Artificial Intelligence (Summer 2017)

From New Media Business Blog

Jump to: navigation, search


Contents

Introduction

The study of artificial intelligence is self-described as "the study of intelligent agents" by experts in the field. Artificial intelligence is generally used to describe machines simulating what humans deem intelligent. With exponential increases in processing power of consumer grade processors throughout recent years and advancements in software systems, artificial intelligence is becoming increasingly capable in wider fields of application. Functions that were once thought uniquely human and impossible for machines to comprehend, such as creativity and empathy are being emulated. More and more businesses are moving in the direction of automation and consequently a large number of the workforce is projected to be replaced by seemingly intelligent machines.

History

The Turing Test

The term artificial intelligence originated from the Dartmouth Conference in 1956. However before the term was coined, the study of artificial intelligence has already been well documented. While there is numerous documentation of ideas of automatons, none are as prominent that the work by Alan Turing. Alan Turing had published his seminal paper, "Computing Machinery and Intelligence" on the issues of artificial intelligence in 1950. Within this paper, Turing noted the difficulty in defining the terms within the question "can machine think" and thus proposed a shift from finding a universally-agreed definition of artificial intelligence to considering a less ambiguous and closely-related question "Can a machine act indistinguishably from the way a thinker acts?". In his paper, he describes a party-game involving three parties as an experiment for the purpose of answering this question. This test is now modified into the form of having a panel of judges distinguish are they talking with a human or machine through a text interface.

The Imitation Game

“Imagine three rooms, each connected via computer screen and keyboard to the others. In one room sits a man, in the second a woman, and in the third sits a person - call him or her the "judge". The judge's job is to decide which of the two people talking to him through the computer is the man. The man will attempt to help the judge, offering whatever evidence he can (the computer terminals are used so that physical clues cannot be used) to prove his man-hood. The woman's job is to trick the judge, so she will attempt to deceive him, and counteract her opponent's claims, in hopes that the judge will erroneously identify her as the male.”

His original "imitation game" eventually evolved into what we know as the Turing test today with the goal of determining "Can a machine imitate a human in conversation?". [1]

Deep Blue

Deep Blue was a chess-playing computer developed by IBM in 1985. In 1996 became the first computer to win a chess game against a reigning world champion but ended up losing the match. However, it went through significant upgrade which leads to its victory in 1997 against Garry Kasparov.

Deep Blue worked by storing all possible chess board positions in memory and recursively choosing optimal moves through min-maxing. Some argue that Deep Blue is not technically AI because it uses formula to predict and choose the best possible countermove against a person and it cannot act based on intuition.


Watson

Watson is a question-answering (QA) computing system built by IBM to apply advanced natural language processing, information retrieval, knowledge representation, automated reasoning, and machine learning technologies to the field of open domain question answering.

The development was led by David Ferrucci, head of IBM Semantic Analysis. In a 2006 test run, Watson was given 500 clues from past Jeopardy! programs but only managed to answer 15% correctly. During 2007, the IBM team was given three to five years and a staff of 15 people to solve the problems. In 2011, Watson managed to beat Ken Jennings and Brad Rutter, two of the most successful contestants in Jeopardy! showcasing its question-answering capabilities.

According to IBM, "The goal is to have computers start to interact in natural human terms across a range of applications and processes, understanding the questions that humans ask and providing answers that humans can understand and justify." It has been suggested by Robert C. Weber, IBM's general counsel, that Watson may be used for legal research.The company also intends to use Watson in other information-intensive fields, such as telecommunications, financial services, and government.

AlphaGo

AlphaGo is part of Google’s DeepMind project. AlphaGo is the first computer program to defeat a professional human Go player in 2017, the first program to defeat a Go world champion, and is probably the strongest Go player in history. AlphaGo combines an advanced tree search with deep neural networks. These neural networks take the state of a Go board as an input and process it through a number of different network layers containing millions of neuron-like connections. Based on reinforced learning, AlphaGo's strategy embodies a spirit of flexibility and open-mindedness: a lack of preconceptions that allows it to find the most effective line of play.[1]

How does AI work?

Conditional Statements

The majority of autonomous machines is written in conditional statements. Conditional statements are based on “If...then...else”. These statements allow programmers dictate how would computers react given certain situation. While the actions of the computer are determined in advance, the computer can still appear to be extremely intelligent. Conditional statements while easy to implement, requires the prediction of every condition before hand and has limited ability to face unexpected situations. Deep Blue is based on an extremely complex version of conditionals.

