Five Predictions for the Future of Learning in the Age of AI

Good, thoughtful piece ...

Five Predictions for the Future of Learning in the Age of AI      by Anne Lee Skates  in A16Z

AI, machine & deep learning  education  Education  Generative AI

When OpenAI released its chatbot ChatGPT last year, proponents were quick to announce the death of various writing-related fields, such as screenwriting, computer programming, and music composition. One particular field stood out as a sector that would feel the power of ChatGPT almost immediately: education. With ChatGPT’s technology, students can now easily cheat on papers and college admissions essays, while on the opposite end, teachers can outsource their curriculums to AI—and no one would be the wiser. 

But ChatGPT is hardly the end of education. Just as quickly as students started passing off the chatbot’s work as their own, new programs popped up to detect AI-written work, and teachers, looking to get ahead of their students, started integrating ChatGPT responses into their lesson planning. 

The truth is, if leveraged well, AI has the potential to greatly enhance students’ abilities to think critically and expand their soft skills. And for skeptics who are worried kids will stop learning basic skills, avoid practicing, and forget general facts if they can rely on an AI to answer for them, psychologists Edward Deci and Richard Ryan posit in their self-determination theory that humans are intrinsically driven by autonomy, relatedness, and competence—that is, they will continue to learn regardless of any shortcuts thrown their way. The creation of Wikipedia is a great example. We didn’t stop learning history or science just because we could now quickly look up dates and formulas online. Instead, we simply gained an additional resource to help us fact-check and facilitate learning.

Seeing as education is one of AI’s first consumer use cases, and programs like ChatGPT are how millions of kids, teachers, and administrators will be introduced to AI, it is critical that we pay attention to the applications of AI and its implications for our lives. Below, we explore five predictions for AI and the future of learning, knowledge, and education.

1. The one-on-one model goes mainstream 

Getting one-on-one support for services like tutoring, coaching, mentorship, and even therapy was once only available to the well-off. AI will help democratize these services for wider audiences. In fact, Bloom’s 2 sigma problem—which found that students who received one-on-one teaching performed two standard deviations better than children in a traditional classroom—has a solution now. AI can potentially act as a live tutor for anyone, with humans supplementing the AI to provide in-depth knowledge and emotional and behavioral support. Academic tool Numerade, for example, recently released an AI tutor, Ace, that can generate personalized study plans, curating the right content depending on students’ skill levels.   .....   (four more predictions follow) 

Predictive Reverse Aging for Archaeology

New three dimensional models predicts past faces from Skeletal Structure.

Faces from Ancient Egypt Coming Back to Life in Extraordinary Detail

By Newsweek, January 23, 2023

Said the Face Lab's Caroline Wilkinson, "We are pretty confident in our ability to predict face shape from skeletal structure."

Researchers at the U.K.'s Liverpool John Moores University (LJMU) and Egypt's Cairo University (CU) used software and a "reverse aging" process to replicate ancient Egyptian pharaoh Ramesses II's face.

CU's Sahar Saleem used a computed tomography (CT) scanner to produce a three-dimensional model of Ramesses' head and skull, which formed the basis of the facial reconstruction.

LJMU's Caroline Wilkinson said, "We have tested our methods using CT [scans] from living donors and we have evaluated the facial reconstruction using geometric comparison that shows approximately 70% [of the] surface of the facial reconstruction with less than 2 millimeters of error."

Wilkinson said ancient Egyptian mummies also preserve features like ear shape, creases, or hair pattern, which "should increase the level of accuracy [of the reconstruction]."

From Newsweek   

Asteroid Hits Earth Shortly After Discovery

With pictures at link.  This happens fairly often,  see the Minor Planet Center, but here apparently with a specific prediction.

Asteroid hit Canada hours after discovery   

Posted by, Kelly Kizer Whitt, November 19, 2022

Asteroid hit Canada hours after discovery

Astronomers spotted an asteroid just hours before it struck Earth Saturday morning (November 19, 2022), near Lake Erie in Canada. This is not the first time this year astronomers have discovered a rock from space just hours before it hit Earth. But this time, it entered Earth’s atmosphere over a populated area, crossing the skies of Toronto, Canada. So we have video and witnesses who saw, heard and felt the impact.

Ye Quanzhi, a University of Maryland astronomer who studies asteroids, comets and meteors, reported on Twitter at 1:13 a.m. EST on November 19:  Looks like a space rock is going to fall into Lake Erie area in ~2 hours (about 3:30 am EST)  ... ' 

Google Algorithm Helped Direct Financial Aid to Hurricane Ian Victims

 An example of the use of Sat imagery to predict emergency need.

