2021 will see businesses embrace AI, not because they want to, but because they have to.
Many years from now, 2020 would be seen as a watershed moment in AI. In the unprecedented year when COVID-19 ravaged the world, a clear answer has emerged about the usefulness of AI — enterprises need to find a way to creatively, broadly, and boldly apply AI to emerge stronger in the short-term and survive in the long-term.
A clear pattern of growth has already emerged in AI: in 2018–19, the phase of experimentation became mature; in 2020, adoptions began in a serious way and suddenly, COVID-19 gave the business leaders an opportunity and impetus to push automation and AI. In 2021, the fallout from a second wave of COVID-19 in the UK and many other countries will eventually become clear, starting with the rapid decline of many traditional, non-digital businesses. As the C-suite takes notice, following are the relevant trends I expect to emerge in 2021:
- Businesses will measure “Real” AI metrics
In 2020 businesses leaped out of experimentation mode, and in a post-COVID world, ground themselves in reality to accelerate adoption. In 2021, they will adopt the business outcome of AI initiatives in terms of traditional metrics such as revenue, churn, customer loyalty etc. While the business will have the responsibility to understand the impact of AI, the onus is on us, the data scientists, to set up the “translation table” to enable the same.
Data scientists evaluate the “success” of their projects in terms of Accuracy, Precision and Recall, F1 Score, AUC, or similar “science” metrics, which is very different from how businesses measure effectiveness of programs. To bridge the gap, data scientists must sit down, preferably a priori, with the business and work on a common framework to understand and measure the impact of their work.
For example, at Airtel, my employer and one of the top telecommunication companies of the world, we use the following simple table to share the outcomes with the product groups and business:
Once these “translation tables” are set up, it’s easy for all the data scientists, product experts, and relevant businesses to focus on the same goal and measure impact.
- Workplace AI and Man-Machine Collaboration will Accelerate
We saw an amusing video of Boston Dynamics robots dancing to the tune of, pick your choice of, Western, Hollywood, Indian, etc. tunes. Besides the social media frenzy of this wildly popular video, there is a long-standing trend on Workplace AI that is emerging and will boost automation and augmentation needs sooner rather than later. The post-COVID world would be more virtual, with workplace disruption for both location-based, physical, or human-touch workers and knowledge workers working from home. It will also be increasingly touch-less, especially in B2C environments such as retail, hospitality, transportation, food and beverage services, and more, leading to the next prediction…
- Computer Vision will be the “Next Frontier”
In the middle of the second wave of the pandemic making rounds globally, a little-noticed announcement made in December 2020 skipped most people’s attention: that Amazon plans to roll out tools to monitor factory workers and machines. Called AWS Panorama, it uses computer vision to analyse CCTV camera footages within facilities, automatically detecting safety and compliance issues such as workers not wearing PPE, or vehicles being driven in unauthorised areas. While this sounds trivial, given that a large industrial warehouse (250,000 sq. ft 0r 25,000 sq. mt.) can have 500–600 CCTV cameras operating at 60 fps (frames per second) — generating about 43 to 52 million images per day, about 1.5 billion per month.
A new generation of technology invented in the past few years makes this possible; federated learning platforms such as S20.AI, which is focused on industrial computer vision or Owkin, focused on medical research data and images, makes it possible to process and make sense of these data in a privacy preserved manner.
This has been aided by rapid growth in hardware capabilities. NVIDIA’s Jetson family of GPUs along with its EGX AI platform open up huge opportunities for computer vision and edge computing for IoT applications. At the same time, NVIDIA A100 GPUs have significantly improved processing heavy computer vision workloads in the cloud with multi-instance capabilities that can run up to seven jobs in parallel on a single GPU. Apple’s latest M1 chips have also shown 3.9x faster video processing and 7.1x faster image processing results.
As AI technology pervades every sphere of our lives, however, it has a significant impact on the future of the economies in the post-pandemic world.