Semantic Network

To address the shortcomings of conditional statements, computer scientists mimicked the thought processes of humans and created semantic networks. Semantic networks are based on pattern matching and association. They are used to classify objects based on attributes. ELIZA, the first program that managed to come close to passing the Turing test, and Watson, who won the game show Jeopardy are based on Semantic Networks.

Machine Learning and Training

Currently, the forefront of artificial intelligence is machine learning. Machine learning allows computers to learn without being explicitly programmed*. Machine learning tasks are typically classified into three broad categories[2], depending on the nature of the learning "signal" or "feedback" available to a learning system. These are:

  • Supervised learning: The computer is presented with training data with different labels. For example, a set of emails that are labeled spam, and another set labeled not spam. The goal is for the computer to be able to find correlations by itself and be able to classify future emails.
  • Unsupervised learning: No labels are given to the learning algorithm, leaving it on its own to find patterns in the input. Unsupervised learning can be a goal in itself (discovering hidden patterns in data) or a means towards an end (feature learning).
    Regression analysis in Excel is an example of this.
  • Reinforcement learning: A computer program interacts with a dynamic environment in which it must perform a certain goal (such as driving a vehicle or playing a game against an opponent). The program is provided a function to maximize, and outcomes in that would increase or decrease points. The goal is for the program to navigates its problem space to maximize the function.
    Example: In self-driving cars, the algorithm is assigning points to different types of scenarios; for instance, crashing to a pole is -100, and crashing to a pedestrian leading to the death of the pedestrian is -1000. Thus, the machine will try to maximize the points in any given scenario.
    AlphaGo is based on this technology.

Business Applications of AI by Type

Rule-Based

Industrial Automation

While the majority of applications are based on conditional, as artificial intelligence become more and more common, it ceases to become AI. Robots replacing factory workers are not seen as highly intelligent anymore. Yet they are actually replacing semi-skilled labour at an alarming rate. Foxconn has replaced 60,000 workers in 2016. [3]

Video Games

AI has been a natural component of many video games, simulating non-player character behaviour. The video game industry has grossed over 91 billion USD in 2016 and continues to grow with further developments in hardware. Physics and behaviour simulation plays a large role in this industry and often computer controlled bots outperform average humans. [4] [5]

Semantic Network-Based

After ELIZA was able to display having human like responses, many chatbots were developed based on a similar model. These chatbot programs became the backbone of many internet firms. Most prominent of all may be Ashley Madison, grossing 115 million in 2014. These firms provide companionship with the aid of an AI and is widespread across the globe [6]



Since Watson made headlines in 2011, it has been developed into assisting data analysis in many fields.

Legal Research

ROSS, based on IBM Watson is gaining traction in legal research. Firms such as Baker & Hostetler, a US based large scale law firm, has been using ROSS to assist in legal research for precedents. [7]

Medical Field

Within the medical field, IBM Watson is in its beta for medical research, genome sequencing, and cancer treatment [8]

Machine Learning-Based

Supervised Learning

Opportunities for disrupting the current workforce by supervised learning are endless. Supervised learning currently powers many technology we currently use. Spam filters, for example, classify various incoming emails with set labels. Instagram is implementing this technology to attempt to filter out cyberbullying. Functions like these, Instagram-implements-an-ai-system-to-fight-mean-and-harassing-comments when done by labour, as time consuming and expensive, This technology, while simple can be used to reduce hours put into data analysis. The need for human resource departments to sorting resumes into desirable and undesirable applicants and the time used on lawyers to go though evidence can be greatly lowered by having such filters discover anomalies.

Unsupervised Learning

Dunnhumby, a UK data mining firm, represents $20 billion in annual sales of a variety of retailers. With the large amount of data, oftentimes data mining firms are looking for trends or correlations that researchers do not even know exist. Thus unsupervised learning is used to a great extent to find relationships between large sets of data. [9]

Reinforcement Learning-Based

AI in Autonomous Vehicles

One of the greatest change to the transportation industry is automated vehicles, automated vehicles are being researched by the major technology firms and car manufacturers in the world. Including Apple, Google, Uber, Tesla, and more. The technology is predicted to replace a large number of our workforce. Self-driving cars requires sensors to perceive the environment, processes and chooses specific actions based on information gathered. Artificial Intelligence is then used to “learn” how to drive using different methods.[10].Combine with interactive communication and active learning, it creates a situation where autonomous cars, through Artificial Intelligence algorithms, can improve their ability to react to situations on the road without actually having to experience those situations first-hand.

