Originally published on ITWeb on 12 April 2022
By Kirsten Doyleg, ITWeb contributor
When it comes to AI, businesses are spoiled for choice. There are multiple technologies, brands, and categories, and over the past couple of years, a few of these technologies have made plenty of progress, such as natural language processing (NLP), machine learning (ML), cognitive chatbots – all are part of AI services. And because of ML techniques, bots are getting wiser over time as they’re exposed to increasing amounts of conversational data. So says Sumit Kumar Sharma, enterprise architect and head of advisory services at In2IT Technologies.
“All these technologies will continue maturing in the coming years. We will also see more implementations of computer vision, deep learning and AI cloud services with big technology players providing more comprehensive AI ecosystems coupled with low code or no code development platforms.”
When implementing AI in your business, it shouldn’t be regarded as a ‘one-size-fits-all’ approach, as each business needs its AI strategy to be personalised to its needs. To begin, define the business objective for implementing AI technologies. Companies need to understand their data maturity, readiness for change and customer needs. For example, if an organisation is a business-to-consumer company, it should look more closely at chatbots and NLP, if it’s seeking to improve communication channels with its customer base. However, if a business is more of a research organisation that needs to understand consumer behaviours or demand patterns, ML and advanced data analytics would be more beneficial.
When it comes to implementing AI effectively, Varsha Ramesar, managing executive, data and analytics at iOCO, says that a 2018 Harvard Business Review survey found that three-quarters of respondents believed that AI will substantially transform their companies within three years. However, the study revealed that highly ambitious projects are less likely to be successful than ‘low-hanging fruit’ projects. Not much has changed in the years since, and many AI projects fail because they’re too ambitious.
She believes that companies should look at AI through the lens of business capabilities rather than technologies. “Before embarking on an AI initiative, companies must understand which technologies perform what types of tasks, and the strengths and limitations of each. There is also a gap between current and desired AI capabilities, but this isn’t always obvious, so companies should create pilot projects for AI implementations before rolling them out across the entire enterprise. Similarly, there’s a stark difference between what you want to accomplish and what you have the organisational ability to actually achieve. Skills, in particular, are scarce, and companies need to leverage the capabilities of key employees, such as data scientists, who have the statistical and big data skills necessary to learn the nuts and bolts of these technologies. If they don’t have data science or analytics capabilities in-house, they will have to build an ecosystem of external service providers.”
AI’s dark side
Sharma adds that businesses also need to know that there’s a dark side of AI and it’s not just limited to transforming the skill requirements to run the business using traditional technologies. There are numerous examples of AI technologies being flawed by human bias, and now businesses are starting to see AI governance and responsible AI brought to the fore.
Don’t reinvent the wheel by building an AI system that has become an industry standard, warns Ramesar. “It’s a waste of your company’s time and resources. Instead, buy it from a company that has done research and development for years, and has launched a product that has been used and trusted by many users. Building an AI system in-house is a costly and risky endeavour, so leverage a partner’s experience.
There are numerous examples of AI technologies being flawed by human bias, and now businesses are starting to see AI governance and responsible AI brought to the fore.
Sumit Sharma, In2IT Technologies
“Many AI systems use deep learning and require thousands, millions, or even billions of training examples in order to perform a particular task. If the AI system that you need is an application that relies on supervised learning, you need to ask if you have the data to train the AI system, and whether the quality and the volume of data is adequate to achieve the expected performance,” she says.
“Uncertainties are inherent in an AI project, so don’t expect it to work first time. Instead, plan AI development to be an iterative process, with multiple attempts needed to succeed. The first and most important step of implementing AI in enterprises is to define specific, measurable, achievable, relevant, and time-bound goals, and start small, with a specific use case and invest in the fundamentals to create a good foundation. This is an opportunity to learn what works and what doesn’t, to learn from mistakes and implement better AI solutions in the future, to refine and improve,” Ramesar says.
AI in banking
Drilling down into some specific industries, Ryan Falkenberg, CEO of Clevva, says for financial services businesses, it’s critical to start by understanding your ‘as-is’ reality, and to be clear on what AI can and can’t do for you. It’s important to know what’s needed to make AI work, particularly considering compliance and regulatory requirements. Try to solve a specific business problem with a targeted capability. “Take customer service, for example. Initially, many companies raced to deploy a chatbot on their website, only to frustrate customers and implode the customer experience. This is because they fell in love with the concepts of natural language and machine learning. The idea of a free-talking chatbot that could self-learn to serve customers seemed too good to be true. And it was. Soon the prescribed nature of query resolution logic, together with the contextual richness of customer realities, proved too difficult to get right using predictive logic. It may have worked for info-bots, but not for resolving complex queries.”
We are seeing a convergence of the physical world and the metaverse, and a convergence of technologies such as AI and robotics.
Ajay Lalu, Q-Hop
Falkenberg says banks operate in a regulated industry, and every customer query, whether sales- or support-related, needs to be handled in a very specific way. Unfortunately, most chatbots aren’t designed to handle the complex rules and contexts that influence how any query gets resolved. As a result, while they embraced one aspect of AI (natural language understanding), they failed in another (getting the chatbot’s brain to resolve contextual, rule-driven queries off a generic knowledge base and simple decision trees). This highlights the importance of starting with the business challenge and finding the right solution. However, financial service companies can now deploy digital experts that can resolve complex, contextual queries in a consistent, compliant and hyper-personalised way. A digital expert is capable of understanding what the customer is asking for, then shifting to navigating them through a structured, contextually-adaptive resolution journey that results in a processed outcome. In short, the magic is in the mix, and we’re now building digital workforces designed to perform a specific job using the required capabilities, not building digital workers who are built from all the best in AI technology, but aren’t fit to do a specific job.
