Aktualisiert: Jan 9
Content: In this blogpost the definition of an AI company is looked at. The strategic importance of the data flywheel is explained and why, from a strategic point of view, it has to be integrated into the concept of an AI company. And last but not least the seven AI patterns will be looked at in this article, to show which areas an AI company can actually consist of and what use cases we have today.
Introduction: In this chapter we'll have a close look at your business, more at the question, if you are an AI company. Is an AI-related company ultimately an AI company? Can you work with AI and still not be an AI company? And most importantly, what does actually AI means, what use cases do we have currently?
What defines an AI company?
Maybe you would say "Company + AI = AI company", but the answer is not that simple. To be an AI company, you must ultimately create business cases out of your models. So to specify, artificial intelligence has to be part of your business model.
What kind of business model will that be, you ask? Right, we talk about the data flywheel - which should be an essential part of your strategy! So have a close look at it:
What is the data flywheel and why do you need it?
The data flywheel is explaining the idea that the more users your company gets, the more data you will receive, which makes you capable to build better algorithms and ultimately you will get a better product. Which leads to more users again, which leads again to more data, ... and so on, until your data flywheel is rotating in such an significant speed, that it is hard for the competition to keep up with your company.
So we talked about algorithms, as in the last chapter we have seen this overview:
Now let us dive a bit deeper into the topic of artificial intelligence and its patterns. Building an business model around artificial intelligence requires you to understand the different benefits of its patterns and to which use cases they belong. Here we go with the question:
What are the elements of the seven AI patterns?
The seven patterns consist out of:
CONVERSATION & HUMAN INTERACTION
PREDICTIVE ANALYSTICS & DECISIONS
PATTERNS & ANNOMALIES
Let's see below the details:
All we need is a camera and artificial intelligence to receive the context of machine vision where we have image recognition, which is the ability of software to identify objects, people, places, writing as well as actions in images. In the recognition process we do machine-based visual tasks, like labelling the content of an image with meta-tags, image content search or guiding autonomous robots, self-driving cars and accident avoidance systems.
Example: Smart photo libraries, targeted advertising, interactivity of media, eccessibility for the visually impaired or enhanced research capabilities.
Here we use machine learning to develop a unique profile of each person so that we can display relevant content to that person. So one could say that the main goal is treating the customer as an individual. Mostly we have personalized content, which has been created after a behaviour profiling, which can be seen in the data of loyalty cards, or which has been transmitted through recommendation systems.
Example: Recommending products or providing personalized recommendations. This goes to the point where we are able to provide personalized finance, healthcare, and one to one insight, information, advice or feedback. So it is of course the advertising industry, but also it is used in medicine for hyperpersonalized medicine where the recommendations as well as the treatment can uniquely be created. In finance a good example is the movement away from the FICO credit scoring system.
Conversation & Human Interaction
The goal of this pattern is to make machines interacting with humans, the way humans interact with each others.
Example: Chatbots, voice assistants, content generation, sentiment analysis, mood analysis, intent analysis and machine translation.
Predictive Analytics & Decisions
The word predictive implicates that this pattern helps us to predict future outcomes by using forecasting, machine learning based forms of regression and prediction, assisted search and retrieval, predicting behaviour, number or value predictions including dynamic or predictive pricing, predicting failures and anticipating trends. Mainly its goal is to help humans make better decisions, so it is using insight which it has learned from interactions, behaviour and data. It is augmenting humans by providing answers to questions, based on selecting best data or optimization activities and giving best possible advice. Important to know is also that we use machine learning to improve predictions and decision making over time, which means it is adaptive and we do not use simple statistical methods.
The difference to autonomous systems is: The autonomous systems do the decision without the support of humans, while the predicitve analytics gives us the data, on which we finally will do the decision - not the machine.
Goal Driven Systems
This pattern of AI is used in combination with agents, where we give them the ability to achieve and find the best optimal solution to a problem through trial and error.
Example: Game playing -> example of AlphaZero, scenario simulation, iterative problem solving, resource optimization, bidding and real time auctions.
If a system is able to accomplish a task or achieve a goal by itself, we talk about an autonomous system. It does not matter if it is software, a virtual autonomous system, or if it is physical and it's a hardware autonomous system. So we can define an autonomous system with its goal to minimize human labour.
Example: Machines, cars or bots which do tasks autonomously. We also count collaborative bots as a form of autonomous system, even if they are operating in an augmented intelligence role. Cognitive automation is also part of autonomous systems. Last but not least we include to autonomous systems all the systems, which independently make decisions.
Patterns & Anomalies
We use machine learning and cognitive approaches to identify patterns in the data. The main goal of this pattern is to identify if there are anomalies in the data or if they fit the rest of the data.
Example: There are several use cases like risk analysis and fraud detection, intelligent monitoring, automatic error detection and correction, allocating hidden groups of data as well as identifying best matches for data or predictive text.
This little introduction should have helped you with the question, if you are an AI company and if you are using the right business model. There is no right and wrong, but for sure we can say that a first mover advantage is necessary in this area of business, as the data flywheel has shown, its hard to keep up with its fast pace.