 Google Algorithm Helped Direct Financial Aid to Hurricane Ian Victims

Wired, Chris Stokel-Walker, October 10, 2022

A Google algorithm deployed in partnership with nonprofit GiveDirectly notified nearly 3,500 hurricane victims in Florida of $700 in financial aid via their smartphones. The algorithm used satellite imagery to identify residents in severely damaged neighborhoods who qualified for emergency assistance. Google's Delphi mapping software powers the Florida project, highlighting communities in need after disasters such as hurricanes by overlaying live maps of damage with information on poverty from sources including the U.S. Centers for Disease Control and Prevention. Google's Skai tool supplies storm-damage data by analyzing satellite images from before and after disasters and estimating the severity of damage. ... ' 

Linkedin Leverages Data for Success Prediction

The Linkedin data begs for such studies to be done, should they be? 

LinkedIn ran undisclosed social experiments on 20 million users for years to study job success

By USA Today, September 28, 2022

A new study analyzing the data of over 20 million LinkedIn users over the timespan of five years reveals that our acquaintances may be more helpful in finding a new job than close friends.

Researchers behind the study say the findings will improve job mobility on the platform, but since users were unaware of their data being studied, some may find the lack of transparency concerning.  

Published this month in Science, the study was conducted by researchers from LinkedIn, Harvard Business School and the Massachusetts Institute of Technology between 2015 and 2019. Researchers ran "multiple large-scale randomized experiments" on the platform's "People You May Know" algorithm, which suggests new connections to users. 

In a practice known as A/B testing, the experiments included giving certain users an algorithm that offered different (like close or not-so-close) contact recommendations and then analyzing the new jobs that came out of those two billion new connections....

Privacy advocates said some of the 20 million LinkedIn users may not be happy that their data was used without consent. .... 

Machine Learning and Firefighting Tech

Contributed to firefighting tech, here a novel application:

 ML-Based Solution Could Help Firefighters Circumvent Deadly Backdrafts

U.S. National Institute of Standards and Technology

October 17, 2022

Scientists at the U.S. National Institute of Standards and Technology have formulated a machine learning-based model to predict potentially deadly backdrafts. The researchers based the model on data from hundreds of laboratory-engineered backdrafts, with the hope firefighters will deploy it to avoid or adjust to hazardous circumstances. The team initially fed the model data on gas levels, fuel richness, and temperature measured at one location in the lab chamber before a door was opened, to calculate the odds of a backdraft occurring. The model predicted backdrafts correctly in 70.8% of the experiments; adding measurements at a second location boosted its accuracy to 82.4%.  .... '

Earthquakes Predictable?

Examined this, even tried some existing test systems. 

Earthquakes Seem to Come in a More Predictable Pattern Than We Thought  By New Scientist, August 30, 2022

A machine learning model developed by researchers at the University of California, Davis (UC Davis) can analyze decades of earthquake data to predict future earthquakes.

The algorithm can determine the probability of a 6.75-magnitude or greater earthquake occurring over a three year-period in a region of California that covers slightly less than 1,000 square kilometers, using records of earthquake since 1970.

The model determined that earthquake data is not as random as previously believed., The researchers used the number of earthquakes per unit of time and had the model identify patterns in the data.

They also applied statistical techniques used in economics to give recent events greater significance than older ones.

The resulting model provides forecast-like probabilities of earthquakes occurring in a short period of time in the future.

Said UC Davis' John Rundle, "What we've done is show that you can see a regional cycle of activity."

From New Scientist  

AI Scours Brain Data to Spot Mental Illness Patterns

Predictive,  accuracy?  

AI Scours Brain Data to Spot Mental Illness Patterns

By Futurity, July 28, 2022  ....  

The research may lead to early diagnosis of mental health conditions.

Georgia State University researchers constructed a computer program that can sift through massive volumes of brain imaging data and identify new patterns associated with mental illness.

The researchers trained the artificial intelligence model on functional magnetic resonance imaging brain scans of people with Alzheimer's disease, schizophrenia, and autism spectrum disorder, and scans of individuals without known clinical disorders. Georgia State's Vince Calhoun said the model yielded new patterns that the researchers could definitively link to each of the three disorders.

"Our goal is to bridge big worlds and big datasets with small worlds and disease-specific datasets and move towards markers relevant for clinical decisions," said Georgia State's Md Mahfuzur Rahman.

