Automation to Autonomous
Waves if AI
Introduction to AI01
Application of AI in various industries
Distribution of job spectrum
Historical context: Agricultural and industrial revolution
AI – Net gain or loss?03
Society & State response04
4. Artificial Intelligence
The use of technology to replace human effort
(physical & cognitive)
Automated –Replacing human effort on pre-
Autonomous – Replacing human effort (mostly
cognitive) by being able to take decisions based
on sensor inputs.
e.g. Loading/ unloading
e.g. Assembly plant
e.g. Waymo’s self driving car
5. Prof John McCarthy
Eliza - MIT
IBM – Language
DARPA – Grand
Google & Stanford:
Feline images –
Imagenet – MIT & Univ. of
Waymo's Self driving
Waves of AI
Addition of common sense and ethics
AI will be able to conceptualise and abstract things
May not require a lot of data points to learn.
Uses a lot of data for machine to extract the rules to learn by itself - (Image/Audio
Collect and prepare data to train machines.
Good in learning by perceiving but cannot do abstracting (A is a better actor than
Problem – Insufficient data or bad data can lead to wrong rules lead to bad
results. Even if 98% accuracy in certain conditions is not good (e.g.
Breaking a bigger problem to smaller problems & define rule for each smaller problem
Defining rule and mimic how humans think.
Problem – If a situation is not defined the system doesn’t work. Works in very narrow problem
set and it doesn’t learn.
8. Applications of AI
Aerospace & Defence
Facilitate ease of use of a product
AI acts as personal assistant facilitating actions of a
user while using a product.
Replacement of manual labour
Repetitive laborious tasks can be performed
by AI powered machines.
Reduction of human exposure in hazardous
With correct programming, Intelligent machines can
perform tasks more accurately than humans thereby
reducing the error rates and increasing the quality.
AI can handle massive computations and work out
various scenarios suggesting the best outcome in a
real quick time.
10. Job functions – Replacement by AI
Support tasks – Clerical data
work, office support, factory
workers, payroll, admin
assistants, general mechanics,
food preparation, cleaning,
drivers, mundane manual labour.
installation and repair
Computer engineers &
specialists, team managers
& executives, sales &
Customer care, food service,
safety and security workers,
Non – Repetitive,
IMF predicts 11% of jobs held by women are at the risk of
Financial services 50% workforce women but only 25% are
at senior positions which may be insulated from AI.
US: 85% of bank tellers are female.
WEF-Linkedin study - Only 22% of AI professionals are
Impact on women
11. AI has made previously
impossible tasks doable now
and opened up a spectrum of
new job functions.
New roles enabled by AI
• Machine Learning Engineers.
• Data Scientists and Data
• Research Scientists.
• Process Automation Specialists.
AI has reduced the cost of
production thereby making
products and services
cheaper further increasing the
demand and consumption.
Job functions – AI Fueling employment growth
AI has enabled better user
experience and enhanced
quality thereby further
increasing the demand and
Reduction in existing roles offset by
increase in job opportunities towards the
cognitive spectrum of job distribution.
• Information Security
• User Experience
• App Developers.
• Robotic Trainers.
• Technical Training Experts
13. Life long learning due to
changing nature of
Short term training
and functions will require
Encourage risk taking
behaviour and exploring
Redesign regulations and
laws for market.
New businesses coming up.
Ethics in AI
New strategy for
Review of schooling and
Society need to motivate
people to take risk and
think out of box.
Agile and flexible mindset.
Society & State Response
Science Technology Engineering Arts Mathematics
Education System (STEAM)
Good Afternoon – This is a presentation about Artificial Intelligence and its impact on employment and jobs.
I shall be covering my presentations under the following heads:
I will discuss the basics of AI, its developmental milestones and the waves of AI. Thereafter I shall cover applications of AI in various industries and the key functions that get enabled as a result of AI.
I will then discuss whether revolution in AI is a net gain or loss for the employment domain and how we should expect a fundamental shift in types of jobs in future?
Thereafter, I will cover the societal and state response towards AI revolution and finally will conclude with certain key takeaways.
