Technological advances in the past decade have created new opportunities, platforms and tools, which influence the way we communicate, travel, shop and work. The proliferation of technology has made one wonder what life would be like without it. With the advent and explosion of big data, society has become more interested in what is likely to happen next, instead of solely analysing what has happened.
Dr Jack Hong is an expert in end-to-end Artificial Intelligence (AI) applications. He is currently the Data Science advisor to Vertex Holdings and AI advisor to Certis Group’s Centre for Applied Intelligence. Jack creates proprietary business capabilities in these roles using cutting-edge research and engineering skill sets in analytics, natural language programing, computer vision, and reinforcement learning. He also leads Research Room, an AI consulting and development outfit that delivers complex decision-making solutions for large organisations and Integrum, the developer of Kailash, a microservices orchestration platform that delivers AI and Data as a service.
Jack is an adjunct faculty with the Singapore Management University (SMU) and has been actively teaching undergraduate and postgraduate programmes including digital transformation, python coding, data science, and financial economics since 2014. In addition to his SMU commitments, Dr Hong trains professionals and interns in AI capabilities, from theoretical foundations to engineering know-hows. His research interests lie in developing new deep learning models that can solve challenging business problems organisations face today.
Jack graduated with a bachelor in Finance at the Singapore Management University (SMU) in 2003. With the technology then, it was difficult to imagine the possibilities of machines providing insights, or even making good financial decisions. However, Jack was intrigued by the interconnection between AI and finance, and it led him to co-found Research Room, an AI consulting and development outfit that delivers complex decision-making solutions for large organisations. Jack also completed his doctorate in empirical finance, where he investigated the relationship between stock prices and management guidance credibility, analyst tones during earnings conference calls, and power dynamics between CEO and his board of directors using social network algorithms.
How does AI aid in an individual or a company’s financial decisions?
Jack: We often assume that the investor or the market is rational, and we analyse how the market should perform under these rational conditions; but when the market does not perform ‘rationally’, we simply deem it as behavioural finance. Nevertheless, finance is not just about managing money. It is about how people make investment decisions, and this is very much based on what they think is going to happen to the company in the future.
Contrary to common perception, accurate predictions of asset prices do not solely rely on historical numbers. It is important to triangulate information from everywhere, how the other competitors in the market are doing, the economic outlook, what other investors are discussing, which influencer said what, etc. This wealth of information extracted from non-numerical data, such as textual data, unfortunately, could not be sufficiently analysed through traditional statistical methods. We require new skill sets to collect and transform unstructured data into mathematical forms, as well as more powerful algorithms to extract complex relationships within these data. This is where computing and machine learning shine and result in a demand surge for data scientists who can do traditional research and code.
How did you find the bridge between AI and Finance?
Back then in school, we talked a lot about traditional economic theories, how markets worked, and how to utilise various statistical methods to explain what happened. It was a lot about learning significant drivers of past events using historical data, and very little about making accurate future predictions. However, as we further immersed ourselves in the practical business world, we realised that it was more productive to make decisions based on measurable predictions than a strategic debate about which combinations of historically known forces are likely to happen in the future.
Traditionally, empirical methods in finance search for a statistically significant relationship between firm performance with quantifiable historical data such as accounting information, analyst forecasts, patent data, and market data. We then try to explain how certain statistically significant relationships support or reject known theoretical models. Decision makers then make decisions based on how convincing these hypotheses are.
AI techniques approach this workflow differently. First, AI models can be constructed to directly predict the set of actions that one should do to maximise profits. Second, various AI algorithms can model complex relationships of various data types such as languages and images. For example, a transformer neural network converts text into numerical representations that preserves contexts, and such networks have a very long memory span. Lastly, AI models go through an automated process of engineering and tuning to maximise its predictive accuracy on future outcomes.
Given how markets are dynamically driven by announcements, influential opinions, and sentiments in real-time, the combination of speed, scale, scope, and precision of AI models is perfect for this use-case.
How should we understand Artificial Intelligence (AI)?
Artificial Intelligence is an umbrella term that describes any solution that automates the human decision process. For a human, the toughest decisions are decisions under uncertainty, which then warrants the use of intelligence augmentation techniques. For anything else, it’s simply too trivial. Thus, for any solution to claim AI capabilities, it must be able to make decisions under uncertainty and achieve better precision at a faster speed than humans.
There are many levels of AI. The state-of-the-art technology that we see today is known as Narrow AI. They can solve well-defined problems very well but are too myopic to be treated as a model that has common sense. The next level is Artificial General Intelligence (AGI), which the current generation of algorithms are not capable of achieving.
Apart from algorithms, AI requires massive amounts of data, and in today’s digital era, data includes texts, images, videos and audio. The way AI works to make decisions under uncertainty relies on massive amounts of data to generate a probability of what is going to happen in the future. To do that, state-of-the-art AI models, such as deep learning, need to build internal representations of complex relationships between data points and store this in its memory. These internal representations are rooted in linear algebra and thus, agnostic about the types of data it is representing. As such, the highly generalisable architecture of deep learning models can be applied to many different sectors and lines of work.
How is AI used in Strategic Communications?
Understanding non-verbal cues is a crucial aspect of effective communication and yet most of our decisions are made by focusing on what was said rather than what we do not hear. Fortunately, AI is here to help. With its long memory and highly contextualised understanding of a given domain, AI can understand spoken context, analyse non-verbal cues such as micro expressions, retrieve more public information about the context, and generate the intended output of its user, such as a press release, tweet, or even a photo-realistic fake image.
