“AI is the fastest growing workload on the planet,” Mike Gualtieri of Forrester Research.
Last week, during H2O World San Francisco, we had the privilege to hear featured speaker Mike Gualtieri from Forrester Research offer tips on how to make AI happen without getting fired. This knowledge, he explained, was acquired by talking to enterprise data scientists and enterprise leaders who embarked on machine learning and AI projects in their organizations. When determining and executing an AI Strategy at your company, the following eight observations, tips and advice may be very useful before getting started.
Tip 1: Set proper expectations for AI by using the proper definition . Forrester recognizes two types of AI: Pure and Pragmatic. The pure AI strives to imitate comprehensive human intelligence (just like we see in science fiction movies), but this is far from becoming a reality at this point. Pragmatic AI, on the other hand, is narrower in scope and complements human intelligence. Focus on Pragmatic AI powered by machine learning. Why? Because machine learning algorithms analyze data and create models that make a prediction, take decisions, or identify the context and supplement what we know about a problem, and therefore we can make a more intelligent decision.
Tip 2: Choose more than one high-ROI use case . Don’t just focus on one use case for your first AI project, because you may not be successful with just one. Cast the net out larger, and run parallel use cases for your use cases. You will be more successful that way, and will likely be able to show ROI as a result.
Tip 3: Insist on comprehensive access to enterprise data . Work across teams to get the right set of data you think might work on those multiple use cases. The old adage of “garbage in, garbage out” applies to data as well. More data and better data leads to better results.
Tip 4: Go faster with AutoML . Data scientists are really good, but what makes them even better is having the right tool (like Driverless AI) to be more efficient. Auto-ML solutions dramatically compress the model building lifecycle, feature engineering, algorithm selection, evaluation, tuning, and model deployment.
Tip 5: Know when to quit . Machine learning is not guaranteed to work for determined use-cases. So, if the data doesn’t fit, you must quit. Go work on a different use-case (That’s why you need to identify multiple use cases – tip 2)
Tip 6: Keep production models fresh . ML models are probabilistic, that means that they are based upon the data that they were trained on, and therefore, over time they will decay. You will need to continuously monitor, retrain, and often rebuild the model And if you are going to do that at scale, you have to protect against undesirable results. You also need to have some form of DevOps collaboration to have a repeatable model deployment process.
Tip 7: Get business and IT engaged early . A machine learning model is only successful if it is deployed. It impacts the business processes and applications and there has to be an assessment of how it will impact design and development too. Work with the business leaders to identify which problems you are trying to solve.
Tip 8: You don’t have to do what the model tells you to do in an application . Trust in AI is a big topic right now. What if the model is wrong? Machines are amazing learners, but so are humans. And it can be a case of garbage in, garbage out. Bad models lead to bad results. Keep in mind AI becomes a lot smarter when there is a collaboration between the humans and the AI, so the path is to bring human expertise in the process to govern, improve and continuously inspect models.
Bottom line, AI and machine learning are new tools for driving a competitive edge for your business. Heed these 8 tips, and you will be successful.