From Futurity

View Full Article   

Predicting Crime

 Where have I seen this before? How quickly will it be knocked down? 

AI Algorithm Predicts Future Crimes One Week in Advance With 90% Accuracy

TOPICS:Artificial IntelligenceCrimeMachine LearningPopularUniversity Of Chicago

By MATT WOOD, UNIVERSITY OF CHICAGO JULY 2, 2022

A new algorithm forecasts crime by learning patterns in time and geographic locations from public data on violent and property crimes. It can predict future crimes one week in advance with about 90% accuracy.

A new computer model uses publicly available data to predict crime accurately in eight cities in the U.S., while revealing increased police response in wealthy neighborhoods at the expense of less advantaged areas.

Advances in artificial intelligence and machine learning have sparked interest from governments that would like to use these tools for predictive policing to deter crime. However, early efforts at crime prediction have been controversial, because they do not account for systemic biases in police enforcement and its complex relationship with crime and society.

University of Chicago data and social scientists have developed a new algorithm that forecasts crime by learning patterns in time and geographic locations from public data on violent and property crimes. It has demonstrated success at predicting future crimes one week in advance with approximately 90% accuracy.   .... '   

Algorithms with Prediction

Useful thoughts, technical.   Considering the worst case. 

Algorithms with Predictions  By Michael Mitzenmacher, Sergei Vassilvitskii

Communications of the ACM, July 2022, Vol. 65 No. 7, Pages 33-35   10.1145/3528087

The theoretical study of algorithms and data structures has been bolstered by worst-case analysis, where we prove bounds on the running time, space, approximation ratio, competitive ratio, or other measure that holds even in the worst case. Worst-case analysis has proven invaluable for understanding aspects of both the complexity and practicality of algorithms, providing useful features like the ability to use algorithms as building blocks and subroutines with a clear picture of the worst-case performance. More and more, however, the limitations of worst-case analysis become apparent and create new challenges. In practice, we often do not face worst-case scenarios, and the question arises of how we can tune our algorithms to work even better on the kinds of instances we are likely to see, while ideally keeping a rigorous formal framework similar to what we have developed through worst-case analysis.

A key issue is how we can define the subset of "instances we are likely to see." Here we look at a recent trend in research that draws on machine learning to answer this question. Machine learning is fundamentally about generalizing and predicting from small sets of examples, and so we model additional information about our algorithm's input as a "prediction" about our problem instance to guide and hopefully improve our algorithm. Of course, while ML performance has made tremendous strides in a short amount of time, ML predictions can be error-prone, with unexpected results, so we must take care in how much our algorithms trust their predictors. Also, while we suggest ML-based predictors, predictions really can come from anywhere, and simple predictors may not need sophisticated machine learning techniques. For example, just as yesterday's weather may be a good predictor of today's weather, if we are given a sequence of similar problems to solve, the solution from the last instance may be a good guide for the next.

What we want, then, is merely the best of both worlds. We seek algorithms augmented with predictions that are:

Consistent: when the predictions are good, they are near-optimal on a per instance basis;

Robust: when the predictions are bad, they are near-optimal on a worst-case basis;

Smooth: the algorithm interpolates gracefully between the robust and consistent settings; and

Learnable: we can learn whatever we are trying to predict with sufficiently few examples.

Our goal is a new approach that goes beyond worst-case analysis.14 We identify the part of the problem space that a deployed algorithm is seeing and automatically tune its performance accordingly.

As a natural starting example, let us consider binary search with the addition of predictions. When looking for an element in a large sorted array, classical binary search compares the target with the middle element and then re-curses on the appropriate half (see Figure 1). Consider, however, how we find a book in a bookstore or library. If we are looking for a novel by Isaac Asimov, we start searching near the beginning of the shelf, and then look around, iteratively doubling our search radius if our initial guess was far off (see Figure 2). We can make this precise to show that there is an algorithm with running time logarithmic in the error of our initial guess (measured by how far off we are from the correct location), as opposed to being logarithmic in the number of elements in the array, which is the standard result for binary search. Since the error is no larger than the size of the array, we obtain an algorithm that is consistent (small errors allow us to find the element in constant time) and robust (large errors recover the classical O(log n) result, albeit with a larger constant factor).  ... ' 

Predicting Impact of Earthquakes

Not predict occurrence, but predict impact.  Quite useful in preparation.  Addressing during an event.  Recall we had looked at  feasibility of prediction.  More training samples needed.  

