AI
Cloud Native
"AI-as-a-service platform as the next big operating system"
Open Algorithm Model
"Numerous startups and boutique software shops are developing AI solutions to meet specific business needs, use cases, and verticalized issues"
Packet Adjunct model
"Several vendors [SAP, Oracle, Salesforce] are putting existing general-purpose AI platforms in the cloud"
Study finds that some smart tech can improve quality of life indicators by _________ %
10-30%
[A]s many as ______ percent of the activities individuals are paid to perform can be automated by adapting currently demonstrated technologies. In the United States, these activities represent about _________ in annual wages
45 $2 trillion
With the addition of _______, the use cases for RPA expand considerably
AI
Areas that Intelligent Interfaces demand more from IT
Bandwidth • Cloud and edge computing • IoT • Ethical risks, privacy, security, regulation
Benefits of AI
Benefits of AI: enhance current products, optimize internal operations, make better decisions, optimize external operations, & free workers to be more creative
AI has gained great prominence due to three factors
Big Data, Algorithims, Computing Power
_______ can be used whenever image recognition is needed
CNNs
___________ have displaced ______________ as the top application of AI
Chatbots, fraud detection
Types of AI deployments
Cloud Native, Package-adjunct model, Open Algorithm model
"Algorithmic Accountability Act"
Concern that AI is magnifying human bias • E.g., bias affecting algorithm on recidivism, crime likelihood, facial recognition, hiring expectations,
The Machine Learning Race Is Really a _________
Data Race"
Big Data
Data generated and stored at a massive scale
Deep Learning
Deep Learning: type of machine learning that can process a wider range of data resources, requires less data preprocessing by humans, & can often produce more accurate results than traditional machine-learning approaches; requires lots of data Convolutional neural network & recurrent neural network & Generative adversarial networks
The Percentage of Enterprises Deploying AI Has ________ in the Past Year (2018)
Tripled
Unsupervised learning is very important for ______________ systems
recommender systems
A typical RPA process deals with very ___________ data
structured
__________ is about using tech and data to improve the quality of life
"Smartness"
The new Nvidia architecture for GPUs is aptly name __________ and Nvidia is reaping profits from its investments in the _____ space
"Turing", AI
Generative adversarial networks combine two neural networks ______ and ____________ to create new content such as _____________
(Generator and Discriminator) (e.g., images)
By 2020 _______ % of operational bank staff engaged in back-office activities will rely on AI to do the nonroutine work.
, 20%
Steps for implementing RPA
1.A process (or a set of processes) is deemed to be a good candidate for RPA 2. A "process developer" creates the detailed set of steps for the automation bot 3. The bot becomes operational and uses software licenses to perform the steps 4. The bot is centrally monitored for performance 5. Performance improves and "everybody is happy"
Help with commuting: ________ reduction in commute times Use of digital signage or mobile apps Ease road congestion
15-20%
In 2020, AI becomes a positive net job motivator, creating _______ jobs while only eliminating __________jobs.
2.3 million, 1.8 million
Through 2020, employers with data- and AI-for-good programs will have _______ higher retention and ________ faster data and analytic job-fill rates.
20% higher retention and 50% faster
AI implementations up _______ in the last four years
270%
By 2023, _______ of smart city initiatives will lose political and public support and be discontinued for lack of integrated services and data analysis.
30%
By 2020, _______ of smart city roadmaps in emerging economies will incorporate strategies on resilience, public health and safety to stimulate economic development.
50%
By 2023, AI combines with human programmers to create "centaurs" performing ______ of traditional programmer workloads — doubling the throughput of stand-alone programmers.
50%
A]bout ____ percent of occupations could have _____ percent or more of their constituent activities automated
60 30
The list of actions that a robot is capable of performing can stretch to over __________ in some solutions, and additional actions can often be custom created using code
600
Training AI in GPUs can be _____times faster than doing the same work with CPUs
70
By 2021, ________ of enterprises will integrate AI to assist employee's productivity, causing 10% of them to add a digital harassment policy to workplace regulation.
