Wearables and the Internet of Things (IoT) may give the impression that it’s all about the sensors, hardware, communication middleware, network and data but the real value (and company valuation) is in insights. In this article, we explore artificial intelligence (AI) and machine learning that are becoming indispensable tools for insights, views on AI, and a practical playbook on how to make AI part of your organization’s core, defensible strategy.
Before we proceed, let’s first define the terms. Otherwise, we risk commingling marketing terms like “Big Data” and not addressing the actual fields.
Artificial Intelligence: The field of artificial intelligence is the study and design of intelligent agents able to perform tasks that require human intelligence, such as visual perception, speech recognition, and decision-making. In order to pass the Turing test, intelligence must be able to reason, represent knowledge, plan, learn, communicate in natural language and integrate all these skills towards a common goal.
Machine Learning: The subfield of machine learning grew out of the effort of building artificial intelligence. Under the “learning” trait of AI, machine learning is the subfield that learns and adapts automatically through experience. It focuses on prediction, based on known properties learned from the training data. The origin of machine learning can be traced back to the development of neural network model and later to the decision tree method. Supervised and unsupervised learning algorithms are used to predict the outcome based on the data.
Data Mining: The field of data mining grew out of Knowledge Discovery in Databases (KDD), where data mining represents the analysis step of the KDD process. Data mining focuses on the discovery of previously unknown properties in the data. It originated from research on efficient algorithm for mining association rules in large databases, which then spurred other research on discovering patterns and more efficient mining algorithms. Machine learning and data mining overlap in many ways. Data mining uses many machine learning methods, but often with a slightly different goal in mind. The difference between machine learning and data mining is that in machine learning, performance is usually evaluated with respect to the ability to reproduce known knowledge while in KDD the key task is the discovery of previously unknown knowledge. Unlike machine learning, in KDD, supervised methods cannot be used due to the unavailability of training data.
Fear of AI
Though perhaps not explicitly stated, you will find that some at your work hold sci-fi views of AI that could hamper proactive exploration of AI and machine learning within your organization. AI, for some, bring images of HAL 9000 from A Space Odyssey or more recent films such as Her and The Machine.
Many futurists have speculated about the future of artificial intelligence that could rival or exceed human intelligence. One of those futurists is Ray Kurzweil, a recipient of the prestigious National Medal of Technology and Innovation honor.
In The Singularity is Near, Kurzweil elaborates on the singularity hypothesis. Kurzweil predicts that accelerating progress in technologies will cause a runaway effect wherein artificial intelligence will exceed human intellectual capacity and control, thus radically changing or even ending civilization in an event called the singularity. During this period, he predicts “human life will be irreversibly transformed” and humans will transcend the “limitations of our biological bodies and brain”.
Kurzweil claims that machines will pass the Turing AI test by 2029, and that around 2045, “the pace of change will be so astonishingly quick that we won’t be able to keep up, unless we enhance our own intelligence by merging with the intelligent machines we are creating”. He further claims that humans will be a hybrid of biological and non-biological intelligence that becomes increasingly dominated by its non-biological component. Kurzweil envisions nanobots inside our bodies that fight against infections and cancer, replace organs, and improve memory and cognitive abilities. Eventually our bodies will contain so much augmentation that we will be able to alter our “physical manifestation at will”.
The artificial general intelligence (AGI) or strong AI community, though varying widely in timeframe to reach singularity, are in consensus that it’s plausible, with most mainstream AI researchers doubting that progress will be rapid.
In regards to feasibility, Microsoft co-founder Paul Allen believes that such intelligence is unlikely in this century because it would require “unforeseeable and fundamentally unpredictable breakthroughs” and a “scientifically deep understanding of cognition”. Roboticist Alan Winfield claims the gap between modern computing and human-level artificial intelligence is “as wide as the gulf as that between current space flight and practical faster than light space flight”. Neuroscientist David J. Linden writes that, “Kurzweil is conflating biological data collection with biological insight”. He feels that data collection might be growing exponentially, but insight is increasing only linearly.
AGI raises difficult ethical questions and risks to civilization and humans. Political scientist Charles T. Rubin believes that AI can be neither designed nor guaranteed to be benevolent. He argues that “any sufficiently advanced benevolence may be indistinguishable from malevolence.” Humans should not assume machines or robots would treat us favorably, because there is no reason to believe that they would be sympathetic to our system of morality, which has evolved along with our particular biology. Hyper-intelligent AI may not necessarily decide to support the continued existence of mankind, and would be extremely difficult to stop.
