1、Powered by DHL Trend Research ARTIFICIAL INTELLIGENCE IN LOGISTICS A collaborative report by DHL and IBM on implications and use cases for the logistics industry 2018 PUBLISHER DHL Customer Solutions systems that approximate, mimic, replicate, automate, and eventually improve on human thinking. Thro

2、ughout the past half-century a few key com- ponents of AI were established as essential: the ability to perceive, understand, learn, problem solve, and reason. Countless working definitions of AI have been proposed over the years but the unifying thread in all of them is 1 UNDERSTANDING ARTIFICIAL I

3、NTELLIGENCE Understanding Artificial Intelligence3 that computers with the right software can be used to solve the kind of problems that humans solve, interact with humans and the world as humans do, and create ideas like humans. In other words, while the mechanisms that give rise to AI are artifici

4、al, the intelligence to which AI is intended to approximate is indistinguishable from human intelligence. In the early days of the science, pro- cessing inputs from the outside world required extensive programming, which limited early AI systems to a very narrow set of inputs and conditions. However

5、 since then, computer science has worked to advance the capability of AI-enabled computing systems. Board games have long been a proving ground for AI research, as they typically involve a finite number of players, rules, objectives, and possible moves. This essen- tially means that games one by one

6、, including checkers, backgammon, and even Jeopardy! to name a few have been taken over by AI. Most famously, in 1997 IBMs Deep Blue defeated Garry Kasparov, the then reigning world champion of chess. This trajectory persists with the ancient Chinese game of Go, and the defeat of reigning world cham

7、pion Lee Sedol by DeepMinds AlphaGo in March 2016. Figure 1: An AI timeline; Source: Lavenda, D. / Marsden, P. AI is bornFocus on specific intelligenceFocus on specific problems The Turing Test Dartmouth College conference Information theory-digital signals Symbolic reasoning Expert systems Source:

8、Nvidia 1950s1960s1990s2010s2000s1980s1970s AI, MACHINE LEARNING Source: Getty Images Sedols defeat was a watershed moment for the prowess of AI technology. Previous successes had depended on what could be called a brute force approach; systems learned well-structured rules of the game, mastered all

9、possible moves, and then programmatically decided the best move at machine speed, which is considerably faster than human decision making. In a traditional Go board of 19 by 19 lines, there are more possible combinations than the number of atoms on planet earth, meaning it is impossible for any comp

10、uting system available today to master each move. DeepMinds AlphaGo effectively had to develop a sense of reasoning, strategy, and intuition to defeat Sedol; something that Go players have tirelessly tried to perfect for over 2,500 years yet DeepMind trained AlphaGo to do in a matter of months. The

11、important outcome from Sedols defeat is not that DeepMinds AI can learn to conquer Go, but that by extension it can learn to conquer anything easier than Go which amounts to a vast number of things.1 Current understanding of AI can quickly become convoluted with a dizzying array of complex technical

12、 terms and buzz- words common to mainstream media and publications on the topic today. Two terms in particular are important in understanding AI machine learning which is a subset of AI and deep learning which is a subset of machine learning, as depicted in figure 3. Whereas AI is a system or device

13、 intended to act with intelligence, machine learning is a more specific term that refers to systems that are designed to take in information, Understanding Artificial Intelligence5 Figure 4: A diagram of a neural network with six inputs, seven tuning parameters, and a single output; Source: Nielsen,

14、 M. DIAGRAM OF A NEURAL NETWORK INPUT LAYER HIDDEN LAYERS OUTPUT LAYER Problem Type Image Recognition Loan Approval Online Ad Placement Inputs Picture(s) Loan application Social media profile, browsing history Hidden Layers Person? Face? Gender? Age? Hair the intention of the system is to learn from

15、 the real world and adjust the learning model as it takes in new informa- tion and forms new insights. In simplified form, figure 5 depicts how deep learning algorithms can distinguish the content of an image, as well as where the elements of the image are in relation to one another, by analyzing pi

16、xel data alone. The human visual cortex is constantly doing this without our conscious awareness; however this perceptive ability in computers is truly novel. This is the type of system that is more useful in addressing real-world data challenges, which is why deep learning systems are the ones that

