Five stages of value for artificial intelligence
(This is the second post in our guest blog series by Esteban Kolsky of thinkJar. See the original post for an introduction and a list of other posts in the series.)
Without going into excruciating detail, the first Encyclopedia of Artificial Intelligence I read was three tomes, each over 1000 pages. There are many, many books that can tell you what the different tools and technologies used for AI are. There are few that will tell you how to understand AI for business – this blog is a start towards that.
Working with myriad organizations across the planet over the last 15+ years to solve knowledge management and artificial intelligence questions, I came up with a framework on how AI can be used in day-to-day interactions with customers.
Each stage of the framework generates value, both for the user and the system. Let’s explore each in more detail.
The start, the genesis of knowledge being used in the organization.
There are many ways to perceive (basically, perceiving is to sense some change in state – a customer has a question, there is a problem arising as reported by a machine, an IoT sensor, etc.) but they can be summarized in three modes:
Descriptive: there is a description of the need or the question.
Predictive: a little more complicated, based on learned patterns of perception. Most common use is optimization.
Prescriptive: the most complicated of the models, the one that leads to better knowledge and machine learning. In this model the computer not only predicts the next best action, but weighs all the different scenarios against each other and recommends one over the other based on personalized profiles and optimized processes.
By perceiving the data – in some cases before the customer even has access to it – the computer can then assist agents to deal with customers more effectively and with a win-win mentality. Even though the answer won’t always be what the customers need or want, it will be the appropriate and correct one, the most logical considering all the available data points and variables in the model.
Value generation in this stage is about vetting and verifying existing models, further optimizing processes and outcomes, and learning more about the customer. Improvement of existing optimization models can only happen when exposed to the real world, and the data generated in this stage is invaluable for that.
Once we, or the computer if allowed to do it automatically, become aware that a potential problem has arisen, we set about collecting the necessary information to solve the problem. There are three parts to capturing the necessary information:
Content: the content is the information that encompasses the question or problem. Content is decoding the problem or issues into the different elements, so it can be understood and solved.
Context: refers to the circumstances surrounding the problem. Context is understanding the variables that make an answer appropriate for a situation.
Intent: the most complicated of the three elements, intent refers to understanding the reason the event took place. Intent goes past context, although it relies on it, to focus on the root cause, the origin of the interaction where AI can be applied.
Capturing the right information for the computer to apply its processing power is further complicated by the simple concept that organizations, until not too long ago, thought in a company-centric manner; they did not seem to care much about the customers’ perspective and reasons to do what they were doing. More recently, with the expansion of ubiquitous customer demands we began to look at interactions as customer-centric, revolving around the customers’ needs.
The value generated by this stage is to both create a broader, more complete knowledge and information about the interaction that needs to be resolved and augment the ability of the computer to both perceive problems earlier in the future and understand concepts more easily.
Understanding is the beginning of the loop of resolution.
Once we understand what the question, problem, or issue is we can begin to resolve it. Until we get to this stage, all we have done is sense there is a problem and capture information about it – we have not fully understood the problem and all its connotations.
There are two classes of tools that are leveraged by this stage:
Linguistics: this is more than likely what you think of when you think of AI (and KM) today: a set of natural language processing (NLP) tools that can take a statement, a sentence and convert it to a problem and associated variables that can be given to a computer in search of a solution.
Classifiers: after we understand the language being used – and the associated content, context, and intent based on existing or new data captured – we then need to classify the problem in search of a solution. To reach a single answer, classifiers are the most valuable tool.
The simplest aspect of AI and the practical application of the complex prior stages: apply all the collected and understood information and do something. It is also the most structured stage of the AI-driven interaction.
There are two potential results when using AI: success or failure.
For an AI-driven transaction to succeed, there must be an outcome that is correlated to the original stakeholder’s needs. AI can only deliver an outcome in one of three ways: optimized workflows, personalized solutions, or automated answers.
This is where AI systems differ from human-to-human interactions. In a traditional knowledge-aided human-to-human interaction, the answer is given by one person to another after adding a value layer of interpretation.
Computers don’t interpret things, not even in AI or machine learning, but rather complete tasks. This is at the core of what AI can do for human interactions, and at the core of how they augment knowledge by learning the best paths to the right answer, and then optimizing the process to get there, personalizing the answer, or automating the interaction completely.
Artificial intelligence is the entry point to machine learning. After a machine becomes sufficiently well versed at perceiving a problem, capturing the necessary information to solve it, understanding the inherent questions or issue, and creating the unique solution for it, it begins to learn.
Learning, represented by the science of cognition, is simply an accelerated representation of applied intelligence.
In the purview of this blog series, and the underlying research, learning is what makes AI augmenting knowledge management possible. Taking stored, static content, learning how to add value to it by leveraging knowledge, wisdom, and data, and finding the right answer are the entry points to the ultimate outcome of knowledge management (and the leverage of AI tools for it): repeatable outcomes.