Stephen Hawking believes that AI has a lot of promising things to offer for future, but not without possible dire consequences. He says that “success in creating AI would be the biggest event in human history,” and “unfortunately, it might also be the last”.
Even Elon Musk, the Tesla and SpaceX billionaire, tweeted recently that “We need to be super careful with AI. Potentially more dangerous than nukes. Hope we’re not just the biological boot loader for digital superintelligence. Unfortunately, that is increasingly probable.”
What’s Reality: Strong AI vs. Weak AI
Before running for the hills, let’s pause for a reality check. It’s important that we don’t confuse AGI with subcomponents AI applications.
AGI or strong AI is defined as the intelligence of a machine that could successfully perform any intellectual task that a human being can. For practical wearables and IoT implementations, we are working with weak AI, which studies a specific problem solving or reasoning tasks and does not attempt to simulate the full range of human cognitive abilities. There are AI applications that exhibit capabilities such as visual perception, speech recognition, and decision-making, but none at human levels. The chasm from stitching subsystems, bottom-up to fully intelligent machines is galaxies wide.
Practical Applications of AI and Machine Learning
From Apple’s Siri, Google Voice Search, Google Brain, Google Translate, Xbox, Netflix, IBM’s Watson, autonomous cars, email spam filtering to credit card fraud detection, AI has already infiltrated into nearly aspect of our daily lives… and our dependence on it is only growing.
So how can AI and machine learning be applied to wearables and the Internet of Things? Let’s walk through a few examples.
Medical Diagnosis & Treatment: Lumiata
Lumiata’s machine graph is based on multi-dimensional probability distribution that contains 160 million data points from textbooks, journal articles, and public data sets to replicate and scale doctor’s knowledge for use by nurses to diagnose and treat illnesses. Add patient-specific data, effects of time and location to Lumiata’s massive data set, the machine learning system is able to generate a clinical model of a patient. In the future, clinically approved wearables can interface with Lumiata’s API to provide a constant feed of a patient’s physiological data, time and place for proactive monitoring and event triggers.
Preventive Health: Google X Nanoparticles
Google X recently announced that they are researching the use of nanoparticles. Released into the bloodstream via a swallowed pill, nanoparticles can proactively detect and diagnose diseases, cancers, impending heart attacks or strokes based on changes to the person’s biochemistry, at the molecular and cellular level. The patient can then use a wearable wristband to view readings of the nanoparticles. Google aims to use nanoparticles to clump around a cancerous cell or identify fatty plaques in the lining of blood vessels about to break free, potentially causing a heart attack or stroke.
Machine learning can be applied to learn to diagnose diseases and changes in biochemistry in the bloodstream through the movement of nanoparticles, as unattached nanoparticles would move differently in a magnetic field from those clumped around, for instance, a cancer cell.
Preventive Health: Entopsis
Another early molecular diagnosis startup is Entopsis, a medical diagnosis platform that can screen for medical conditions using nano-engineering and machine learning. Entopsis is able to detect patterns and biomarkers that point to specific conditions by analyzing the protein composition of biofluids.
After incubating a biofluid sample using their Nanoscale Unbiased TExtured Capture (NUTeC) process to capture molecules, Entopsis applies its signature analysis based on machine learning algorithms to analyze the molecular signature on a NUTeC glass. The scanned signatures are then uploaded to the cloud to run signature comparison against others in the database to find similar profiles.
Entopsis desires to give consumers direct access to NUTeC dishes to collect biofluids and send it in for molecular analysis. Could NUTeC glasses be equipped with IoT sensors to optically scan and transmit signature data to the cloud remotely?
Body Movements: Atlas Wearables
Atlas Wearables is a fitness band plus intelligence platform, powered by the Motion Genome Project database of movements. Aside from measuring heart rate and calculating the calories burned, Atlas’ claim to fame is their machine learning algorithms that automatically classifies your exercise routine in 3D vector, being able to decipher the difference between push-ups and triangle pushups. Exercise detection is just the beginning. In speaking with co-founder Peter Li, the startup’s aspiration is to bring “intelligence into body language and movements”. Machine learning algorithms and datasets can be extended to understand how you are walking, sitting, moving or interacting with others, that can give clues about your mood, physical reaction, energy level and even context.
