Recently, I was invited to speak at the AI Squared Forum which was held Last Sunday(September 22nd 2019) at the Bahen Centre for Information Technology at the University of Toronto. I spoke on the topic of “Potential Impact of AI on Jobs: A View into the Telecom Industry”. This blog entry is based on that talk.
Good Afternoon! If you look at the business press, there are screaming headlines about robots taking away jobs. There have been multiple reports and estimates on the potential impact of AI on jobs. At the same time Telecom operators are facing multiple pressures and are unprepared for major changes. This creates a situation for significant disruption of the telecom industry by AI adoption particularly when it comes to jobs. We don’t want to be caught unprepared. Every one of us in the telecom industry or any other industry can take actions which will help us prepare and benefit from the adoption of AI and automation. I am Vijay Dakshinamoorthy and in this talk I will present some insights into some of the impacts of AI on jobs, with a focus on the telecom industry.
Before my current role, for most of my career I have worked in the telecom industry. When I look back at this experience, I see that there has been periodic restructuring and turn over in the industry whether it was on the consulting side or on the client side. There have been job losses as well as job growth. So with this current cycle of AI adoption and automation what will be the impact on jobs in telecom?
Here is a brief outline for the talk. We will begin by looking at jobs and AI. Specifically, we’ll look at job classes in telecom. We’ll then look at the estimates and projections for the AI impact on jobs. We’ll look at overall estimates as well as projections and trends in telecom. Next, we’ll look at a couple of case studies in areas that are prime candidates for AI in telecom. We’ll then look at the jobs of the future and how job redesign should be undertaken along with AI adoption. I will then leave you with my calls to action and conclusion.
I believe that a good portion of this audience has or will work in the telecom sector. I also assume that there will be three groups in this audience. One group that is comprised of graduate students, academics and researchers, another group that is comprised of entrepreneurs, executives or managers and the final group that is comprised of innovators who are building or will build AI products. Let us see how each of these groups can shape the coming impact of AI on jobs.
AI and Jobs
What can AI automate? AI can perform physical tasks through fully autonomous robots or vehicles. Of course it would have to be a narrow set of tasks. Admin tasks and decisions as well as business process management can be automated through robotic process automation and rules engines. AI can digest words and images through application of natural language processing and image recognition. And finally, it can analyze numbers and make predictions by the use of neural networks and machine learning including operational analytics and scoring. At a fundamental level though AI provides prediction and automation of certain tasks.
Jobs in Telecom
Having looked at the tasks that AI can automate, let’s look at the job classifications in telecom. For this I looked at all the jobs and roles in telecom according to the Bureau of Labor Statistics in the U.S. and grouped them into 5 broad categories. The 5 broad categories are Technicians and Frontline, Engineers and Analysts, Middle Management, Others and Human Resources, and these are the percentages of jobs and positions that fall into those categories. Almost all of the classes here are vulnerable to automation. I have also listed some of the tasks that could be automated with current AI technologies or in the near future. But there are opportunities to get ahead of the curve and identify opportunities for job growth.
The dreams of yesterday are the hopes of today and the reality of tomorrow. Science has not yet mastered prophecy. We predict too much for the next year and yet far too little for the next ten.
While AI could lead to the automation of some of the routine jobs which involves mechanical calculation or data entry or other clerical work, it could lead to a growth in demand for tasks and jobs that are complementary to the adoption of AI. Decision making is fundamental to all the job classes listed here. However, current day AI and the automations listed here are one component of the decision making namely prediction. As outlined in the book Prediction Machines, decision making involves other components including judgment, action, outcome and three types of data (input, training and feedback). According to the book, having better predictions and more predictions increases the returns to complementary activities such as judgment. So in each of these job categories, there will be opportunities to exercise judgement and decision making. For instance any of the decisions that human resources personnel take such as a candidate hire or middle management makes such as capital allocation involves payoffs and a careful analysis and judgment to come up with the right decision. So there will be increase in demand for the higher value tasks and jobs that involve judgment and decision making.
Impact Projections
Let us now look at the impact projections and employment statistics. When we look at studies estimating the impact of AI on jobs, some are more pessimistic, and some are optimistic. On the pessimistic side there were estimates from the Oxford economists that about 47% of jobs in the U.S. are automatable. On the optimistic side, the World Economic Forum study estimates that there will be 133M new jobs worldwide by 2022. In any case, the net consensus is that there will be a displacement from lower skilled jobs to higher skilled jobs.
