Depending on what news headline you have read, you may have perceived an Artificial Intelligence (AI) system as either an Alexa or Siri assistant that understands all your commands, a deep learning system that can recognize dog or a cat from image, a system that recommends personalized medicine, or an intelligent, overpowering machine that can overtake all human tasks and render humans useless. Few of these definitions can be termed as visionary, few fear mongering and rest of them being evolutionary.
Last month, I was at Artificial Intelligence (AI) Summit 2018 in San Francisco. The event highlighted the state of AI business as it stands today and real-world examples from enterprises who are using AI to transform their business. I want to give a shout out to organizers of AI Summit – they did a fabulous job in bringing a highly diverse set of speakers across a variety of verticals. I will attempt to capture my takeaways from the conference with different narratives, infusing with some of my thoughts and experience in this space.
What is AI and Why it matters
You might have heard about this saying from Eric Raymond on what we should expect from a machine –
“ A computer should never ask the user for any information that it can auto-detect, copy or deduce.“
If you use this definition of AI and connect this with innovations in last few decades, you will realize that AI is nothing but an evolving science of computing that has been around in some form or other. With the advent of Information Revolution, our collective minds have changed how we work, live and interact with our ecosystem. Software and Hardware assets across both consumers and enterprises alike have made human lives easier in many aspects and allowed humans to focus on the higher order of things.
So, what is different now? Two striking differences:
1. We are in a different phase of this computing world where Cloud, Mobile, and IoT have accelerated the pace of information processing
2. Fueling on the above, we are automating more of complex decision-making tasks
Take an example of ride-sharing: Uber solved a complex decision-making problem of how to bring a car to a consumer at the spur of a moment. The disruption came about because of mass adoption of Mobile devices, coupled with super-powered Cloud-based applications. Beyond bringing a car, Uber has ushered into Uber Eats and other services. Uber Eats has become a $6B run rate business as reported last quarter. These AI services, in turn, bring a variety of enterprises under the AI value chain. Hyper-personalization could emerge soon to improve the consumer experience. This could include a personalized order of your favorite food and setting the ambiance with your like of songs while you take the ride.
Benedict Evans (investor @Andreessen Horowitz) talked at the summit on why AI matters. AI has played an integral role for recent innovations, including Uber/Lyft ride-sharing, Instacart, and Instagram. Although these innovations are solving today’s problems, it may open and lead us to a totally new set of innovations in the future. Ben gave an example of how Mobile, which we once thought as a simple utility in the past for Email, Stocks, and Weather are now opening additional frontiers unthought of. A self-driving vehicle is a classic example of a future innovation S-curve on top of the ride-sharing business model that Uber and Lyft are venturing into. Consider self-driving vehicle as a behemoth IoT use case where tons of sensors, actuators and motion events are analyzed to arrive at a real-time decision using onboard AI algorithms and Cloud acting as a data store for continuous deep learning.
Value Generation and Adoption
Decision making is an integral part of any business and naturally, any capability that helps in moving the needle by using automation/ data-driven insights is going to bring great value. According to Gartner, AI-derived business value is forecast to reach $3.9 trillion in 2022. McKinsey, on the other hand, predicts that AI-generated business value in next few years could be in the range of $2.5 trillion to $5.8 trillion. We will discuss real-world examples of how enterprises are leveraging the technology in my next article.
Like any other new technology adoption, enterprises are taking time to adopt and fully embrace AI. While business case and stakeholder buy-in remain a critical factor for any technology adoption, enterprises will also need a well-established and a mature big data setup on top of which they can experiment with the AI models. Depending on which survey you have read and the kind of AI use cases being talked about, the adoption numbers vary from one research to another. The good part is that AI is front and center of C-suite discussions. According to the Economist Intelligence Unit, approximately 75% of 203 executives surveyed claimed that AI will be actively implemented in their organizations in the next three years.
How can enterprise become successful with their AI initiatives
This poses a question: How can an enterprise be successful in harvesting the outcomes promised by AI? Here are my top three takeaways:
- Clear articulation of Outcomes: First and foremost, identify and clearly articulate what outcomes your organization is trying to solve with AI. Presenting a story and getting buy-in from wider stakeholders is a key to success. Investment in AI because of its coolness will not lead to a better outcome, and perhaps would lead to negative outcomes. I heard this over and over again from many enterprises at AI Summit
- AI is more than models: Expect to spend the time to collect/prepare the right data and ensure that appropriate processes are in place before AI models are churned out. The ability to shorten time-to-market for a new model from experimentation to production may require close interlock between stakeholders. Leverage the likes of AIOps or MLOps that defines a disciplined tooling and approach around this. ParallelM, one of the startups in this space, helps enterprises to offload the automation, scaling and optimization aspects of running ML operations in production.
Gil Arditi, Head of Products, ML @Lyft acknowledged that the hardest part for them was to solidify the data architecture and AIOps processes. AIOps focuses on process interlock across Data Science, Software development and Operations teams. A new model from the Data Science team may work well in his setup but may not scale in production or may produce different outcome in production based on real data. After considerable effort, the AIOps at Lyft has been fully operational. The effort and time have now paid off, says Gil. Lyft ML team will be able to devote 100% of on model experimentation going forth.
3. AI Bias & Human touch: AI models are as mature as the data and decision thresholds fed to it. Automated decisions do require some form of assumptions, and there could be a thin line between assumptions and bias. Eliminating bias is a harder problem to solve. Many organizations believed that having a human-assisted approach right from defining an outcome to its realization works best, at least in the initial stages of AI maturity curve
Here is a recent example that highlights what happened when Boston Public Schools tried to solve their problem purely using algorithms. Last year, Boston Public School officials asked MIT students to build an algorithm to reconfigure school timings that would help high school students to improve their health and academic performance and the school to optimize on buses. The algorithm rightly did its job but failed to evoke a favorable response from most parents as it recommended to start elementary schools by 7:15 AM instead of 9:30 AM. The parents revolted as it would require them to change their schedules for family time and work by a great deal.
Another example where Amazon had to scrap its AI recruiting tool, because the training data seemingly introduced bias against women.
Is AI to fear?
Among the promise of AI also lies a fear that AI is going to bring mass unemployment. If you look back at the past ten thousand years or so, the kind of activities that humans perform has evolved across Agricultural and Industrial revolutions. These revolutions automated a lot of activities and have helped transform the human lives and the ecosystem around us. At the same time, the kind of activities in Agricultural revolution didn’t exist in Industrial, and same is the case when we moved from Industrial to Information Revolution at the end of 19th century. Humans have shown tremendous ability to re-skill itself at every phase of this transformation.
AI as a growth component of this Information Revolution is going to do something similar and our work will definitely get transformed, but not lost. As it happened in the past, the tasks that are better suited for automation (such as storing and processing large volumes of data) will be offloaded to machines, while humans will be freed to focus on the higher order of problems that are unsolved yet. And these tasks could take the shape of a voice assistant, a computer vision system to detect objects, an autonomous vehicle that is always alert to obstacles around itself, an automated decision support system that can forecast supply and demand, predict churn or a myriad of other data-intensive tasks within and outside of an enterprise.
Stay tuned for a follow-up article where I will highlight how some of the enterprises are using AI to transform their business. Please drop a note on the article and/or comment on how your organization is transforming itself using AI.
P.S. This is a repost from my LinkedIn article
from Featured Blog Posts – Data Science Central https://ift.tt/2PKSxU6