Artificial intelligence (AI) refers to the development of machines that are capable of perceiving, thinking, learning and adapting their behavior, just like biological organisms.
The concept of intelligent machines has been around since Ancient Greece, and human beings have been fascinated with it ever since. The term “artificial intelligence” was coined by cognitive scientists at a 1956 conference at Dartmouth University, which kicked off fresh waves of AI research. After an uneven history including two “AI winters” where research mostly halted, development of artificial intelligence has grown more in the last three years than in the three previous decades. The market for AI-related hardware, software and services is expected to grow from $8 billion in 2016 to $47 billion by 2020.
Artificial intelligence takes various forms. Below are some of the terms and categories that arise in discussions of AI.
First coined in 1959, the term “machine learning” refers to computers that have the ability to learn without being explicitly programmed. This is the field of AI which is most promising today, in terms of enabling the tools that are catalyzing change across industries and throughout society at large. There are three primary types of machine learning:>/
- Supervised: For example, input/output signals are fed to the program, enabling it to create a rule that maps the inputs to the outputs.
- Unsupervised: The program is tasked with discerning the structure through which data inputs are organized.
- Reinforcement Learning: The program performs in a changing environment and receives feedback (rewards or punishments) based upon its actions.
Deep learning is a term for today’s leading edge of machine learning, based on “neural networks” that function analogously to the human brain. Essentially, this approach is a further exploitation of machine learning that is able to take on dramatically larger datasets and make adaptive decisions based on that data. Some of its expressions are contained within the following three concepts:
1. Neural networks
The concept of neural networks has become more common in AI. It refers to a set of connected logical gates which sort incoming data according to binary criteria. These networks contain multiple hidden layers between the input and output, in which each layer changes or interprets the data for the succeeding layers. They were developed as a way of classifying massive datasets, and their sophistication makes it possible to sort through complex information such as elements of images, in order to identify which ones are meaningful within a given context. Additionally, neural networks also allow software systems to learn and refine their identification processes. The shift in the direction of organic brain-like learning is being further expanded by AI researchers who also study human physiology and neuroscience.
2. Natural Language Processing
Natural language processing (NLP) is one example of the value neural networks offer in analyzing massively complex systems. NLP enables software systems to understand the meaning of human language, instead of relating to it only as a set of symbols. This technology gives rise to numerous new possibilities across every industry, because it lets systems rise above just searching for specific words. NLP allows the computer to actually understand a person’s spoken or written intent, thus giving rise to new levels of personalization. While this field is still evolving, it has the potential to revolutionize the way that people interact with digital devices.
3. Computer Vision
Just as NLP is leading computers to gain a human-like perception of the meaning of language, Computer Vision (also called “perception”) is a method by which computers can ascribe inherent meaning to visual images. Just as the human brain makes sense of the patterns of light that enter through the eyes, computer vision allows software systems to differentiate between elements of an image and pick out the consistent or important parts. As this capacity becomes increasingly sophisticated, there are many potential applications. Self-driving cars depend on computer vision, for instance, and startups such as DeepScale are receiving venture funding to improve the sensors that enable this potentially enormous industry. The ability to automatically derive meaning from visual information also has potential usefulness in healthcare applications.
Block chain technology was originally created for the purpose of creating bitcoins, but it has other potential applications as well. By enabling the distribution of digital information, without allowing that information to be copied, block chain creates an “incorruptible digital ledger.” Information that exists on a blockchain is public, existing in millions of simultaneous locations, and is constantly reconciled in real time. It can’t be hacked because it isn’t centralized. Just as a Google document can be shared in real time between several people, instead of having to be revised and sent back and forth sequentially, block chains are shared among any number of users. The robust, decentralized nature of block chains could create secure new models for stock trading and other financial transactions.
AI Applications in Use Today
While automated customer service has been around for some time now, NLP is taking it to new levels of responsiveness. Chatbots are becoming better at perceiving user needs and intent from language cues, and to use machine learning to improve their comprehension. Increasingly able to recognize emotion, they are facilitating numerous transactions on company sites, social media platforms and messaging apps.
A voice-driven form of chatbots, these smart devices are listening and interacting with human speech, taking on increasingly more complex tasks and leveraging what they learn to become indispensable to their owners. BotCube’s exhaustive directory of tools, platforms, newsletters, influencers, and more provides some sense of how large the field is now. It extends far beyond Alexa, Siri and Google Home.
Cognitive Copy Generation
The ability to analyze the intent and sentiment within language means that written content can now be optimized to inspire action in each individual reader. The marketing potential offered by this emerging has sent AI-generated content leader Persado to garner $66 million in venture capital and to achieve a spot on CNBC’s Disruptor 50 list.
The core of marketing has always centered around perceiving and responding to customer needs, and AI now automates this learning curve. Marketing software such as that offered by Albert.ai offers a “self-driving solution for cross-channel campaign execution, testing, optimization, analysis, and insights.” Indeed the entire art and science of marketing, including customer segmentation and profiling, is undergoing a profound digital transformation. Predictive analytics, dynamic pricing, and algorithms for ad targeting are only some of the innovations that are driving change in this field.
International corporate taxes are another area where massively complex information is currently being handled by highly paid human beings. Rainbird is in the vanguard of intelligent tax platforms, using AI to “make complex, human-like judgments based on a combination of real-world expertise, the applicable rules and regulations and client-specific data.” This scalable service provides a clear audit trail to promote compliance and risk management.
While manufacturing has been moving toward automation for some time, supply chain management is now being disrupted through the introduction of AI. More than 30,000 Kiva robots fulfill orders at 13 Amazon fulfillment centers, but that’s only the beginning. Amazon is actually positioning itself to provide what AI experts term “contextual awareness,” linking every piece of its supply chain in just-in-time inventory by making increasingly smart predictions of what customers will order.
Self-Driving Cars and Automated Navigation
Deep Learning and Machine Visualization have led to “convolutional neural networks” which “map the raw pixels from a front-facing camera to the steering commands for a self-driving car.” The car’s learning requires only minimal input from human beings, as it continually integrates the countless fragments of behaviors it was exposed to during “training,” such as the angle of human steering. The end-to-end learning system creates simultaneous internal representations of the numerous steps involved in driving, without needing a human being to break down and categorize them. This organic style of learning is increasingly characterizing AI.
The role of AI in security is still in the formative stages, but at the 2017 RSA security conference in San Francisco, Mike Buratowski, senior vice president of Fidelis Cybersecurity, stated that its most crucial use lies in its ability to handle the massive quantity of data. “Right now, it’s an issue of volume. There’s just not enough people to do the work.”
Additionally, MIT is developing a cognitive security system that can predict 85 percent of cyber attacks by incorporating what it learns from its human collaborators. Once an anomaly is detected, then the program brings that to the attention of a human being and also refines its own detection algorithm in the process.
One of the most promising developments in Health AI comes out of Canadian company Deep Genomics. This business uses AI to extract key information from the massive amount of data in the human genome and to predict how genetic traits will actually be expressed. These new packets of information have enormous potential for developing precisely targeted drugs based on genetic personalization.
The growth of AI creates a positive feedback loop enabling further growth, because the larger the dataset (by order of magnitude), the better the outcome. Evolving AI systems offer progressively better efficiency and lower error rates, as they transform the digital environment by infusing it with true intelligence.