Talk on Potential Impacts of AI on Jobs at the AI Squared Forum at the Bahen Centre of Information Technology at University of Toronto
Here are the BLS projections specific to Telecom. According to the BLS projections about 100,000 telecom jobs will disappear in the wired telecom category by 2026. On the other hand only about 5,000 jobs will be added in the wireless category. However, these job projections are based on a multitude of factors not just AI. These factors include the maturing of the industry, pressure on EBITDA and ARPU and competition from OTTs.
AI Impact on Economies
When we look at the impact of AI on the world economy, it is predicted that AI will generate about $13-15T in economic output as contribution to the world GDP. Of this it is projected that about 70% will be accrued to the AI superpowers U.S. and China. According to conventional wisdom, since U.S. is more reliant on service oriented and white collar jobs the impact of AI on jobs would be lesser whereas because of China’s reliance on manufacturing jobs the impact will be greater. Dr. KAI-FU LEE who is the chairman and CEO of Sinovation ventures in China, and is a former AI researcher who has worked with the top tech companies has a different take. In his book AI Superpowers – China, Silicon Valley and the New World Order, he describes the advances in AI, growth of AI tech companies and startups, impacts that AI will have on jobs and contrasts the impact in the U.S. with the impact in China. According to him, the impact of automation on jobs in China will be slower and arrive later in China than in the U.S. He cite’s Moravec’s paradox which suggests that AI is good at succeeding at intelligence and computational tasks but it is difficult to get it to do tasks that require perception and mobility i.e., to get it to be good with its fingers.
Employment Statistics from Major Carriers
Let us now look at the employment numbers for the major carriers. For the most part the employee count in the major U.S. carriers has been flat or declining. The upward trend in AT&T is due to the takeover of Time Warner in 2018 but even then there was a reduction of about 11K in headcount in the first nine months of 2018. AT&T also found that nearly half of its 250,000 employees lacked the skills needed to keep the company competitive. So it launched a massive $1B web-based training program in partnership with Coursera, Udacity and leading universities. Verizon is in the process of reducing headcount by about 10,400 jobs this year. The original voluntary buyout deal was actually sent to about 44,000 employees. The reduction will save about $10B for the firm which it plans to deploy in building transmission towers for 5G. The reductions are happening in the content and cable area.
When we look at the major Chinese carriers, only China Mobile has grown significantly in terms of employees and it is worthwhile to note that it also entered into the wireline communication market recently. Starting from 2014, China Mobile converted its contractor/dispatched workers into formal employees on a massive scale, which caused a significant jump of its total number of employees, followed by a growth plateau when the conversion was completed.
When we look at the Canadian telecom providers, TELUS has been growing because of its investment in broadband network and other business lines. There has been drop in employee counts for both Bell and Rogers. Again this could be because of any number of factors such as industry maturing or declining, pressure on EBITDA and ARPU as well as competition from OTT etc.
Case Studies of human AI collaboration
Having looked at jobs and AI and impact projections and employment statistics, we can look at two case studies where AI is being adopted in a telco context. Let us look at the network operations center. As you might know, the Network Operations Center or NOC is essential to monitor and manage the network. It keeps track of events on the network such as disruptions or breakdowns and high bandwidth usage etc. According to Tractica, network operations monitoring and management is likely to account for 61% of telecom AI spending in the coming years. Elisa Finland which is a telecommunications and digital service provider is a pioneer in this space. It uses AI and ML along with advanced analytics to run self-optimized, self-healing and autonomous transport networks and can facilitate enhanced business agility, superior scalability and optimized performance. Such cognitive NOCs provide correlated alarms, alarm and event prediction, automation of NOC tasks, automated inventory and network discovery and chatbots and task bots. The goal is to apply AI to help NOCs achieve operational maturity by moving from present day reactive mode to proactive, predictive and ultimately automation based mode. This is also an opportunity to make the NOCs adopt customer centricity to enhance customer experience and prevent service degradation. It has completely automated its network operations where 90% of network incidents are resolved automatically and as much as 75% are proactively corrected before they become visible to customers. They have also involved the people in the design of the automation and in the daily practices and by employing the Elisa academy they were able to bring along the people in the transformation.
