Exclusive Interview with Stephen Brobst
Stephen Brobst is an advisory board member at Analytics Private Limited (APL) (analytics.com.pk) and the Chief Technology Officer of Teradata Corporation.
He is widely regarded as a leading expert in data warehousing and artificial intelligence.
He holds a Ph.D. from MIT and an MBA from Harvard Business School and is an author of numerous journals and papers.
Stephen has served in President Obama’s Innovation and Technology Advisory Committee (PITAC) and was ranked the number four U.S. Chief Technology Officer by ExecRank. Stephen is a regular visitor to Pakistan and enjoys the natural beauty of our northern areas.
All technologies go through a maturity cycle and many don’t make it past their initial hype. How do you see AI in this context?
AI has survived through multiple cycles of hype and disillusionment. Most of the ideas that we associate with AI today have been around for more than 50 years. The term “AI Winter” was coined to describe the first cycle of disillusionment with AI that occurred in the 1970s. This period was marked by a disappearance of funding and public call-outs of over-promise and under-delivery. During this period the UK Parliament commissioned a study by professor Sir James Lighthill from the University of Manchester and he chastised practitioners for “grandiose objectives” and lack of delivery.
A second AI hype cycle related to the development of “expert systems” occurred in the 1980s and then crashed in the 1990s due to unreasonably high expectations related to these emerging technologies. The hype cycle that we are currently experiencing started in around 2012 and is based on huge advances in both hardware and software technologies to enable machine learning and deep learning. Some technologies never emerge from the trough of disillusionment as they fall to their death from the peak of over-inflated expectations. However, AI has continued to evolve and re-invent itself many times over the past 50+ years and is here to stay.
Being the trusted advisor to the top tier organizations in Pakistan as well, what level of adoption and success of AI do you see in Pakistan?
The adoption of AI in Pakistan is highly variable across industries and even within an industry. In general, deployment of AI will be most aggressive where competition is the most intense. Competition has clearly driven early adoption of machine learning in the telecommunications industry in Pakistan for purposes such as churn prediction, top-up offer optimization, device up-selling, and so on. The financial services industry in Pakistan has generally been more conservative in AI adoption versus the rest of the world. Areas such as risk scoring, behavioral segmentation, and fraud detection are widely deployed with machine learning in the United States and Europe, but are still in pilot stages in many of the Pakistani banks.
At the same time there are AI start-ups launching all over the country and incubation programs in collaboration with some of the top universities are in high gear. The government in Punjab has done some very interesting things with analytic solutions in areas such as security and disease management- AI adoption is going to be the next logical step in their journey of data driven governance.
Should organizations be focusing on AI adoption today and can they drive actual business value from it?
The answer is “it depends”. Not all organizations benefit equally from the deployment of AI. To obtain value from AI, an organization must have access to detailed data to allow proper training of models specific to its business. Organizations that have instrumented their operations to collect detailed data and whom make lots complex decisions on a daily basis are good candidates to create value from AI solutions. Areas that are particularly well aligned to high value creation from AI solutions include recommendation engines, demand prediction, fraud detection and prevention, and optimization of asset maintenance.
What do organizations need to do to be ready for AI adoption? Are there any specific skillsets, processes or hierarchies that are needed?
The most important thing that an organization must do to be ready for AI adoption is put into place a strong infrastructure for data management. Big data is the fuel for AI. Data must be collected at the most detailed level possible and organized for efficient access by data scientists. Without data, AI algorithms are nothing more than mathematical equations in a text book. The most successful AI projects are narrow in scope and driven by a strong business sponsor rather than by technologists.
Do not try to solve “big” problems for which there is no specific definition of success – focus on targeted use cases, especially those that relate to operational decision-making. Buying a “tool” is not enough; organizations need to invest in data scientists and data engineers so that skill sets are in place to ensure that feature engineering, model training, and deployment of AI models is executed in a way that will bring positive business outcomes.
Are the potential benefits of AI limited to the corporate world or can other segments like Governments also benefit from AI?
Governments have always been big investors and also big beneficiaries from AI technologies. Governments typically have lots of complex data in situations where AI provides huge value. Some of the most important areas where AI has benefited government programs include improving quality and cost of national security, healthcare, energy production and distribution, education, and transportation. For example, using AI software as “Augmented Intelligence” can help doctors be more effective in diagnosing and prescribing interventions to treat patient ailments. Note that the AI software does not replace the doctors, but rather makes them more effective. Many studies have shown that doctors and AI software perform better when working together than either one can deliver in isolation.