- The growth of the “K Economy”, driven by AI, will Accelerate
In 2021, the re-building of the ravaged economies will begin, and growth in parts of the world that have recovered from COVID the earliest, such as India and China, would be the fastest. 2021 will be the beginning of the “K economy”, loosely defined by the growing difference between the performances of the economies and companies that digitise and drive faster AI growth. Grittiest among the companies will push AI to new frontiers, for remote collaboration, on-demand manufacturing, and move onto intelligent experiments on digital experiences, recommendations, etc. on the edge.
Given that AI tools have been significantly democratized over the past two years with top-notch algorithms being available quickly after they are developed in Stanford, Google, or China, laggards still have a possibly one-time opportunity to aggressively implement their top AI 10, 20, or 50 use cases, possibly with no-code AutoML tools.
As the AI models start to impact many aspects of our lives,…
- AI models would have to be assured for trust and fairness, quickly
As AI permeates our lives, the systems have to be fair, accountable and reliably reproduce results. Consider an AI system that credit scores unbanked customers, based on “alternate” credit data. Such scoring models rely on access to, for example, gig-economy paychecks, usage and repayment of micro-loans, social profiles, usage of smartphones (calls, data), shopping on eCommerce sites, to name a few. While these products are created with the noble intention of bringing millions of people under a formal credit, both the lenders and users have to trust the system for the products to be effective.
As eCommerce, banking, entertainment and other everyday systems get incorporated with AI, businesses will have to ensure that the public can be certain that the AI technologies being used are transparent, secure and that its conclusions are not biased or subject to manipulation. In 2021, technologies that provide a measure of trustworthiness and “fairness” will start getting incorporated into AI lifecycle to help us build, test, run, monitor, and certify AI applications for trust, not just performance
And this is not limited to AI algorithms only… .
- The Dark side of AI
With the volumes of data increasing, 2021 may showcase the good, bad, and ugly use of artificial data, injected in models to cause harm. In a sparse data environment, for example, the number of people who upgrade their phones calling plan in a given month, “synthetic data” allows scientists to create extended datasets training AI. Fake data are created for exactly opposite purposes: it’s meant for perturbing AI training to create tainted models and results.
As the 2020 US elections demonstrated, AI bots disseminating misinformation and fake contents were much harder to detect. In fact, the fear of Deepfake AI text contents manipulating election outcome was so high this time, that OpenAI, the creator of GPT-3 AI has pledged to restrict the availability of its use for ethical uses only, closely monitoring its application programming interface (API).
The development of AI in the next few years will be very different from any other recent development of technologies. AI is different from any other powerful technology that’s been developed in the past — its ubiquitous, top quality algorithms reach globally within weeks, cheap computing power including GPU are available along with no code technologies such as AutoML. To counter this menace, I expect that AI professionals in 2021 will look for a consensus methodology for recognizing and unmasking adversarial threats in AI apps. Adversarial AI Threat Metrics, an open, extensible industry framework for classifying the most common adversarial tactics that have been used to disrupt ML systems will likely be adopted by ML and the DevOps engineers.
2021 will be a watershed moment for AI, with the technology out of the experimental cycles in 2018–19, adoption in 2020 and start being part of daily activities in all types of businesses, processes, products, and services. Once-in-a-century, “Black Swan” like events such as COVID will push the growth of the “K economy” forcing businesses to start adopting AI quickly or risk perishing.
The post-COVID world would be touch-less and digital where services are automated and remotely driven. Consequently, Workplace AI and Man-Machine collaboration will accelerate, putting computer vision technologies at the forefront.
As AI pervades every sphere of our lives, consumers’ understanding of technology will start changing. However, to be widely accepted by the public, the systems have to be fair, and accountable. Else, expect to see major backlash in adoption of AI — delaying, but still not ultimately preventing an AI-driven world.
Author: Dr. Santanu Bhattacharya
Chief Data Scientist at Airtel, Prof/Scholar at IISc & MIT, worked for NASA and Facebook, built start-ups