Today, just like the word Cyber, Artificial Intelligence or AI is also a buzzword. It is about the use of technology to replace human effort which can be either, physical or cognitive in nature. While it is a buzz word, it is also sometimes the most misused word in discussions and debates. There are various levels of AI and here I have segregated them into four different aspects:
The key here is to understand the difference between automated and autonomous. Automated means replacing human effort in pre-defined tasks and autonomous means replacing human effort in the domain where decision making is involved based on certain sensory inputs.
With this distinction in mind, these are various levels of automations; from being semi-automated to being fully autonomous.
In the most basic form, an “If-then-Else” statement is also some sort of artificial intelligence but now with maturity in basics around AI, the bar has been raised to consider something as true AI or not. –
The term AI was coined by Prof John McCarthy of Dartmouth college in 1956 during a session on the subject of development of machines to be able to do tasks like human beings.
In 1966, ELIZA was developed at MIT that represented a text based chat bot meant to converse with computer using natural language processing. IT wasn’t capable of learning through these conversations but could converse through pattern matching.
In 1980, Dec system deployed a computer program for order processing called XCON to automatically select computer components based on customer requirements. This led to savings of millions od dollars and was the first example of commercial exploitation of artificial intelligence in some sense.
In 1988, IBM brought the concept of probability of outcomes into machine learning from the previous approach of being just rule based machine learning.
Internet came into existence in 1991 and that started onset of an explosion of digital data that could be used in artificial intelligence in future.
In 1997, using brute force mechanism Deep Blue defeated Gary Kasparov but it was not based on learning and adaptation and the victory was more symbolic rather than AI.
While in 2004, all the vehicles failed DARPA grand challenge. This was a100km race for autonomous vehicles but in 2005, 5 vehicles completed the race.
In 2012, Google and Stanford researches showcase unsupervised learning to identify pictures of Cat by machines.
In 2015, Researchers at MIT and university of Toronto showed that machines were better than humans in identifying handwritten characters using AI.
In 2016 Alphago created by Deep Mind defeated the Go champion Lee Sedolf in 5 matches using neural networks.
Last year we had Waymo (a google spin-off) launching a self-driving taxi service in Phoenix.
Therefore, AI has come a long way from being a rule based computational approach to neural network based machine learning approach.
Today, there is a lot of talk around AI replacing human beings or being able to take over critical decision-making positions.
Let’s look at the evolution and development of various stages in AI and where is it headed to in future. We can see three different waves of AI.
First one is about handcrafted knowledge. Experts broke down larger problems into smaller ones and computers then executed these rules. E.g of such system is a scheduling algorithm used in transportation systems for sequencing movements. The problem here was that if the system encounters a situation that is not defined, then the system would not be able to work. It cannot create a rule by itself and cannot perceive external inputs on its own, can’t learn or abstract things on its own.
Now the second wave is characterized by a lot of statistical learning. A lot of effort goes behind the scene in training the machines. It requires a lot of data that needs to be curated and cleansed. Therefore, in this wave, the systems are very good in perceiving and learning. Lot of applications like natural language processing, image recognition, audio recognition etc are part of this wave. The problem sometimes here is the insufficiency of data that can lead to wrong rules and incorrect learning. Although the algorithms are fairly accurate but sometimes even 98% accuracy is not enough. E.g. nobody likes an algorithm which is 98% accurate while facing a life and death situation.
This takes us to the third wave of AI where the systems will have underlying explanatory models that allow them to characterize real world. These systems will not require a lot of training data and will use these explanatory models to further learn and abstract higher order tasks.
The fourth wave will be about AI applications considering common sense and ethics.
Next we look at how AI is helping industries.
AI has slowly embedded itself into a lot of sectors for various applications. On the left side of the screen you will see the sectors where AI is being used and on the right-hand side, you will see various domains that are being affected by the use of AI.
Banks for example take help from predictive analytics while profiling a customer who has applied for loan. Data points from regulatory database are assessed against the predictive model to understand the creditworthiness of a customer and associated risk for the bank.
Ecommerce portals use AI to suggest various products to buyer looking at the past purchase profile and similar items bought by various buyers.
Autopilot in aircraft may not be that advanced version of AI but self-driving cars are using advanced version of AI to navigate through traffic and avoiding obstacles.