70%
By 2021, the social impact of a digital economy will force ______ of cities with smart city ambitions to continuously audit for biases toward citizens
75%
Help with determining crime detection strategies __________reduction in fatalities _______ lower left _________ reduction in emergency response times
8-10% 30-40% 20-35%
________ %of marketing leaders surveyed expect that by 2019 they will compete, either mostly or completely, on the basis of CX
81%
recurrent neural network (RNNs)
A recurrent neural network (RNNs) stores data in context nodes to learn data sequences
Difference between RPA and IPA
Algorithim based decision making, self learning, subject matter experts, semi-structured data
Common Myths of AI
All black-box AI needs to be explainable AI is all about algorithms & models We don't need AI strategy AI can be free of bias AI will only augment/automate repetitive jobs that don't require expert skills There is one best AI vendor Deep neural networks are the best AI AI works in the same way the human brain does Intelligent machines learn on their own
Three stages in journey for full utilization of AI
Assisted Intelligence Augmented Intelligence: Autonomous Intelligence:
Five things to keep in mind about AI:
Develop a thesis on time to AI impact, AI growth will likely be exponential, Trust the machine, Know what to predict, Manage the learning loop.
AI's organizational Impact
Displacement effect and Productivity effect
___________ are ramping up AI investments, launching more initiatives, & getting positive returns
Early adopters
Computing Power
Far beyond what we were able to do in the past (Cloud)
Automation's 5 areas of impact
Fewer workers, Scaled capabilities, Greater speed, High quality, new capabilities
Business functions most impacted by automation
Finance, Administration, Customer service, Faclities management, sales
Convulutional Neural Network (CNN)
For convoluted neural networks (CNNs) we use a multilayered neural network in which each layer gets more complex features form the previous layer
___________ can create text that mimics a writer's style
GANs
___________ architectures are particularly good with data operations at scale because of __________
GPU, parallelism
Industries investing low in AI and seeing low return
Gov., Consumer Products, Financial services and Insurance
Big data has not slowed down partly because of Additional developments in_______ and Highly scalable storage based on ___________
Hadoop, NoSQL
_________ may result in unintended consequences that can actually inhibit AI
Hype
What does augmentation really mean?
Identifying situations in which humans can really be more productive when AI takes care of some of the work
Behavior Based Work
In behavior based work, the task is done simultaneously by both agents, cooperatively • So, it is not like "let's have the machine getting data and passing it to the human" • Instead, it should be "let's have both the machine and the human getting data based on what they can do best and work cooperatively based on the problem"
Reinforcement Learning
In reinforcement learning, an algorithm learns to perform a task simply by trying to maximize rewards it receives for its actions
Task Based Work
In task-based work, the action alternates between humans and machines, sequentially
Unsupervised Learning
In unsupervised learning, an algorithm explores input data without being given an explicit output variable
Graphcore is an "unicorn" developing an ___________
Intelligent Processing Unit (IPU)
Example of unsupervised learning
K-means clustering, hierarchical clustering
Top 3 barriers to AI implementations
Lack of staff skills Defining AI strategy Identifying use cases for AI
5 imperatives driving edge computing
Latency / determinism, Limited autonomy, privacy / security, Data / Bandwidth, Local interactivity
Industries investing heavily in AI and seeing low return
Life sciences and healthcare
Augmented Intelligence:
ML capabilities layered on top of existing info management systems work to augment human analytical competencies Lane detection drivers
Additional refinements in data access used by IBM watson
Massively parallel processing (MPP), In memory processing
Motivation to augment RPA with AI
Most documents in an organization are not structured (that is, not in a database or excel) • That percentual might be even bigger in areas such as legal departments, scientific institutions, etc.
Who in particular has been successful in the GPU domain, both with its architecture and accompanying software
Nvidia
Potential benefits for the consumer from AI and related technologies
Personalized products • In so many industries, such as apparel and fashion, but also including food and beverage (among others) • Online product recommendations • Making the experience more efficient and enjoyable • Timely service • Reducing turnaround, delivery, and consumer service times
Narrow AI -
Playing strategy games, writing newspaper articles, etc.
Algorithim improvement
Revolutionary approaches in how machines learn
RPA
Robotic process automation (RPA) tools perform 'if, then, else' statements on structured data, typically using a combination of user interface (UI) interactions, or by connecting to APIs to drive client servers, mainframes or HTML code. An RPA tool operates by mapping a process in the RPA tool language for the software "robot" to follow, with runtime allocated to execute the script by a control dashboard
Use cases for intelligent things
Robotic-assisted surgery • Greetings at a hotel • Prevention of crime through patrolling autonomous robots • Advanced agriculture • Cities of the future • Safer law enforcement • Self-driving taxis • Safer automobile transportation • Intelligent things on the battlefield
Intelligent interfaces can combine multiple stream of information ___________, __________, and ______.