Finance

In finance, not only bookkeeping and tax preparation are increasingly automated, but a number of hedge funds are currently using AI to do day trading [11]. These AI are able to be extremely rational, and analyze large amounts of data. With machine learning, they are capable of changing their behaviour immediately to respond to the opportunities and risks faster than any human trader would. Furthermore, robo advisors are starting to replace financial advisors at an increasing pace.

Other examples of using artificial intelligence in the financial industry includes fraud prevention, operations managements and property management.

Specifically, credit card companies can improve the accuracy of real-time approvals and reduce false declines using artificial intelligence.[12]

Furthermore, the use of high frequency trading is also a rising trend in the financial industry, in which artificial intelligence is be used to automate stock trading.[13]

Health Care

Google’s DeepMind health is an example of the application of AI in the healthcare industry. As part of Google’s DeepMind project, DeepMind health works with hospitals on mobile tools and AI research to help get patients from test to treatment as quickly and accurately as possible [14] DeepMind health aims to help with challenges facing healthcare systems, including empowering patients to look after themselves and their families’ health, and supporting coordinated ongoing care around patients’ needs. Specifically, it can send immediate alerts to doctors when a patient deteriorates, furthermore, it is also proven to save at least two hours of work per day for nurses.


Shopping and Payments

Combined with artificial intelligence technology, the shopping experience could be upgraded in the future. The technology helps the company to solve the pain points and introduce refreshed shopping experience. Within this new field, as the first mover, Amazon launched the new retail concept of “Just Walk Out Shopping”, which introduces the shopping experience with no lines and no checkout process. [1] The further implication of this new type of shopping experience is that the refresh processes can be applied into more other services such as car rental, cafeteria, clothing store, etc.




Other Applications

AI Companionship

AI can also be used as a companion for children, the elderly, or people with disabilities. For example, children with autism.[2]

A study in New Zealand indicates that interaction with robots can significantly decreases loneliness for the elderly, such companionship has the same impact as companionship with residential dogs.Surprisingly, the study also shows that people tended to spend more time interacting with robots than with dogs. The cause was found in the conversational capacity that robots had to engage their partners.[3]

Automated Armies

Artificial intelligence is one of the most attractive fields that militaries want to combine with. In recent trends, many nations in addition to the U.S., have already implemented automated weapons capable of tracking and firing on a target if authorized. [4] These include South Korea’s Samsung SGR-A1 “Intelligent Surveillance and Security Guard Robot” deployed for perimeter defence at military installations, and Israel’s See-Shoot border defence system capable of establishing a mile-deep kill zone along the Palestinian border. Integrating AI weapon systems into military platforms has a broad range of applications, but numerous concerns, both practical and ethical. Autonomous weapon platforms have the potential to significantly reduce the manpower required to performing a myriad of tasks. This is important given that some of these tasks like patrolling and mine clearing, are exceptionally dirty, dull, or dangerous. Autonomous weapon systems can perform the same task for longer a duration more reliably. They lack human limitations such as fatigue, boredom and injury.





Ethics

AI is able to mimic empathy, creativity, and values. Mimicking empathy is demonstrated through the success of Ashley Madison, creativity through music bots such as Emily Howell and Band in a Box, and values could be easily be programmed. But it could be argued that AI can only mimic, but never comprehend these concepts.

Empathy

Empathy are associated with feelings. When humans say “How are your pets doing.” They are expressing interest and concern for someone’s pets. But to a machine, this is a pre-programmed input to display some certain symbols. [5] Machines as of yet are unable to express interest nor concern as they do not have to ability to intrinsically feel for others. This leads to the issue of empathy based jobs such as customer service or psychologists. Automated customer services had been widely employed for a long time, without the human interaction, it could be argued that customer services are merely self help or automated services. Furthermore, services such as those provided by Ashley Madison, which business model is based on getting customers to buy virtual gifts for woman being impersonated by chatbots further calls into question how society should view the advancements of AI.