Many people believe that with the advancement of AI a lot of human jobs will be lost. Fact or Fiction?
This is absolutely fact. Many jobs that humans do today will be replaced by AI software. Driverless cars powered by AI software will make human taxi drivers largely obsolete. Robotic process automation (RPA) will replace human jobs related to paper-based workflows that are more efficiently (better accuracy, lower cost) executed with AI software.
There are many more such examples ranging from the automation of manufacturing jobs to customer service to insurance underwriting where AI software will eliminate or reduce the number of humans required to accomplish important tasks. However, AI will also create many new jobs – but with different skill set requirements. Humans with skill sets related to data management, mathematics and statistics, software engineering, and so on will be in very high demand.
What do you believe are the biggest deterrents for AI adoption?
Fear and lack of understanding. IDC has predicted that here will be a backlash from consumers to prevent the use of AI technologies due to lack of transparency. Because deep learning techniques (multi-layer neural networks, in particular) behave as a “black box” for decision-making in many applications, the lack of transparency creates fear among the persons impacted by such AI software. When people do not understand how a decision is made, there will be fears regarding bias in those decisions.
Bias in AI solutions is illegal in laws and policies put into place across the globe – such as GDPR (EU), CCPA (California), Algorithmic Accountability Bill (New York City), Personal Data Bill (India), Personal Data Protection Bill (Pakistan), and many more. However, enforceability of such laws is dubious at best. Some have proposed the use of AI software to detect bias in AI software. But how will we check for bias in the AI software that is supposed to check for bias in other AI software? The solution becomes circular. Moreover, humans
will be reticent to take actions based on the recommendation of AI software without explanations of “why” the actions that they are taking are the right ones – especially in areas where the actions can impact the health or well-being of another human. Investment in eXplainability in AI (XAI) is critical for adoption. The development of explainability algorithms, such as Shapley, is a very active area of research for the industry.
Are there real-life examples today where AI is making a difference?
I can think of many, many such examples. One case that I was involved with was the deployment of machine learning on top of sensor data taken from the operation of very expensive machinery where the company was spending lots of money every year on maintenance and parts replacement according to milestone-based care after a pre-defined number of hours of use for its assets. When we replaced this approach with condition-based maintenance we not only reduced maintenance and repair costs by over 40%, we also reduced the number of unplanned outages (which are very bad for their operations). Another prolific use of AI is in product recommendation engines where conversion rates are at least 500% better with machine learning versus simple rule-based recommendations.
How do you see AI evolving over the next 5 years?
Right now AI is still in the early adopter phase. Most industries are exploring the possibilities, but are not yet committed to the technology. Within five to ten years I believe that we will evolve from early adopter and bleeding edge deployments to mainstream adoption of AI technologies. AI will no longer be a special project in an advanced development laboratory, but will be infused across all industries and job functions. In some ways, AI will become less visible than it is today because it will be more deeply embedded into all software solutions that we deploy rather than being treated as something “special”.
Can you share any interesting experiences you’ve had in Pakistan?
I first came to Pakistan for work purposes 20 years ago. I keep coming back because of the beauty of the people, culture, and landscapes. Over the years I have made lots of friends throughout the country. Some of the most amazing experiences I have had are with a group of friends who make an annual trip into remote areas of northern Pakistan. We are self-named as “The Northern Overlanders” and I am very grateful to be accepted into the crew as the only non-Pakistani participant in these treks. We camp in pristine nature next to beautiful rivers and mountains (always leaving no trace), hike onto glaciers, take 4x4s where there are no roads, and meet amazing local people in small villages that rarely get outside visitors.
What is your view on Pakistan’s tech talent? Where do you see strengths and potential areas to be worked on?
Tech talent in Pakistan is amazing. As CTO of Teradata I worked with professors at a dozen universities in Pakistan back in 2001 to develop a course curriculum aligned to data science and analytics for both computer engineering and management students. This investment created a pipeline of great talent that Teradata used to create a Global Delivery Center in Pakistan. We started with one location in Islamabad and later expanded to Lahore to increase our capacity.
While many companies have focused on India and China for offshore resources, we have diversified ourselves into Pakistan with great success in hiring and retaining top notch talent to serve our customers in Pakistan as well as the rest of the world. I have also had the privilege to serve in an advisory role to many start-ups coming out of Pakistan, such as Analytics, TenX.ai, and others. I have consistently been impressed with the ambition, innovation, and technical expertise coming from Pakistan – but, I must admit, I am a bit biased by my wonderful experiences in the country.