Sensory • Physical • Biological
Voice use cases for intelligent interfaces include ______ & __________
Siri & Alexa
5 Steps to Understand Customers More Efficiently Through AI
Step 1: formulate your customer experience strategy Step 2: map and analyze your customer journeys Step 3: familiarize yourself with the AI solutions (s) suitable to understand your customers - faster and more efficiently I Step 4: decide whether to build or buy Our enterprise AI survey says that 45% of businesses would buy an AI solution — either buy and then customize it to their industry or buy an already customized solution. This is compared with 33% that build their own custom solution internally. Step 5: track and measure success
_____________ is Used when the meaning of the relationship is theorized but calculation is needed.
Supervised
Machine Learning Training techniques
Supervised, unsupervised, & reinforcement learning
T/F AI & ML are not the same
T
T/F As such, automation becomes a strategic priority, especially for successful organizations
T
T/F CNN receives an input and each layer of the input is able to become more specific in the features it identifies
T
T/F Depending on the type of work, one can have machines leading, being led, or working collaboratively with humans
T
T/F Diversity counteracts the biases of data, algorithms, and people who select them
T
T/F Implementing the algorithms used to make machines intelligent about a data set or problem is the easy part
T
T/F It is impossible to eliminate bias, but it is possible to anticipate and understand it
T
T/F Job reduction does not necessarily means layoffs
T
T/F McKinsey expect the job losses to be offset by job growth
T
T/F Most ML libraries support GPUs
T
T/F One misunderstanding is that sophisticated AI algorithms alone can provide valuable business solutions without sufficient data
T
T/F Opportunities for Intelligent interfaces include tracking customers' offline habits, new products & solution sets (think, micro-personalization), & efficiency
T
T/F Other use cases for intelligent interfaces include Computer vision, gesture control devices, embedded eye-tracking platforms, bio-acoustic sensing, emotion detection/recognition technology, & muscle-computer interfaces
T
T/F RPA and AI go together in successful organizations
T
T/F Reinforcement Learning can be adapted from other models
T
T/F Several companies working on neural network chips, Including Intel's Nervana
T
T/F Smart cities don't create jobs, but make local labor markets more efficient and lower the cost of living
T
T/F The underlying technology [in intelligent things] is improving at an unprecedented rate
T
T/F There is potential for companies that already have access to vast amount of data to become even more powerful
T
T/F These are all important steps for Computing Power • Nvidia releases first true GPU in 1999 • Amazon launches AWS and cloud computing in 2002 • MapReduce is released in 2004 to use parallelization • Storage and computing power costs keep decreasing • Hadoop evolving from the early 2000s to Spark in 2009
T
T/F This could be an analogy for GANs the Generator tries to create fake money, the discriminator tries to find out the problems, then the cycle continues until the Generator learns to create perfectly realistic money
T
T/F We might need to truly redesign certain jobs and, with that, better train people to conduct those jobs
T
Google investing in chip-design with its new __________. Based on the TensorFlow architecture
TPU (Tensor Processing Unit)
Three layers of a smart city:
Technology base (smartphones and IoT sensors) Applications (translate data into alerts, insights, and actions) Usage (by cities, companies, and the public)
Used when data classification is unknown and we need to find patterns
Unsupervised learning
Edge computing
a part of a distributed computing topology that has depth and where information processing is located close to the edge"
AI
ability of a machine to perform cognitive functions we associate with human minds, such as perceiving, reasoning, learning, interacting with the environment, problem solving, & even exercising creativity
Computer vision:
ability to extract meaning & intent out of visual elements - faces, objects, scenes & activities
Natural language processing:
ability to extract or generate meaning & intent from text in a readable, stylistically natural, & grammatically correct form
Operating AI as part of _____________ system is greatest barrier to AI success
automated decision-making
Ideally we would move to _________ instead of _________ work
behavior-based work instead of task-based work
Recommendation: Maximize business value by pairing ____________ w/near-term _________ - especially those that __________ human work, decisions, & interactions
business priorities, opportunities, augment
With edge computing, Processing can take place at any or all of the layers between the ______ and the __________
cloud (or central enterprise data center), edge
Recommendation: Leverage AI tech from __________ for integration, deployment, & infrastructure scalability
cloud vendors
Risks
cybersecurity vulnerabilities, making the wrong strategic decision based on AI, legal responsibility, failure in mission-critical context, regulatory non-compliance risk, erosion of customer trust from AI failures, & ethical risks
Recommendation: Secure _________ and _______ budget in advance & w/domain experts
deployment & infrastructure
Machine Learning
detect patterns & learn how to make predictions and recommendations by processing data & experiences, rather than by receiving explicit programming instruction; also adapt in response to new data & experiences to improve efficacy over time
Understanding customers in real time will become a challenge for organizations with _____________ and ___________ solutions
disparate data silos, static business intelligence/CRM
Displacement Effect
displacement effect — removing or reducing the need for certain types of jobs.