Neural Network Model Helps Predict Site-Specific Impacts of Earthquakes

Hiroshima University (Japan),  April 15, 2022

A neural network model developed by researchers at Japan's Hiroshima University can calculate how soil layers can amplify the seismic waves of large earthquakes. The researchers analyzed microtremor data from 105 sites in the Chugoku district from 2012 to 2020 using a generalized spectral inversion method. Data from each site was split into a training set to educate the neural network, a validation set applied to iterative model optimization, and a test set to assess the model's performance. The model performed well on the test dataset, although, said Hiroshima's Hiroyuki Miura, ”the number of training samples analyzed in this study sites is still limited.” Miura said more training samples needed to be considered “before assuming that the neural network model applies nationwide or globally.”

Full Article

Ray Predicts the Technology Future

Good thoughts, below just the intro, much more at the link:

Ray Kurzweil Predicts Three Technologies Will Define Our Future  By Sveta McShane and Jason Dorrier -Apr 19, 2016

This is the last in a four-part series looking at the big ideas in Ray Kurzweil’s book The Singularity Is Near. ​Be sure to read the other articles:

Will the End of Moore’s Law Halt Computing’s Exponential Rise?

Technology Feels Like It’s Accelerating — Because It Actually Is

How to Think Exponentially and Better Predict the Future

Over the last several decades, the digital revolution has changed nearly every aspect of our lives.

The pace of progress in computers has been accelerating, and today, computers and networks are in nearly every industry and home across the world.  Many observers first noticed this acceleration with the advent of modern microchips, but as Ray Kurzweil wrote in his book The Singularity Is Near, we can find a number of eerily similar trends in other areas too.

According to Kurzweil’s law of accelerating returns, technological progress is moving ahead at an exponential rate, especially in information technologies.

This means today’s best tools will help us build even better tools tomorrow, fueling this acceleration.

But our brains tend to anticipate the future linearly instead of exponentially. So, the coming years will bring more powerful technologies sooner than we imagine.   As the pace continues to accelerate, what surprising and powerful changes are in store? This post will explore three technological areas Kurzweil believes are poised to  change our world the most this century.

[Read more about exponential progress in computing, Kurzweil’s law of accelerating returns, and how to think exponentially and better predict the future.]

Genetics, Nanotechnology, and Robotics   ......   (Read the whole thing and articles linked to)   ....... 

Machine Intelligence Builds Soft Machines

 Softer is naturally more adaptable in a context. 

Machine Intelligence Builds Soft Machines

By University of Maryland A. James Clark School of Engineering, February 1, 2022

A machine learning framework developed by researchers at the University of Maryland (UMD) aims to accelerate the design of soft machines.

The framework can be used to build a prediction model that perform the two-way design task, predicting sensor performance based on a fabrication recipe and recommending feasible fabrication recipes for adequate strain sensors.

UMD's Po-Yen Chen said, "What we've essentially created is a high-accuracy prediction software – based on a machine learning framework – capable of designing a wide range of strain sensors that can be integrated into diverse soft machines." ...

Explained the Universary of Maryland's Po-Yen Chen, “What we’ve essentially created is a high-accuracy prediction software–based on a machine learning framework–capable of designing a wide range of strain sensors that can be integrated

From University of Maryland A. James Clark School of Engineering

View Full Article    

Predictions from O'Reilly Group

 From  OReilly AI: emerging tech Radar  

1. What’s ahead for AI, VR, NFTs, and more?

Mike Loukides makes some Predictions:

Thoughtful looks.  

Predicting Future of Covid

 Interesting predictive approach,  but don't understand its basis.

Predicting the Future of COVID

Boston College, January 6, 2022

A new analytical tool developed by a research team led by biologists at Boston College (BC) uses quantum mechanical modeling to predict future mutations of SARS-CoV-2. BC's Babak Momeni said, "We computationally predict what mutations allow better binding to host receptors and better evasion of antibodies." The goal is to prepare for future COVID variants of concern. Said Momeni, "We use a fully quantum mechanical model to theoretically assess how different mutations in the spike [protein of the coronavirus] can contribute to its increased, or decreased, binding strength to human ACE2." The study also found that factors other than binding may be involved in determining how a variant evolves.

Model Predicts COVID Outbreak

No idea how well this actually works in practice, but interesting.   Could be useful for future needs.  Examining the stated methods.