Another example is in call centers. T-Mobile is humanizing the customer experience. Instead of connecting a customer to a chatbot or IVR, T-Mobile is connecting them to customer service agents who are equipped with the information relevant to the customer’s call. They are using Natural Language Understanding and machine learning to comb through hundreds of thousands of messages and knowledge repositories to get to the information most relevant to the call. The customer service agents are also part of a team of experts who know their customers very well so that T-Mobile is able to offer personalized service to the customers. This is a clear example of humans and AI working together to improve the service to customers. Additionally, instead of hand labeling the thousands of messages from customers, T-Mobile is using Amazon Sagemaker’s Ground truth to label the messages. Ground truth consists of a machine learning component where it learns from the annotations in real time and automatically applies labels to the remaining dataset. Ground truth also has a human component where a pool of humans (like Mechanical Turk) label a portion of the dataset and the machine learning component learns from that. You can learn more about this story on the T-Mobile site or on AWS site.
Job Redesign and AI Adoption
Next, let’s look at job redesign and AI adoption. According to an Accenture global study, new human roles and positions will emerge as automation increases. These roles namely trainers, explainers and sustainers will have an important role in the successful deployment and maintenance of AI solutions. Trainers are humans training AI for performance. The tasks here might involve having the machine observe decision making, redesign processes in the workplace and correct errors. Explainers are humans making AI explainable. This might involve making the algorithms explainable and interpreting the machine insights and outputs. Sustainers are humans making AI sustainable. These roles might involve making sure the AI actions and decisions are ethical and fair, making sure objectives are aligned etc. For instance, when we look at the Cognitive NOC of the future, trainers will be necessary for training the chatbots and task bots on the workings of the NOC. Explainers will be required to clarify the decisions made by the cognitive network and the reasons for why the NOC classified certain alarms in a particular way. Finally sustainers will be necessary for the ongoing maintenance of the NOC and in ensuring the right actions are taken when customer impacting events occur. So there is an opportunity to leverage AI in the NOCs to benefit companies, customers and employees.
I believe that AI adoption and job redesign will be an iterative process. Particularly when it comes to moving people to higher value jobs and tasks it involves a number of steps where the organization will need to review and revisit some of the earlier assumptions and observations. By identifying higher value tasks and redesigning jobs accordingly and adopting AI to augment humans, organizations can be successful in their AI adoption and in moving people to higher value tasks. We can start by identifying tasks that involve decision making and require judgment in each job class. We can then redesign the job to incorporate the human component and the machine component. We might create new job descriptions as a result. We can then train and reskill employees. We can then go about adopting AI for the machine component of the tasks and orient people for the human component of the tasks. Finally, review the original assumptions and observations and the value equation and iterate.
Conclusion
So what are some actions we can take. Identify the higher value tasks – tasks suitable for the trainers, explainers and sustainers in your industry and organization so that we can move people to those tasks along with the adoption and implementation of AI. Studies show that there is value to be uncovered in terms of 5-12% increase in profits and a similar increase in employment in telecom alone.
As an entrepreneur, executive or manager – identify higher value tasks in your organization or enterprise and begin training people. In addition to the tasks that fit the categories of Trainers, Explainers or Sustainers identify tasks and jobs that involve empathy and communication, critical thinking, creativity, strategy, imagination and vision. If you want a resource for understanding this topic more read the World Economic Forum report titled The Future of Jobs. As a graduate student or academic seek opportunities to train others on the cutting edge techniques and methods that you are learning or uncovering. Also as an individual seek opportunities to learn about AI and the skills for the future through online MOOCs or other avenues. If you want to learn the skills in demand and benchmark against the world and competition, take a look at the Coursera Global Skills Index report produced from a set of 38M learners. As an innovator or inventor find ways to come up with ideas and products that serve organizations, customers, employees and communities. And also ask and encourage government to come up with innovative solutions to mitigate the societal impacts.
In conclusion, I would like to leave you with a quote from Dr. KAI-FU LEE’s book. In the book, he describes his cancer diagnosis and how that changed his perspective on life and his outlook for the role of AI. He concludes his book saying, “Let us choose to let machines be machines, and let humans be humans. Let us choose to simply use our machines and more importantly, to love one another”.
I think that is an ideal we can all hope for in the future.