Welcome to the first post in Scaling AI series, where we will talk about transforming a company into an AI-driven one.
The term Artificial Intelligence has been used and abused so much that you can literally find a bunch of Python scripts with
if-else statements which people call AI for some reason. This is great for selling snake oil, but we are not in this kind of business. Before we start developing an AI strategy, we need to figure out exactly what we want to achieve, and learn how to make our arguments effective and concise.
Today we will veer from the popular way of defining what AI is, and instead talk about two most vivid misconceptions. After that we will try to figure out what AI is really used for.
Misconception 1: AI is about getting rid of the humans
Throughout many years of delivering AI projects and working closely with business stakeholders, I found that many people — even more technically inclined — view AI as “Automation on steroids”. The train of thought is pretty clear:
- AI is about emulating human thinking.
- A lot of work done by humans in modern businesses is based on thinking (as opposed to physical labour).
- We can replace humans with AI.
- Hey, this is the same thing as automation when we replace humans with machines or software!
As much as this argument is compelling, using this narrow definition of AI can seriously hinder your efforts in transforming your company into an AI-enabled business.
The main premise of automation is that we take an existing business process highly dependent on human labour and install some machinery or software scripts that enable that same process to run faster, more safely or more efficiently. The point here is that the outcome of this process rarely changes when automation is applied.
Common examples of traditional automation would be installing industrial robots at a production line or replacing an employee, who manually enters data into spreadsheets, with custom software.
Artificial intelligence is much more than that.
Misconception 2: AI emulates human thinking
First of all, I would strongly discourage comparison between current AI technologies and human intelligence. Many media outlets overhype AI so much, that people start to believe we are standing on a brink of Artificial General Intelligence. The misinformation is so prevalent that almost every discussion I have with a new business stakeholder starts with defining what AI is really capable of. Sometimes it's really, really hard to explain that no, AI can't generate five paragraphs of cohesive legal text based on a one-line question.
Educating people about what's possible is a complicated topic that we will discuss in later articles. The key concept here is “educating” as opposed to just saying: “OK, we can do this and this, but definitely can't do that”.
When you talk about AI and Machine Learning, it's much better to present these technologies for what they really are: an exceptionally good statistical toolbox that can learn patterns from existing data and make decisions based on new unseen data. It's extremely important to manage people's expectations and highlight limitations of current approaches.
Take, for example, the same people who asked me to make a system that generates legal text. They listened intently while I explained what AI in its current state is capable of. And then they said something like: “OK, so we get it. AI is not like a fully grown human, it's like a 3 year-old. Can we teach it some reeeally basic legal stuff”? Don't settle for this, however tired you might feel. AI is not similar to humans — adult or otherwise. The most profound concept that current AI technology still cannot tackle is causality, while humans start to grok cause and effect as early as 8 months in. So no, AI will not learn all the intricacies of our legal system just yet, but it doesn't mean it can't be useful.
AI is about new opportunities
It's much better to talk about AI in light of new opportunities it provides. Let's look at Amazon: 35% of everything people purchased came from personalized recommendations (as of 2013, source). Think about it: this story is not about automating an existing process or doing something that humans already do. It would be impossible to manually sift through all of Amazon's products and constantly keep “similar products” section updated. Likewise, it would be impossible to create some fixed linear script which would cater personal recommendations to users, while constantly improving itself. Such effective recommendations are possible exactly because of advanced statistical capabilities of AI: there is a lot of data about products and purchases, and we want recommendations to maximize revenue. We have the data and clearly defined objective — a perfect environment for AI to thrive in.
So what is AI about if not replacing humans or automating existing processes? The answer is: it's about enhancing humans and creating opportunities for things that were not previously possible. AI is about making decisions based on data and being able to continuously improve performance based on feedback loops. AI can enhance almost every aspect of company's operations:
- It can help humans make decisions faster.
- It can help humans make more informed decisions based on more factors as opposed to using intuition and gut feelings.
- It can make decisions completely autonomously, freeing humans to do more meaningful work.
- It can create entirely new products and services when applied creatively: think Prisma, Alexa, and numerous other products where AI is the primary value proposition.
- It can help discover new avenues for business development based on data that is currently being generated.
So next time when you are discussing AI with C-suite execs, and they say something in the lines of: “Yeah, but we already do automation, how is this different?”, you will know what to say.
Great, we need AI. Where can we buy some?
When you deside to embark on a journey of adopting AI at your company, it's easy to feel lost and unsure of the next steps. You can't just go and buy a perfect AI solution from a software vendor — despite all their efforts to convince you otherwise. If you want AI to have a meaningful impact on your company, you must understand that getting there will take lot of time and effort. You will have to open up data that is currently locked in silos throughout the company; make sure data is available and updated regularly; create infrastrcture that will make use of data; transform people's views on decision making; ensure that feedback loops exsist and can be easily added to new products.
All of these activities need to be part of a comprehensive AI transformation strategy, which needs to be validated, supported and communicated from the top levels of company's hierarchy. AI can start as a small guerilla operation inside one branch of the company, but in order for it to become scalable and pervasive, you need to lower the adoption bar for product teams and make sure tools and infrastructure are in place to make deployment of AI products easy and profitable.
In next articles of this series we will discuss different aspects of AI transformation: from data availability to internal communications and promotion. Stay tuned!