17、 have been directed at extremely large and fast-moving datasets typically found on social media platforms and in autonomous vehicles. Deep learning is typically done with neural networks. Neural networks are humanitys best attempt to mimic both the structure and function of the human brain. As new d

18、ata is fed into a neural network, connections between nodes are established, strengthened, or diminished, in a similar fashion to how connections between neurons in the human brain grow stronger through recurring experiences. Furthermore, each connection in a neural network can be tuned, assign- ing

19、 greater or lesser importance to an attribute, to achieve the quality of the output. Figure 5: Deep learning goes beyond classifying an image to identify the content of images in relation to one another; Source: Stanford Instance Segmentation Object Detection Classification + Localization Classifica

20、tion Single Objects Multiple Objects Understanding Artificial Intelligence6 1.2 How Machines Learn: Three Components of AI Despite the oversimplification that tends to define AI in the popular press, AI is not one single, unified technology. AI is actually a set of interrelated technology components

21、 that can be used in a wide variety of combinations depending on the problem it addresses. Generally, AI technology consists of sensing components, processing components, and learn- ing components (see figure 6). Sensing: The Fuel of AI To be able to understand or “sense” the real world, AI must tak

22、e in information. As real-world information comes in many forms, AI must be able to digest text, capture images and video, take in sound, and eventually gather information about environmental conditions such as tem- perature, wind, humidity, etc. everything that is typically understood by humans thr

23、ough our sense of touch. One of the most mature AI sensing capabilities is text- based processing. While AI systems have been processing structured data from databases, spreadsheets, and the internet for many years, recent advances in deep learn- ing have improved AIs ability to process and understa

24、nd unstructured data. Comments online, in social media, and even within apps are unstructured, so this critical capa- bility dramatically increases the amount and diversity of inputs that AI can leverage to understand the world. Putting it all together A FULL AI LEARNING CYCLE Figure 6: A full AI le

25、arning cycle; Source: IBM / DHL 1. Training data 2. Data gathered continuously from the environment, sensors and online behavior 3. Data is aggregated and harmonized 4. Machine learning framework processes data 5. Patterns and trends are revealed, generating insight 6. System takes different actions

26、 to drive value. New action is used as input to improve self- learning of the system. TextImageSoundMachine Sensing Processing Learning Much of our spoken interaction can now be captured by microphones and made sense of by AI systems. AI systems can consider the context in which spoken words were ca

27、ptured and, with access to large enough datasets containing similar and related phrases, can transform this once unusable data into valuable insight. Advances in speech-to-text technology are significantly enabling voice-driven AI. Today the comprehension ability of AI-driven voice assistants is sur

28、passing that of humans. The key metric for measuring speech-to-text performance is word error rate; effectively, how accurately does a per- son or system translate and interpret the words contained in a given voice sample. In a typical interaction between two people, the average percentage of words

29、misunder- stood by each person is 6%. Today the best AI-driven voice assistants are able to achieve a word error rate of 5%. Figure 7: Smart speakers with AI-driven voice assistants; Source: Heavy Understanding Artificial Intelligence7 2 Chen, F. (2017). 3 Meeker, M. (2014). 4 Chou, T. (2016). And a

30、s AI-driven voice assistants improve their models and comprehension with each new query (in other words, as they are given new data to learn from) their word error rate continues to fall.2 Images are another rich source of insight from unstructured data. It was estimated that even four years ago 1.8

31、 billion images were uploaded to the internet daily, and this num- ber continues to grow.3 Fortunately, many AI capabilities have been developed to process information from images. Companies like Google have leveraged this type of AI for years in consumer settings, and an increasing number of compan

32、ies are deploying static and video-capable systems in their daily operations. As AI continues to get better at turning the vast sea of visual information into system- usable content, the accuracy with which these systems understand our world is also increasing. The Internet of Things (IoT) is alread

33、y making machine data available for consumption by AI-based systems, often for the first time. IoT involves collecting large heterogeneous datasets from vast fleets of heterogeneous devices, but making sense of this information and learning from it can be a challenging task even for the advanced dat

34、a analytics tools of today. The computer understands the picture and sees that a woman is throwing a frisbee in a park DEEP LEARNING IN ACTION Figure 8: The evolution of picture understanding with deep learning; Source: IBM 1980s2012 Scanned DigitsNatural Photos Vision Source: DHL AI IN THE INTERNET

