a holograph. It naked-eye XR is the most natural output, then the most natural input must be brain-computer interface (BCI).
XR will make us rethink what living means. Humans have been seeking immortality for thousands of years. With these technologies, we can ponder the possibility of a digital immortality.
The Holy Driver
Autonomous vehicles (AV) are actually one of the Holy Grails of AI. Driving is very complex task with many subtasks and inputs, as well as the potential for uncertain environments and unlikely events.
AV is a computer-controlled vehicle that drives itself.
AV will mature one step at a time. These steps are classified by the Society of Autonomous Engineers as Level 0 to Level 5:
- L0 – no automation.
- L1 – hands on.
- L2 – hands off.
- L3 – eyes off.
- L4 – mind off.
- L5 – steering wheel optional.
Again the biggest obstacle is the amount of data needed to achieve L5. Driving data.
A confluence of improved L5 technologies, augmented roads, and 6G communications connected using AR should be experimentally deployed around 2030. L5 could be deployed by 2040.
Governments will approve pervasive AV only when it is safer than people.
Quantum Genocide
We don’t need Ais to destroy us; we have our own arrogance.
In the quantum world, causality worked counter to human intuition. Cause and effect were intertwined.
Technology is inherently neutral – it’s people who use it for purposes both good and evil. Autonomous weapons, like all technology, will also be used for good or evil.
Quantum computing has an 80-percent chance of working by 2041. And if that happens, it may have a great impact on humanity than AI. It is truly general-purpose technology.
Quantum computer uses qubits instead of bits. They are typically subatomic particles such as electrons or photons. They include unusual properties that give them super-processing capabilities. The first such property is superproposition, or the capability for each qubit to be in multiple states at any given time.
Even slight vibrations, electrical interference, temperature changes, or magnetic waves can cause superposition to decay or even disappear.
Th IBM researchers acknowledge that control of errors caused by decoherence will get much worse the more qubits are added.
Most experts believe it will take ten to thirty years to get a useful QC. One world changing application will be drug discovery.
Programming a QC involves giving it all potential solutions represented with qubits, and then scoring each potential solution in parallel. Then, the QC will attempt to find the best answer in very little time. This could potentially revolutionize machine learning and solve problems that were viewed as impossible before.
What could be done to upgrade our cryptography? Quantum-resistant algorithms exist. But they are very expensive computationally, so they are not being considered right now by most commercial and Bitcoin entities. Perhaps only after inevitable quantum bitcoin heist happens will people wake up to revamp the algorithms.
Autonomous weaponry is the third revolution in warfare, following gunpowder and nuclear arms. Their benefits are to save soldiers lives, they can be used to help soldiers target only combatants and avoid inadvertently killing friendly armed forces, children and civilians. One issue is having a clear ling of accountability.
The Job Savior
AI can perform many tasks better than people can at essentially zero-cost. When it comes to work, is AI ultimately a blessing or a curse?
Will we ever reach a point where human work is so scarce that future work will turn into some kind of simulated game.
AI’s main advantage over humans lies in its ability to detect incredibly subtle patterns within large quantities of data.
Optimists argue that productivity gains from new technologies almost always produce economic benefits – that more growth and more prosperity always mean more jobs. But AI and automation differ from other technologies.
A growing pool of unemployed workers will compete for an ever-shrinking number of jobs, driving down wages. Wealth inequality will go from bad to worse, as AI algorithms destroy millions of human jobs, while at the same time turning the tech titans who harness these new technologies into billionaires in record time.
Even more problematic than the loss of jobs will be the loss of meaning.
The enormous challenges of AI and job displacement have breathed new life into and old idea called universal basic income (UBI).
In order to help guide people through AI displacement, we need to first understand what kinds of capabilities and tasks AI cannot do. Where I see AI failing short:
Creativity. AI cannot create, conceptualize, or plan strategically.
Empathy. AI cannot feel or interact with feelings like empathy and compassion.
Dexterity. AI and robotics cannot accomplish complex physical work that requires dexterity or precise hand-eye coordination.
I propose the 3R – relearn, recalibrate and renaissance – as part of a gargantuan effort to deal with the central issue of our time, the AI economic revolution.
In addition to relearning skills, we need to recalibrate what today’s jobs look like with the help of AI moving toward a human-AI symbiosis. A deeper interdependence between AI optimizations and human touch will reinvent many jobs and create new ones.
Isle of Happiness
Back in the 1970s, American psyhologists Philip Brickman performed an experiment. He brought a group of lottery winners together with a group of people paralyzed by accidents. The lottery winners were no happier than control group. While accident victims were less happy at the time of evaluation, their hopes for future happiness were no different from the controls.
The brain measured its level of sensory stimulation against the level of stimulation it is already used to.
When everything is possible, nothing is interesting.
Can AI optimize our happiness? This is an incredibly complex and tough problem. What is happiness exactly and how to measure it.
Psychologists Michael Eysenck introduced the term hedonic tread-mill to describe our tendency to always readjust to a fixed level of happiness, despite monetary and possession gains or losses.
By 2041 AI ability to read human emotions should be quite advanced, well beyond human capabilities, and there should be prototypes that try to improve human higher-level happiness.
The fundamental issue is that when the interest of the AI owner diverges the interest of the AI users, the users lose.
Over time, there may also be technology solutions that will allow us to have our cake (powerful AI) and eat it too (with data protection even from the AI owner).
Dreaming of Plenitude
We humans have long fantasized about the day when we no longer have to work and everything is free.
While AI and other technologies are bringing about the fourth industrial revolution, a clear energy revolution is under way. As the cost of energy plummets, it will also bring down the cost of water, raw materials, manufacturing, computations, logistics, and anything that has a major energy component.
Synthetic biology will revolutionize the food industry. It will revolutionize many industries by making them more sustainable while dramatically lowering the overall cost.
Robots and AI will take over the manufacturing, delivery, design, and marketing of most goods. We will have AI assistants, AI teachers, AI doctors, AI entertainment, AV and AI service robots.
Post-scarcity describes a world where nothing is scarce, and everything is free. For millennia human economic systems have evolved under one fundamental premise – scarcity. When there is no scarcity, then all mechanisms such as selling, buying, and exchanging will no longer be needed.
What if we don’t need money. The Basic Life Card (BLC), you can think of it as essentially universal basic service. BLC gives holders credits that can be exchanged for services that fulfill basic needs as well as allow for a comfortable life.
In a world transitioning into plentitude, we cannot simply assume everyone falls into useless class, nor that everyone will strive for self-actualization.
Abraham Maslow has said: “One’s only failure is failing to live up to one’s own possibilities.”[1]
I have painter a grand road map to plentitude. But this road is full of obstacles and even death traps. Reaching plentitude requires nothing short of a total financial overhaul. Corporations will refuse to accept the end of scarcity. The transition to plentitude requires a successful societal overhaul.
[1] In the book on page 434
