What is the difference between data and information? And how do you turn information into knowledge with AI?
Artificial intelligence and data-driven organizations – these are probably the hottest topics for most corporations at the moment.
But what do these terms actually mean? C-level managers and corporate strategists need answers when shaping innovation and transformation programs.
In short, the term “data-driven organization” covers two aspects—first, the more technical need for collecting data systematically throughout the entire company. Second, the organizational shift in decision making, focusing more on facts and less on seniority. Artificial intelligence itself is about outperforming humans when turning data into insights.
In our daily life, we use terms such as data, information, knowledge, or insights interchangeably. Everybody understands, based on the context, what is actually meant. However, precise definitions ease the understanding of data-driven organizations and the role of AI (Figure 1).
Data is the raw form: 1s and 0s on disk, as part of a stream from an IoT device, or strings and numbers in databases. On this level, we know nothing more about the data. Data itself does not carry any meaning. It might or even might not help to understand the physical or digital world.
Information is contextualized data. A stream of 1s and 0s becomes a video signal from a camera overseeing an assembly line. In a database, a string now represents a customer name or order ID. A number becomes the profit per year, another the turn-over last month. Data becomes information – and has meaning now.
In the foreseeable future, contextualization is and remains a predominantly a human task. Are these employee or customer names? Are these order IDs of open orders or of orders where a customer complained? Data catalog tools might provide some support. However, only humans understand complex data model semantics. They can intellectually penetrate all subtleties of similar-looking, but different-purposes serving data elements such as open, delivered, and returned orders.
Figure 1: AI and Data-Driven-Companies – The Big Picture
Actionable insights – also named “knowledge” - represent collections of information that guides decision-making. Are there irregularities on the assembly line? Which customers return orders most frequently? As the term “actionable insights” suggest, someone has to decide whether and which actions to perform – but the options must be clear.
Historically, generating actionable insights is the domain of white-collar office workers. Managers and analysts transform large Excel tables full of numbers into nice-looking diagrams. However, there are two limitations. First, humans can only comprehend and experiment with a limited number of influencing parameters. Second, while there are examples for statisticians and mathematicians working on core business questions (e.g., actuaries in insurance companies), this is different for most other industry sectors.
Most managers and analysts have heard about statistical significance at university years ago but never applied the concept in practice. Thus, for turning information into actionable insights, artificial intelligence is revolutionary. AI understands highly complex correlations, including non-linearities; plus, AI development methodologies emphasze robust quality control measures.
As a decade-old research, development, and engineering discipline, AI has many facets, with three often-mentioned directions or schools of thought:
Artificial narrow intelligence. Narrow AI singles out and automates limited, particular tasks that only humans could previously perform: identifying objects in images or playing chess, for example. Thanks to deep neural networks, narrow AI drives innovations today and transforms society and corporations.
Artificial general intelligence aims to build a human-like intelligent system capable of learning, understanding, and behaving like humans and the human brain.
Artificial superintelligence wants to build even better “brains” that surpass the human brain and its intelligence.
Reminiscences of these concepts are already part of popular culture. Artificial superintelligence was a topic of the movie HER a view years ago. An innovative new mobile phone generation comes with an AI assistant. These AI assistants are intellectually more advanced than humans– and eventually depart from them and their limited intellect.
The Turing test, a 1950s concept from academia, is widely known today. For passing the Turing test, an AI system must be indistinguishable from a real human conversation partner in a chat. Is this an example of general AI? Not quite. Building an AI system that mimics a human reflects only a very limited aspect of general AI. The Turing test does not require any learning capabilities. Today, commercially successful chatbots do not aim to mimic a human. They make it clear that they are just bots. They are not designed to pass a Turing test or achieve the level of general AI. They are perfect examples of narrow AI. They simplify buying or booking processes for online customers or query support databases, thereby providing a more human-friendly interface than, e.g., an SQL command-line interface.
Narrow AI improves the creation of actionable insights. These insights are the base for decision-making, which comes next and require some kind of judgment. Should the bank clerks call the ten or thirty potentially most interested clients to sell a credit card? There is an intruder in the building. Should we call the police, or should the night security guard check the situation first? A patient might have diabetes. Do we believe the test results (yet)?
The rise of narrow AI changes corporate decision-making. An AI model beats specialists with 20 years of experience and a senior title. In the past, there was often no other option than seniority-based decision-making, and most experts have years of experience in their field. They do not just command in dictator-style to prove their power. However, they have no chance against a well-designed AI model.
At the same time, AI is not the end of senior experts. They remain essential for companies, just with a new focus: for deciding in areas without a dedicated AI model, for challenging whether specific AI models benefit the organization, and for final decisions based on proposals from the AI.
Besides the seniority-based decision pattern, three more approaches are worth mentioning:
Decisions based on summarized information: A domain specialist gathers all relevant data, looks at correlations and dependencies before deciding himself, or prepares a proposal for a management decision.
Combining the power of AI with human judgment: AI generates the insights, the final decision stays with humans. This pattern brings benefits for one-time strategic decisions, such as whether to enter the Russian or the Australian market next year.
Autonomous algorithms: The AI model generates proposals for various situations and directly triggers the execution – without any human intervention. This pattern fits routine decisions, especially when human judgment is too expensive or too slow. Placing online ads is the perfect example. Algorithms must choose a suitable ad within milliseconds, plus a single ad generates only low revenues – a few pennies or sometimes Euros. Human intervention is too expensive and too slow for such use cases.
When considering how data becomes information and information actionable insights and how the different decision patterns work, the concepts of becoming data-driven and the role of AI get evident in the big picture (Figure 1).
Data-driven organizations require data to be able to decide based on data. They invest in the collection of data and its contextualization to get information. Furthermore, data-driven organizations want to decide based on facts. Either humans or algorithms turn information into actionable insights. Relying on humans is one option, but AI brings more benefits: better prediction or classification models that no human can create using, e.g., Excel. Furthermore, after data scientists trained an AI model, the model delivers results for new situations and data more or less immediately.
Figure 2: The Three Challenges for Data-Driven Organizations using AI Capabilities
When companies become data-driven and introduce AI, their business cases must not cover only the AI capabilities. They must incorporate the expenses for becoming data-driven: establishing capabilities for collecting data from data warehouses or business applications, from IoT devices, and from all systems in engineering or service departments – plus expenses (and time and resources) for re-engineering the decision-making processes (Figure 2). Looking at these three challenges – data collection, building AI models, transforming corporate decision making – holistically together is the key to success.
Klaus Haller is a Senior IT Project Manager with in-depth business analysis, solution architecture, and consulting know-how. His experience covers Data Management, Analytics & AI, Information Security and Compliance, and Test Management. He enjoys applying his analytical skills and technical creativity to deliver solutions for complex projects with high levels of uncertainty. Typically, he manages projects consisting of 5-10 engineers.
Since 2005, Klaus works in IT consulting and for IT service providers, often (but not exclusively) in the financial industries in Switzerland.