Updated: Feb 10
AI is as important as ever. But what is the staying power and value of skills in this space? I am going to argue that it is declining. Case in point: in 2015, Tensorflow (called DistBelief at the time when it was internal to Google) sucked, according to engineers at Google. Using Inception in Tensorflow (image classification) was complicated and took a lot of trial and error. By 2018, using Keras, (a framework on top of TensorFlow), you could do image classification in just a few lines of Python code.
As another example, think of model tuning now. We have AutoML, Sagemaker, Databricks. All this innovation is a good thing and a sign of fantastic progress. But it means that we must keep our skills and those of our AI group up to date.
I keep telling people that Data & Analytics is innovating in the cloud. Many Data Analytics products and solutions are complementary products to the cloud, meaning they go together and get bought together. If the price goes down in one, it will help the other. You get the idea. By being cloud-native, Data & Analytics solutions inherit its benefits (if designed to take full advantage of them, of course). I just wrote about Snowflake in a previous post, which is a great example. The pattern is clearly not going away. We are seeing everything getting automated, and the skills required to scale AI in the organization will shift.
But how quickly are skills required changing?
SKILLS WE SHOULD FOCUS ON
If you need to have continuity in your career, never forget the big picture and why you are doing what you are doing. Tools and technology will change, but the primary objectives are eerily stable: save money, sell more, and reduce risks.
If, as a leader, you need to keep your people up to date, perhaps the training must be more strategic. X.AI mentions the following skills as most important to take full advantage of the benefits AI can provide:
Notice that pandas, tuples, Naive Bayes and Random Forest are strangely absent. For one, if we lose sight of business objectives, we will never reach them, no matter how knowledgeable our teams are. Furthermore, the hardest part is getting quality data to work with. No amount of fancy algorithms can do anything without it. Finally, between complex and convoluted AI models, the preference is to remain simple because while we may lose "some" accuracy, the explainability part is so essential when we try to apply AI in the real world.
Train your teams in the above skills and hire temporary help for the technology skills, perhaps.
DATA MORE IMPORTANT THAN FANCY AI ARCHITECTURES
The media pumps up the excitement and euphoria about AI, sometimes forgets that it is all about the data. Data is much more important than fancy AI architectures. Asking the right questions, choosing the appropriate data for the problem at hand, and eliminating useless studies (I will write a post about this, promised) take other skills (the ones above) that will be much more long-lasting in terms of value to organizations. Problem solving and critical thinking will help people do their jobs forever.
The media tends to pump up the excitement and euphoria about AI, sometimes forgets that it is all about the data. Data is much more important than fancy AI architectures. Asking the right questions, choosing the appropriate data for the problem at haned, and eliminating useless studies (see this previous post) take other skills (the ones above) that will be much more long-lasting in terms of value to organizations. Problem-solving and critical thinking will help people do their jobs forever.
Building AI models may be exciting, but what really matters is having better data than the competition; this is where organizations must put some of their best minds. It's a long game to be sure. But rest assured that at some point, we will only need to press a button, and AI will get done. Really. The competitive advantage will shift elsewhere for organizations. I argue that better data, better quality data, based on a better, smarter, longer-term Data Strategy that is ruthlessly focused on the business strategy and takes into account all the various data use cases. What is becoming possible is impressive, and it is happening fast, but it needs data.
Buckle your seatbelts!