Emotion Measurement: BrandEmotionsTM
BrandEmotions solves the problem of quantifying consumers’ emotions. Sentiment analysis and surveys provide positive/ negative or stacked ranked results but brands still can’t classify nor measure the emotions of their consumers. BrandEmotions enables brands to measure how consumers feel about their brand experience, from retail, live events, movies, hotels, cruises, amusement parks to advertising. BrandEmotions, a product of Amyx+, visualizes the emotional reaction of participants to brand engagement, allowing brands to optimize the brand experience, increase brand loyalty and accurately target products and services at the right time. BrandEmotions’s emotion sensing, machine learning platform measures physiological data captured through a broad range of wearable devices and Internet of Things connected devices to translate data into emotional classifications and intensity using its proprietary EmotionIQ methodology.
Medication Compliance: Vitality
Vitality, acquired by NANTHEALTH, addresses the billion dollar medication adherence market with an Internet-connected pill cap called the GlowCap that blinks and sounds when it’s time to take medication. Vitality’s compliance-enhancing system tries to change patient behavior through a combination of feedback, reminders, education, and incentives to improve patient’s medication adherence. The GlowCap provides real-time data to caregivers, such as when medication has been removed or a dose skipped. Patients can reorder by simply pushing a button on the cap.
Speaking on a panel at the Wearables + Things 2014 conference, Dr. Yan Chow, the former Medical Director of Kaiser Permanente’s IT Innovation group, stated that drug compliance is a complex, multi-layered issue. “It’s not simply that patients forget to take their medication. Some patients disregard advice from doctors and family members for irrational reasons,” asserts Dr. Chow. Blinking pill bottles may not be enough. IoT startups need to understand at a deeper level how to overcome stubborn resistance to drug compliance, whether that’s through reminders, education, gamification, or something else.
Farming: ENORASIS and SCRI-MINDS
The third year of record drought in the farmland of the San Joaquin Valley in California is forcing growers to rely almost entirely on well water and farmers are worried that groundwater will run out. California produces more than 90 percent of the broccoli grown in the United States and just about every other fruits and vegetables. No other state can match California’s output per acre. Hence, the drought condition in California affects national agricultural output and commodity prices that we personally feel at the grocery store. So how can IoT and machine learning help?
The ENORASIS project uses a network of sensors in the fields to determine how much water to give their crops through subsurface drip and micro-irrigation systems. The sensors collect environmental and soil conditions such as soil humidity, temperature, sunshine, wind speed, rainfall and the water valves to quantify water already added to the fields. ENORASIS combines weather forecast and sensor data about the farm’s crops to create a detailed daily irrigation plan that best suits the needs of each crop. The model also includes crop yield data and energy and water costs, helping farmers decide whether extra irrigation will increase yields profitably or cause a loss.
Another project is the SCRI-MINDS project, a research collective comprised of scholars from the University of Maryland Center for Environmental Science, Carnegie Mellon University Robotics Institute, Colorado State University, Cornell University and the University of Georgia that applies wireless sensor networks and environmental modeling to conserve irrigation water for nurseries and greenhouses.
Over time, researchers can amass a rich dataset of geographic-specific irrigation, weather, environment, soil, and crop yield data by plant and tree varieties that machine learning algorithms can use to determine the best crops to plant for the next farming cycle.
Petronics recently launched and funded Mousr on Kickstarter, the first robotic mouse that can see and react to a cat’s movements, bringing to life the Tom & Jerry chase for your cats. Equipped with a 360 degree camera, motion sensing technologies, and Bluetooth, Mousr responds to motion and external forces to escape a cat’s paw. According to co-founder David Jun, next on their product roadmap is “to develop AI algorithms that will help Mousr to outsmart a cat every time”. Petronics introduces robotics into the home in an affordable, gamified way that opens the door for other startups to pursue robotic AI dogs and eventually child-sized AI robots with machine learning-based movements and natural language process capabilities. This is validated by JIBO, the world’s first family robot, pioneered by a social robotics MIT professor.
Other noteable AI-based home robotics include Anki Drive and WowWee. Anki Drive combines virtual car racing games with physical RC cars powered by artificial intelligence. WowWee’s MiP self-balancing robot with GestureSense technology not only responds to hand gestures but is aware of its surroundings.