Model Predicts COVID-19 Outbreak Two Weeks Ahead of Time

Florida Atlantic University, Gisele Galoustian, August 6, 2021  via CACM

Researchers at Florida Atlantic University (FAU) and Lexis-Nexis Risk Solutions have crafted a long short-term memory (LSTM) deep-learning model that could potentially predict a COVID-19 outbreak two weeks in advance. The team blended driving-mobility data compiled by the Apple Maps application, COVID-19 statistics, and county-level demographics from 531 U.S. counties. Researchers trained the model to record the impact of government responses and age on COVID-19 cases and viral spread, respectively. Results indicated that average daily cases declined as the retiree percentage expanded and increased as the youth percentage grew. FAU's Stella Batalama said the research "has significant applications for effective management of the pandemic and future outbreaks, which has the potential to save lives and keep our economies thriving."  .... 

Flawed Algorithms

Always looking at data and context involved when algorithms fail.  Like any kind of predictive approach, its rarely perfect.   The same for classing analytics and for machine learning based methods.  And for that matter for human predictive methods as well.   And that can change as data and context changes over time.    So these kinds of test examples are useful.

Algorithm That Predicts Deadly Infections Is Often Flawed   By Wired

A study using data from nearly 30,000 patients in University of Michigan hospitals suggests Epic Systems' early warning system for sepsis infections performs poorly.

An algorithm designed by U.S. electronic health record provider Epic Systems to forecast sepsis infections is significantly lacking in accuracy, according to an analysis of data on about 30,000 patients in University of Michigan (U-M) hospitals.

U-M researchers said the program overlooked two-thirds of the approximately 2,500 sepsis cases in the data, rarely detected cases missed by medical staff, and was prone to false alarms.

The researchers said Epic tells customers its sepsis alert system can correctly differentiate two patients with and without sepsis with at least 76% accuracy, but they determined it was only 63% accurate.

U-M's Karandeep Singh said the study highlights wider shortcomings with proprietary algorithms increasingly used in healthcare, noting that the lack of published science on these models is "shocking."

From Wired

Sample Patterns and Predictions

 Like to see examples of this type to show what can be done with emerging tech. 

Open Source AI Can Predict Electrical Outages from Storms with 81% Accuracy  by Anthony Alford in Infoq

Development Group Manager at Genesys Cloud Services

A team of scientists from Aalto University and the Finnish Meteorological Institute have developed an open-source AI model for predicting electrical outages caused by storm damage. The model can predict storm location within 15km and classifies the amount of transformer damage with 81% accuracy, allowing power companies to prepare for outages and repair them more quickly.

The work was described in an article published in the European Geosciences Union's (EGU) Natural Hazards and Earth System Sciences (NHESS) journal. The model predicts damage to power transformers from large low-pressure storms up to 10 days in advance, categorizing the results as either no damage, low damage (less than 140 transformers damaged), or high (more than 140). The predictions are based on a support-vector classifier, which achieves 81% precision and 61% recall. Using this model, power companies can prepare materials and repair crews, restoring power to customers more quickly.

Because Finland is a heavily forested country, its overhead power lines are often damaged by falling trees, especially during strong extratropical storms; on average, about 46% of the country's power outages were caused by these storms. Because the power suppliers are required by law to provide their customers with financial compensation for prolonged outages, the companies maintain a large workforce for rapid repair. While several researchers have applied AI techniques to predict power outages from hurricanes, as well as damage to trees (not surprisingly, random forests work quite well for this task), there has been little work specifically on power outages due to extratropical storms. ... '

Predicting COVID Spikes

Hmm, caution with such causal seeming predictions.   But interesting.    

Google Search Can Help Predict Covid-19 Spikes   The Jerusalem Post, Zachary Keyser, February 8, 2021

Researchers from the U.K.'s University College of London (UCL) and Israel's Bar-Ilan University have found that spikes in Covid-19 cases could be predicted an average of 17 days in advance using online search data. The researchers developed an analytical model that compares Covid-19 symptom-related Google searches against spikes in Covid-19 cases. Symptoms identified by the U.K.'s National Health Service and Public Health England are used by the model and weighted according to how frequently they occurred in confirmed Covid-19 cases. UCL's Vasileios Lampos said, "We have shown that our approach works on different countries irrespective of cultural, socioeconomic, and climate differences. Our analysis was also among the first to find an association between Covid-19 incidence and searches about the symptoms of loss of sense of smell and skin rash."  ... ' 

On Predictability of Elections

Well done with useful links

On Predictability of Elections

By Irving Wladawsky-Berger

A collection of observations, news and resources on the changing nature of innovation, technology, leadership, and other subjects.

Are There Limits to the Predictability of Elections?