I’m a big fan of the US TV series Lie To Me and am particularly interested in micro expressions. In the show, Tim Roth plays the role of a highly trained behaviourist who can read human intents by changes in their expressions that stays for only a fraction of a second. In reality, non-verbal cues such as gazes and gestures serve to augment and reinforce spoken communication. Non-verbal communication can reveal a person’s state of mind, intentions and their thoughts and feelings. Without training, it is easy to overlook the importance of inflections, word choices, word emphases and body language that reveal emotions, depth of feelings and hidden intents, but trained AI machines will not miss most of these subtleties. They are also way faster, more accurate, scalable, and instantly deployable compared to human experts.
Ultimately, a leader’s effectiveness, or rather social influence, is shaped by the interaction and communication with its followers. It is not the sole determinant of a business decision, but it certainly gives potential investors and collaborators an additional consideration to make the business call. The usage of AI technology in these circumstances is becoming ubiquitous. Natural language, video and audio AI models are already deployed in company conference calls and public appearances by influential leaders. Trades are being made based on the real-time results of these models. I think that leaders need to start working with AI to up their communications game.
Will an AI takeover occur?
It is inherent that with the advent of technology comes a new set of challenges the virtual community face – the issues of data privacy and the sentiment of an AI takeover. However, while AI harnesses great potential in the work that humans do, it is important to see AI technology as a complement rather than a replacement. AI helps us to make better decisions under uncertainty, but they are still quite myopic.
Today’s AI models perform exceptionally well, but only to well-defined problems and only with sufficient amounts of unbiased data. They do not own the problem and cannot see outside of what its creators want it to focus on. Given the way we design our AI algorithms today, an AI takeover can occur only if humans train it to do so, and I’m not even sure if it’s capable enough for such a ‘general’ task. And since we don’t have a good idea of what the next generation of algorithms that could tell us how Artificial General intelligence (AGI) looks like, it’s too premature to even speculate about AI takeover.
What is more troubling is biased or unethical AI. Today, many AI companies play the role of the middleman and develop AI models to make decisions for its users. As users, we don’t have a choice on how we want these models to make decisions for us. Social media is a good example, it shows you posts that will make you stick around longer so that they could make more money from advertisers with more clicks and impressions. The scary thing is that this outcome is not intentional. The owners simply give the models more personal data and tell it to figure out how to monetise better. They never intended for it to create the polarising society, alarming teen suicide rates, plummeting self-worth, and smart phone addiction that we experience today.
The AI models that we wield now are simply tools, but they can become weapons of mass destruction. We still need humans to guide its training towards a fair, just, and equitable outcome for all, guided by the right incentives, values, and of course, unbiased data.
Having helped many organisations in digital transformations, what are some common misconceptions you observe among companies when we talk about the integration of AI?
One of the biggest misconceptions companies have about AI is that AI is cheap because it can reduce expensive manpower costs through the automation of labour-intensive work, including decisions that middle managers are employed to make.
AI is not cheap by any standards. The talents you need to build and operationalise AI capabilities have salary ranges in the top 20 percent of the labour market. The average senior data scientist in Singapore makes much more than the median household income. And you need a team of talents from data analysts to software engineers. Next, the infrastructure and operating expenses, even when deployed on Cloud, are not cheap. The promise of AI is not cost cutting, but margins. The cost of AI does not scale linearly with the size of the company’s operations, but it can help companies create new value propositions, and generate additional revenue from existing and new sources with lower expansion costs.
The mindset of companies that can deploy AI successfully is thus different. They no longer think about economies of scale but economies of scope and speed. They are quick at conducting rapid experimentation to discover what users really want. They are even faster at deploying products and services by co-opeting (cooperate and compete) competitors, partners, consumers, merchants, and suppliers to build a larger pie, an ecosystem that can disrupt businesses in other industries.
Without these changes in business mindsets, owners and managers are going to face nothing but disappointments with AI solutions.
What are some challenges faced in the AI sector today?
One of the greatest challenges we face is to put into control the unintended harm in which these advanced technologies may bring. Take the commoditisation model, which many social media companies have adopted as an example. The typical data algorithm tells you the browsing patterns, demographics of users and audiences etc, and the main mechanism is to create impressions and clicks to achieve higher conversion rates and optimise profit. However, the machine does not decipher the truth and reliability of the data, and certainly does not take into account of protecting the social fabric, non-discrimination and other intangible assets of society. What it simply does is to ensure that the user stays on the screen longer every day.
In finance, I am always wary about these kinds of issues. If one day we are going to make decisions based on incomplete data, just because it is easy to do and easy to automate, we are going to cause unnecessary discrimination on generations of people. Now it’s no longer about unequal outcomes, it is about unequal access to opportunities. In finance, it will be in terms of your mortgage, in terms of the credit lines, just because of the unintended discrimination. This is one of the critical reasons why human insights remain indispensable, and the increasing importance of strategic communications when there is technological progress.
How do you think videography complements strategic communications?
Creative design is an important element of clear and effective communication. Videography, as a creative tool, captures and communicates emotions. The connection and resonance videos create help to capture the attention of your audience, and build trust and credibility among all. For corporates, videography helps to bridge the gap between how others perceive the organisation and how it lives up to its values.
From statistics, it is clear how videography has transformed consumers’ behaviour, content curation, how we run advertising campaigns etc. I have always liked this analogy from one of my mentors: If photography is an expansion of time, videography is a compression of time. The longer we look at a photo, the more we think and imagine about what happens behind the scene. On the other hand, videos require you to compress all the events across a long duration to present a succinct story in a few minutes, sometimes even in seconds. That is one of the reasons why videography is sometimes said to be information-heavy.