Creativity

Creativity is another area where AI would be unable replace. The current method which programs generate art are based on procedural generation [6]. This process is done by having variables generated while still following certain set criteria. Art exist to communicate emotions. While there is an objective scale of beauty known as golden ratio [7] if a machine is unable to comprehend emotions, it can not create but only mimic what already exist.While some may argue that art is always about generating similar patterns, real creativity can be seen in the work of genre defining artists, such as Beethoven, whom singlehandedly ended the Classical era by starting the Romantic era.

Values

Personal values exist along with cultural values. Personal values agree or diverge from cultural values based on the worldview of the individual, which determines the actions of the individual. The issue lies within what values does AI follow. The creators may input their values, the society may mandate their values, the user may choose value. For example, whose safety should autonomous vehicles prioritize when there is an accident. What values should an AI adopt, and should it reflect the creator, the society or the user. These are issues that have to be addressed. Values of AI, or lack thereof has been a common topic for works of fiction detailing the doom of mankind. With militaries starting to incorporate AI into their systems, the issue of what values should AI have will be even more complicated in the future.

Economics

As more and more of the labour force has been, and are projected to be replaced by automation, including skilled and professional labour, issues have arose regarding the ethics and economics. According to a study from Oxford University, 47% of US jobs would potentially be automated in the next 10 to 20 years. [8] Should business replace what could be replaced simply it can because it is more profitable. What then happens when half of the population are unemployed. Elon Musk and Bill Gates among others have advocated for taxing robots for a univeral basic income [9] . Essentially turning our economy into a leisure economy.

Security and Privacy


With the rise of personal digital assistants such as Apple’s Siri, Microsoft’s Cortana, Amazon’s Alexa, etc. These virtual helpers use artificial intelligence to parse what users say or type, and return useful information. They have been taught to guess at what users want to know before they are asked, providing notifications on road conditions at the appropriate time. However these services collect vast amont of user data. Queries are often stored with associated location data connected to user accounts and can paint a very accurate picture of users’ habits, travels, and preferences, which may cause a problem as they would be vulnerable to hackers who could gain access to sensitive servers, leading to potential for crimes such as identity theft. Furthurmore, the vast amount of data collected poses issues of privacy and monatizing of personal information.

Limitations

Image Recognition

reCAPTCHA is a free service that protects websites from spam and abuse. It uses an advanced risk analysis engine and adaptive CAPTCHAs to prevent third-party automated software from engaging in abusive activities while allowing valid users to pass. Each time a CAPTCHA is solved, that human effort helps digitize text, annotate images, and build machine learning datasets. This in turn helps preserve books, improve maps, and solve hard AI problems. Here are a few examples of reCAPTCHA, one of the method of detecting bots. It is mostly used on signup pages where it will test whether or not the user of the computer is actually a person. There are several different types where there are some that uses words, numbers, and pictures. However, all of these are in image form to make it difficult for bots to pass this test.

With the current limitation in image and speech recognition, Nvidia graphic has been developing project Xavier based on a new Volta Architecture. This will significantly improve graphic cards performance by up to twenty times stronger than the best GPU we have today.

Current Developments

With the current limitation in image and speech recognition, Nvidia graphics has been developing project Xavier based on a new Volta Architecture. Graphics cards will be at least twenty times stronger than the best consumer GPU we have today to address these limitations. Tesla is the first company to preorder and work with this technology to address major issues of auto driving techniques such as recognizing road conditions and extreme weather conditions.

References

  1. https://www.theverge.com/2016/12/5/13842592/amazon-go-new-cashier-less-convenience-store
  2. http://www3.imperial.ac.uk/newsandeventspggrp/imperialcollege/newssummary/news_2-2-2017-15-28-34
  3. http://www.cnn.com/2016/10/03/health/robot-pets-loneliness/index.html
  4. https://ww2.kqed.org/education/2016/09/14/are-artificially-intelligent-military-systems-worth-the-risk/
  5. https://books.google.ca/books?id=somvbmHCaOEC&pg=PA157&redir_esc=y
  6. https://stackoverflow.com/questions/180858/procedural-music-generation-techniques
  7. http://www-history.mcs.st-andrews.ac.uk/HistTopics/Golden_ratio.html
  8. https://www.cnbc.com/2016/11/18/elon-musk-says-robots-will-push-us-to-a-universal-basic-income-heres-how-it-would-work.html
  9. https://www.forbes.com/sites/ianmorris/2017/02/17/tax-the-robots-says-bill-gates/

Personal tools