What essentially AI is doing here?
AI powered machines are replacing repetitive laborious tasks. They are also reducing human presence in hazardous zones by automating operations in such areas.
AI is predicting and prescribing output to facilitate faster decision making.
It is making a product easy to use and is acting as a customer’s personal assistant.
Finally, intelligent machines are performing tasks accurately thereby reducing errors in this process.
Now we will cover whether AI is a net gain or a loss to the employment domain?
The tasks highlighted on the slide from left to right represent the entire spectrum starting from repetitive and predictable tasks to Non- repetitive and ones that require cognitive effort each time they are performed. The repetitive tasks like clerical jobs are easier to be replaced by rule based computational approach and as we move towards the right side, the tasks are more and more customized to each situation and require a different response each time they are performed. They are more difficult to be done via AI algorithms. E.g. computer engineers, managers, artists, childcare professionals.
We had two analogous scenarios in the past where major revolutions led to employment changes. If you see the chart by McKinsey at bottom left hand side, you will see how employment of labour has decreased in agriculture and manufacturing and mining since 1850 to 2015. But simultaneously there are other sectors like trade, healthcare, education and construction that have increased employment.
Women may be at a disadvantage as far as job impact due to AI is concerned. IMF estimates that about 11% of jobs held by women are at a risk of elimination. In financial services women represent 50% of workforce but only 25% senior positions that are insulated from redundancy due to AI. In US, 85% of bank tellers are females. A study by world economic forum and Linkedin puts women population in AI jobs to be at just 22% which means that they may miss the upside of AI.
Now let’s see how AI will be fueling job growth.
AI has enabled new opportunities, as now we can perform the tasks, that were previously impossible to think off. This has led to opening of new types of roles which were nor present few years back.Some of them being, AI engineers, data scientists, robotic trainers, app developers etc.
The other two effects of AI are reduction in cost and ease of use of products and services and that has increased the demand and inturn the consumption. This has increased the quantum of job opportunities and has offsetted the reduction in jobs by more jobs on the cognitive side of job spectrum.
An example can be seen by the chart on bottom right for the percentage of queries received on coding website stackoverflow where Python has gradually moved to the top of the most queried language and c# has moved down between 2009 and 2018.
Now let’s look at societal and state response to this AI revolution in to enable people to reap full benefits of this AI revolution.
The concept of learning needs to change to lifelong learning and continuous short terms training programs. The education system may also need review that includes formal schooling and college education. New disciplines and subjects need to be made part of education. The subjects of Science, Technology, Engg, Arts and Maths need to be followed more by the population.
Society needs to encourage risk taking capability of people, agility, flexibility and an entrepreneurial mindset for people to be able to survive this wave of AI. Comfort and stagnation in mindset is a sure shot recipe for unemployment.
State would need to review legislations around new businesses, redesign regulations for the market, code and ethics of AI need to be formulated.
Different economies have different distribution of type of jobs and labour rates across job spectrum. Therefore, if job losses happen due to repetitive and mundane jobs being taken over by AI, then there will be more job displaced due to automation in emerging economies like China and India where the jobs are more tilted towards more of manufacturing, agriculture and services sector. McKinsey observes that between 2016-2030, a total of 75-375 million people may need to switch occupational categories. The size of the circle on the graph shows the number of job displaced for each country during this period.
As previously described, the impact of AI will be like agricultural and Industrial revolution but this will be at a much rapid pace as we are already ahead in the technology lifecycle as compared to 19th and 20th century.
Less cognitive jobs will be replaced first followed by the ones that require a different human effort each time. But by the time 3rd and 4th wave of AI comes; more cognitive jobs will also be affected.
Society, State and corporations need to change the status quo and adapt towards this changing job scenario. Education and careers in STEAM (i.e. Science Technology Engg Arts and Maths) will be more resilient to the shock of AI in this domain.
While job losses may happen AI will also enable higher productivity and lowering of cost enabling more businesses to flourish.
Last but not the least, AI and machines will always complement human effort but may not take over critical applications and replace humans in such scenarios.
With this I conclude my presentation. I will be happy to answer any queries on it.