35、 OF THINGS 5. Action New action taken to drive value by orchestrating assets differently. New action is used as input to improve self-learning of the system. 3. Data How data is gathered from the connected devices. 1. Things Billions of connected assets equipped with sensors, carrying out various ta

36、sks. Internet of ThingsArtificial Intelligence 4. Insight What an AI model uncovers and learns from patterns within large volumes of complex data. 2. Connectivity How the devices are connected. Understanding Artificial Intelligence8 Figure 10: An overview of machine learning techniques; Source: Jha,

37、 V. TAXONOMY OF MACHINE LEARNING METHODOLOGIES Game AI Real-Time Decisions Robot Navigation Learning Tasks Skill Acquisition Identity Fraud Detection Image Classifcation Customer Retention Diagnostics Advertising Popularity Prediction Weather Forecasting Market Forecasting Estimating Life Expectancy

38、 Population Growth Prediction Classifcation Regression Supervised Learning Reinforcement Learning Meaningful Compression Structure Discovery Big Data Visualization Recommender Systems Targeted Marketing Customer Segmentation Feature Elicitation Dimensionality Reduction Unsupervised Learning Clusteri

39、ng Machine Learning Processing Source: Nvidia THE DIFFERENCE BETWEEN A CPU AND GPU CPU Multiple Cores Optimized for serial tasks GPU Thousands of Cores Optimized for many parallel tasks Figure 12: Ray Kurzweils Law of Accelerating Returns depicts the exponential growth of computer processing power a

40、nd technology innovations throughout history, and anticipates computers will exceed human intelligence in the future; Source: TIME / Wikipedia LAW OF ACCELERATING RETURNS THE ACCELERATING PACE OF CHANGE 1 AND EXPONENTIAL GROWTH IN COMPUTING POWER 2 Computer technology, shown here climbing dramatical

41、ly by powers of 10, is now progressing more each hour than it did in its entire first 90 years 0.00001 10 20 10 15 10,000,000,000 100,000 I 10 26 2045 Surpasses brainpower equivalent to that of all human brains combined Surpasses brain- power of mouse in 2015 Surpasses brainpower of human in 2023 WI

42、LL LEAD TO THE SINGULARITY 3 Power Mac G4 The first personal computer to deliver more than 1 billion floating-point opera- tions per second Apple II At a price of $1,298 the compact machine was one of the first massively popular personal computers UNIVAC I The first commercially marketed computer us

43、ed to tabulate the U.S. Census, occupied 943 cu. ft. Colossus The electronic computer with 1,500 vacuum tubes helped the British crack German codes during WWII Analytic engine Never fully built, Charles Babbages invention was designed to solve computational and logical problems Computational calcula

44、tions per second per $1,000 Agricultural Revolution 8,000 years Industrial Revolution 120 years Lightbulb 90 years Moon Landing 22 years World Wide Web 9 years Human Genome Sequenced 52000120201900 Understanding Artificial Intelligence10 Big Data: The existence of plentiful and

45、 easily accessible data is not a new phenomenon, however its ever-increasing volume, velocity, and variety is a key part of the AI story. Even though AI could exist on a smaller scale without these advances, AI requires data to demonstrate its full power. While new types of data have emerged in the

46、past few years, and while there is a significant increase in the pace at which data is created and changes, AI systems are currently consuming only a tiny fraction of available data. This has been true for a long time. So even if data quantities were to stagnate, and the rates of data volume and vel

47、ocity were to remain constant, AI would still have a lot of data to ingest, contextualize, and understand. Algorithmic Improvements: The increasing abundance of data being created every day has invited researchers, data scientists, and software engineers to conceptualize sophis- ticated new algorith

48、ms capable of ingesting large volumes of complex data. Because of this, today AI is not merely- capable of handling the rapid assembly of large and quickly changing datasets but in fact thrives on this. These big datasets make the best contribution to AIs ability to learn when they are complex, so t

49、he more diversity in the data domain the better. This is an advantage AI systems have over other data processing methods: whereas standard systems get bogged down with large complex datasets, algorithmic improvements in recent years have improved significantly to be able to handle large volumes of hetero- geneous data, enabling the detection of patterns and discovery of correlations that