Every business will soon be an intelligent business in the same way every enterprise is a digital enterprise. If they don’t become intelligent, they will simply cease to exist. Artificial intelligence, a revolutionary cluster of technologies, is a game changer and has the potential to make fundamental changes in how we function. However, an AI strategy, if it's just an AI strategy, won’t take you very far. An AI strategy that’s just an AI strategy is an odd managerial notion that by sprinkling this magical AI sauce over everything will turn everything more magical and awesome.
Fight against this temptation of plunging into things AI-first. Instead, look to your strengths first and then bring in the AI twist. You need AI applications to work smart, to take advantage of your data, to learn about and improve on your past performance. You don’t need AI to become something new that you don’t yet comprehend. You require AI to build on the strengths you already have, and transform from what you already are, but better. And along the way, if you come across products and ideas which disrupt your current practices and your industry, great.
Where to start?
What are you exactly looking for? Do you want to better internal processes like Accounts? Are you looking to improve customer experience? Are there features in your product that would benefit with AI or do you need some insights into product design? Are there specific tasks that are repetitive, error-prone, and basically boring, or tasks where a little help can make your employees more efficient? AI is a tool that can help you with all of these things and much more, just like a chisel can be used to build any kind of statue. But to succeed, an AI strategy has to be part of an overarching business plan. Whether you’re improving your current business or building out a new business, AI should serve your business plan, not the other way around.
What do I need to understand before I go any further?
AI can do amazing things. But before your team can use it to do amazing things, you need to look at it realistically and understand its challenges. For example, AI has proved to be really good at classifying things like, tagging pictures, but it really can’t tell you a lot about what’s in those pictures. In a business angle, an AI application might be able to tell you whether users look delighted when they see a new product design, but they can’t answer a bigger question, like if the product will be a success.
Now how do you define A.I.?
There are enough blogs and articles out there defining artificial intelligence and machine learning. All the definitions may seem kind of disjointed, yet they all overlap. Moreover, for better or for worse, we have made a habit of using artificial intelligence and machine learning interchangeably. For now, let’s just abandon these terms and call it automation. When you’re effectively automating a process, it won’t matter what technique you use to automate, be it an older, simpler technique or a neural network.
However, it’s still important to start with a definition of AI. So here’s ours. Artificial Intelligence is known by its output which needn’t be particularly dependent on the input or on the algorithm, but rather the training process in which the program learns how to perform its task. Training is what differentiates AI from data analysis and other traditional software applications.
What does AI do well?
AI is great at solving specific, well-defined problems. Think self-driving cars, developed by Tesla, Uber, and others. Driving a car might not seem like a specific, well-defined problem. But it can be broken down into specific, well-defined problems like identifying road signs and signals, including other vehicles and pedestrians, planning a route, identifying obstructions, managing the brakes, detecting skids, and so on. None of these issues are easy to solve, but that doesn’t make them unsolvable. Break that big problem down into smaller more solvable problems. This process is what engineering is all about.
Artificial intelligence can be effectively used to assist and augment humans, not replace them. Companies that view intelligent machines merely as a cost-cutting tool are likely to push them into all the wrong places. AI does away with the repetitive, dull parts of the job, but this doesn’t always mean you’ll require less staff. In fact, you may need more highly trained employees to work on the more creative, less routine parts.
What makes adopting AI so hard?
AI products are data products. Training your AI product requires data, a lot of data. Valid data. If you don’t already have solid data practices in your organization, the harsh reality is that you’re unlikely to have useful data. You’ll need to identify data sources, build data pipelines, clean and prepare data, identify potential signals in your data, and measure your results. Organizations that flourish with AI will become proficient at acquiring data strategically. They won’t merely have data. They will understand where and how to get more. Most companies with AI aspirations do not have right data practices, even though they believe they do. Take a close and skeptical look at how you presently use data as part of creating your AI strategy.
Are you ready for A.I.?
We don’t want to be a wet blanket in the presence of all this excitement, especially when we share that excitement, but more often than not, companies are not ready for AI. Maybe data literacy is not at the heart of your culture, or maybe you hired your first data engineer to lesser-than-great outcomes. But the most typical case is that you haven’t yet laid down the infrastructure necessary to sow and reap the advantages of the most simple data science operations and algorithms, much less machine learning.
Hiring the right talent required to deploy an AI strategy
There are broadly three parts to an A.I. approach: generating data, interpreting that data and making judgment about that data. Keeping this in mind, an AI team requires at the least three separate roles: a data engineer to organize this information, a data scientist who investigates this information and a software engineer who implements the applications.
Hiring can get a lot harder when it comes to attracting researchers, data scientists and software engineers with experience in building AI-enabled software. Either embrace aggressive poaching or take the long-term route and work on bottom-up development by training the engineers you already have on the new paradigm. Sometimes this method might be quicker than bringing in outside talent since new hires often don’t work out. And if you’re a smaller company, remember, you don’t need to do your own research. Ride on the backs of the others. Given, we’re still watching out for the Ruby on Rails of the AI world, however, we’re getting closer as more companies on open source licenses, release their own tools.
Now, AI projects need to be trained before they can do any useful work. By training, we mean running the application on a bunch of known data and allowing it to create a model. In other words, automatically tweaking internal limits until it gets satisfactory results on your test data. After which you run it on a different set of test data, one it has not seen before, to figure out if the results are adequate. Training works alongside writing and debugging the software. It’s possible that more effort will go into training than in the remaining the development process. So account for training time if you’re planning an AI project.
You also need to understand how training can go wrong. Your application has probably memorized the training data, if you get hundred percent accuracy on training data, and will most probably perform terribly on real-world data. This is known as overfitting. It’s a never-ending issue. On the other hand, if your application isn’t hundred percent accurate on training data, it will never be hundred percent accurate on real-world data. Too bad. However, a few applications need a lot more accuracy than others, for a self-driving car, ninety-nine percent isn’t good enough, but sixty percent accuracy is alright for an app recommending products to users. Secondly, your application’s results will be biased, if your training data is biased. This feels obvious, but bias could sneak in unnoticed. For instance, AI systems tend not to show technical job postings to women as there are fewer women in technical jobs. And bias can have legal consequences.
Lastly, it’s simple to think training is one-time, get it done and forget thing. It isn’t. Business conditions change, customers change, products change, changes in your environment can affect your application. Its performance will gradually degrade over time, even though you might not notice. If you’re planning an AI project, you need to account for retraining.
Can you afford not to get on the AI train? You can’t. But don’t jump on the wagon blindly, map out where you’re headed before taking to the road. Looking at artificial intelligence as a magic ingredient that makes all things better will lead to costly mistakes. Understand what you’re doing, why you’re doing it, and the limitations you face. Both the limitations of AI itself and the limitations of your organization.
Finally, if the challenges of doing AI doesn’t scare you away, start planning your AI project. AI does present some huge challenges. But, these are all challenges you can meet, provided you haven’t started your project blind. Build a good data team, and develop good practices for working with data. Allow adequate time for training. Stay aware of the problems you’ll encounter along the way. And remind yourself, there’s nothing magical about AI, and similarly, there’s nothing magical about what can go wrong.