AI governance should be based on ________, _______, and ________
diversity, transparency, trust
As value of machine prediction goes down, value of human prediction goes _______while value of data, human judgment, and action goes __________.
down, up
Popular & easier paths
enterprise software w/AI, co-development w/partners, cloud-based AI, open-source dev tools, automated ML, data science modeling tools, & crowdsourced dev
Recommendation: Document __________ needs & put AI models in context
explain-ability
For the Discriminator, given the _______, tries to predict __________
features, labels
Recommendation: Create __________________ leveraging traditional algorithms 1st & then utilize DL if not good enough
flexible ML pipeline by
Issues companies are facing resulting in less return
implementation challenges, integrating AI into roles & functions, data issues, cost, lack of skills, & challenges in measuring & proving business value
Productivity Effect
increased demand for labor for nonautomated tasks
Industries investing low in AI and seeing high return
industrial products and services
For the Generator, given a certain ________, it tires to predict _______
label, features
Assisted Intelligence:
large-scale data programs, power of cloud, & science-based approaches to make data-driven business decisions, Cruise control
Intelligent interface:
latest in series of major tech transformations that began w/transition from mainframes to PCs w/emergence of web & mobile - more natural, contextual, & ubiquitous
Established techniques for supervised learning
linear and logistic regressions
The race between companies is to have access to _________ AND _________ data
meaningful, unique
City leaders are finding smart-city strategies start with ______, not ________
people, technology
Intelligent things are __________ things that go beyond the execution of __________ to interact more _________ with their surroundings and with people
physical, rigid programing models, naturally
Autonomous Intelligence
processes are digitized & automated to degree whereby machines, bots, & systems can directly act upon intelligence derived from them Self-driving cars
In supervised learning, Inputs are _________ and outputs are _________ and the algorithm is trained on the data. When algorithm is sufficiently accurate, it is applied to new datasets
provided, defined
Moving forward companies should
pursue execution excellence, address cybersecurity risks, apply AI beyond IT function, buy some off-the-shelf apps, staff wisely, decide where to automate vs augment
Quantom computing uses _______ to deal with superposition, entanglement, and interference
qubits
More specific techniques for supervised learning
random trees, AdaBoost, Gradient-boosting trees)
Understanding your customers in near _____ time will become the new norm over the coming ______ years
real, five
In reinforcement learning, the algorithm _____________ if the action maximizes the total amount of _________, allowing the algorithm to _______________ when more actions are pursued
receives a reward, rewards, correct itself over time
Used when the only way to learn from the environment is to interact with it
reinforcement learning
Needed Skills
researchers, software developers, data scientists, UX designers, change management experts, project managers, business leaders to interpret results, subject-matter experts
Foundational tech for Intelligent Interfaces
robotics, 5G networks, IoT, AI, digital reality, cloud & edge computing
The key of intelligent interfaces
seamlessly integrate the physical and digital worlds, the communication should go both ways
RNNs __________ and ______ the data that it identifies to assign the likelihood of the next item in the sequence
stores, combines
General AI (AGI)-
systems that have human level problem-solving across many different domains
T/F Early adopters need the right mix of _________, not just tech skills, to accelerate progress
talent
Industries investing heavily in AI and seeing high return
tech/media & entertainment/telecommunications & professional services
In Unsupervised learning, the algorithm infers structure from _______, identifying ____________________
the data, patterns with similar behaviors
We have always generated data but the last ________ decades have brought an explosion in both generation and storage
three
RNNs are used when dealing with ________ data or ____________ (like ___ or _____)
time series, sequences (like recordings or text)
The most profound disagreements are over two things:
timelines and dangers
CNN's are particularly good to deal with __________ data sets like ________
unstructured, images
AI has massive potential, but enterprise architecture & tech innovation leaders struggle to identify ___________; overwhelmed by number of available AI tech
value-adding cases
RNNs can be used in a variety of scenarios dealing with _________ and___________ and _________.
voice, speech recognition, patterns of transactions
Supervised Learning
• In supervised learning, an algorithm uses training data and feedback from humans to learn relationships
Even the most advanced smart cities have a long way to go, best cities only ____ of the way there
⅔