AI and machine learning techniques are being actively applied in many areas:
- Affective computing (Affectiva)
- Bioinformatics (Classifying biological sequences, clustering biological entities)
- Brain-machine interfaces (Emotiv)
- Brain neurons connections (EyeWire -crowdsourced)
- Cheminformatics (Thomson Reuters Systems Biology)
- Classifying DNA sequences
- Computational advertising (Microsoft, Yahoo)
- Computational finance (Algorithmic trading, quantitative investing, high-frequency trading)
- Computer vision and object detection (Dropbox/ KBVT, Occipital, Google+ photo)
- Facial recognition (Emotient)
- Fraud detection (First Data, Fiserv)
- Game playing (The Last of Us, Halo, Sim City)
- Information retrieval & search engines (Google, Yahoo, Bing)
- Inputted facts (IBM Watson)
- Optical character recognition (Google Docs)
- Machine perception (Computer vision, machine hearing, and machine touch)
- Market segmentation (IBM)
- Medical diagnosis (Entopsis, Google X nanoparticles)
- Natural language processing (Google Translate, IBM Watson — Fluid, MD Buyline, Welltok, Healthline, Elance)
- Protein prediction (Noble Research Lab)
- Recommender systems (Amazon, Netflix)
- Robot locomotion (Honda ASIMO)
- Sentiment analysis (Twitter, Google Prediction API, AlchemyAPI, BeyondVerbal)
- Speech and handwriting recognition (Google Translate)
- Text categorization (Gmail, Outlook)
Playbook for Integrating AI As a Core Business Strategy
1. Get clarity on the business problem that you are trying to solve.
As a senior executive, you shouldn’t be trying to figure out which AI or machine learning approach to apply but rather to determine what are the actionable insights that will make your offering defensible in the marketplace. For Google it meant better search results.
2. Familiarize yourself so that you can competently evangelize the value of AI within your organization.
Remember, you’re not trying to become an AI expert but to have a reasonable understanding of AI and machine learning concepts
For an introductory overview, consider these resources:
- Artificial Intelligence: A Modern Approach (3rd Edition) by Stuart Russell and Peter Norvig
- Machine Learning by Thomas M. Mitchell
For a cliff note version, refer to AI co-founder John McCarthy’s article: http://www-formal.stanford.edu/jmc/whatisai/whatisai.html
For more in-depth, search for specialized books that cover machine learning, natural language processing, robotics, computer vision, neuroscience, probabilistic reasoning/ programming, logic, bioinformatics, etc.
Another great resource is the Association for the Advancement of Artificial Intelligence (AAAI) non-profit scientific society.
3. Recognize challenges and limitations.
As alluded in the beginning of the article, we are far from reaching strong AI. That means that business expectations have to recognize the limitations of AI applications.
In some cases automated learning systems need to be combined with hand-coded knowledge to produce better results.
Depending on the subfield, some systems cannot reach a high degree of accuracy without human assistance, such as in the case of recognizing images. In those cases, a crowdsourcing approach like the Amazon Mechanical Turk, reCAPCHA, and EyeWire helps refine the model further through human input.
A real challenge is data integration, integrating across different data sets. Google’s Alon Halevy notes that “no matter how much you speed up the computers or the way you put computers together, the real issues are at the data level.” The relationship between the different schemas must be understood before the data in all those tables can be integrated. Additionally, companies are shifting to using both SQL and NoSQL, structured or unstructured relational database, formats for data storage depending on the application.
4. Partner with AI research institutions.
MIT (CSAIL), Stanford (SAIL), Carnegie Mellon, UC Berkeley, University of Toronto, University of Washington, to name just a few, are the world’s most renowned institutions for AI research. Leverage their expertise by partnering with them on your next AI project.
5. Hire the right talent.
To do it right, your organization has to commit to building the right team. For instance, adding to Google’s already deep AI bench, Google hired futurist Ray Kurzweil in 2012 as the Director of Engineering to oversee Google’s most forward-leaning ideas.
Because the AI field is interdisciplinary that crosses computer science, mathematics, psychology, linguistics, philosophy and neuroscience, your resource matrix has to be multi-disciplinary.
In some cases, it might make more sense to hire a consultancy with the expertise in AI rather than to build from ground up.
6. Start experimenting.
Get your team to start experimenting with open source code and libraries on GitHub and other sources.
It’s easy to get caught up on the wearable and the Internet of Things sensors, hardware and communication protocols but the key differentiator to your solution will be the actionable intelligence that it derives from data.
Spend the time to seek the truth about AI and appropriate the fundamental knowledge. Equipped with a powerful arsenal, you will then be ready to craft a defensible business strategy. In turn, you have the potential to create a higher valuation firm that will lead the competitive pack. Start applying the scale and power of AI today.
Originally published on Wired on December 4, 2014. Author Scott Amyx.