AI in retail
On the subject of how AI is being used to transform the retail sector, Ajay Lalu, co-founder and director of Q-Hop, a division of CIRT, says AI is the golden thread running through all the revolutionary changes taking place in global retail. “Retail of the future will be seamlessly omnichannel, hyper-personalised and convenient in ways we cannot yet imagine – and this will all be enabled by AI.”
Lalu cites several examples. “AI can allow retailers to make personalised offers to upsell and build loyalty, identify and retain customers that retailers risk losing, streamline supply chains, and predict changes in markets and patterns of behaviour, to adapt manufacturing processes and stock levels timeously.”
Q-Hop is currently launching pilots of a use case for AI – analysing shopping preferences and using this data to make hyper-personalised and real-time offers to the customer at the point of decision-making.
“We simulated over 20 million transactions to train the algorithm, and its predictability looks very good. For example, it will find a pattern in which a customer buys milk and bread every Sunday and offer a discount on bread next time they buy milk on a Sunday. This builds customer loyalty. But retailers can go a step further by upselling to the customer, offering them a three-for-one discount on bread when they buy milk, or a special on the brand of chocolate they occasionally treat themselves to.
Lalu says retailers – particularly in SA – have not fully adapted to the fact that the power now sits in the hands of the consumer. “They still tend to do ‘spray and pray’ advertising on leaflets, and some don’t know what a customer’s preferences are – even with decades of loyalty card information at their disposal. They’re moving too slowly to embrace AI, even though they typically would like to use big data analysis to improve operations and business. Most retailers have vast volumes of data, but the challenge is that it’s all over the place in legacy systems. These were never built to be data-oriented businesses. Turning that situation around could take years if managed in-house, whereas change must happen now. Retailers need to move into broader ecosystems and collaborate with organisations that are built from the ground up to manage and analyse data. Instead of thinking in silos, retailers should be part of intelligent ecosystems including analytics, payment, manufacture and logistics. AI is the golden thread that binds all these ecosystems together, making each component smarter and more efficient.”
AI in healthcare
AI is transforming drug discovery, innovation and analytics in the medical sector too, says Kale-ab Tessera, a key member of InstaDeep’s AI research engineering team in SA.
“We’ve seen things like the launch of solutions such as DeepMind’s AlphaFold, which solved a 50-year-old protein folding prediction problem, or Google’s Pathways, a system that can solve multiple tasks at once.
“InstaDeep is working with leading vaccine maker BioNTech, including applying the latest advances in AI and machine learning technology to develop novel immunotherapies for a range of cancers and infectious diseases, and applying InstaDeep’s DeepChain protein design platform to engineer new mRNA sequences for protein targets. For advanced analytics, a prominent example is InstaDeep and BioNTech’s high-risk Covid variant early warning system.”
The system identified more than 90% of variants of concern, on average two months before their designation by the World Health Organisation. It detected the highly transmissible Omicron on the day its sequence became available among more than 70 000 novel variants discovered in October and November 2021.
AI in cybersecurity
This year has brought a fresh wave of cybersecurity attack examples that demonstrate the ever-changing landscape of threat detection and hacker threat mutations, says Yashmita Bhana, CEO at Nihka Technology Group.
In short, the magic is in the mix, and we are now building digital workforces designed to perform a specific job using the required capabilities.
Ryan Falkenberg, Clevva
Self-learning AI in the cybersecurity world is the new foundational mechanism to drive what is fast becoming known as AIOps. These tools combine AI and ML to determine threat severity in real-time. Gartner predicts that 40% of companies will be using AIOps by 2023 for application and network infrastructure monitoring.
AIOps is a natural companion to process ever-growing datasets and behaviours to detect known and unknown threats. Using the internationally recognised and trusted MITRE ATT&CK Framework to process and learn from the data sourced from endpoint telemetry, AIOps tools are the natural addition to any cybersecurity strategy.
“In essence, the use of self-learning AI transcending into artificial general intelligence in cybersecurity surpasses the traditional methods in detection and prevention of threats and vulnerabilities through speed, quicker learning, prediction.”
While traditional methods require security analysts and teams to go systematically through a framework, AIOps tools like Fortitude does this seamlessly in a day and can assess the entire security posture, both internal and external.
AI in the future
While it’s very hard to predict how AI will take shape in five years, looking at the current landscape, we can see a few patterns, says In2IT’s Sumit Kumar Sharma.
“We will start recognising AI as an architecture domain and not just an enterprise initiative. We will see lot of successful projects that will deliver cross-functional AI capabilities to the enterprises, which will result in formal AI architectures, strategies, roadmaps, maturity models and suchlike.”
Enterprises would also need a matured AI delivery model. It will be worth exploring ‘ModelOps’ for speedy deployment of AI from pilot to production.
AI biases will become a serious issue in coming years, as AI continues to replace humans in decision-making. In coming years, we will see more deliberate efforts from organisations to bring fairness and transparency by making AI a ‘responsible AI’. Another related upward trend in AI we will witness is ‘digital ethics’.
In the next five years, first-line support agents are going to be the ‘digital workforce’, a combination of hyper-automation, cognitive bots, supported by matured NLP, advanced analytics and more.
Highly anticipated and closely watched, autonomous vehicles will continue maturing as well.
There are a few other players in AI that are comparatively still less popular today. “I will eagerly keep an eye on the evolution of ‘Transformers’ (a neural network architecture used for language understanding),”says Sharma. “It’s an existing AI breakthrough from Google’s stables, with a lot of promise. Although we’re still scratching the surface, in the coming years, we will see this area develop a lot.”
AI will be more commoditised in lot of ways. It will be within the grasp of small enterprises and individuals and a catalyst for growth in developing economies.