I came across a Quora question recently that asked, “What makes AI projects fail?” It’s a valid question: The vast majority of artificial intelligence (AI) projects fail, as reported by Harvard Business Review.
Copyright by www.forbes.com
It’s a bit of a trick question, though, because there are as many reasons for failure as for success, so there’s no single, all-encompassing answer.
That being said, here are four common mistakes that kill AI projects. Avoid them, and you’ll be more likely to succeed.
Humans have a “complexity bias,” or a tendency to look at things we don’t understand well as complex problems, even when it’s just our own naïveté.
Marketers take advantage of our preference for complexity. Most people would pay more for an elaborate coffee ritual with specific timing, temperature, bean grinding and water pH over a pack of instant coffee.
Even Apple advertises its new central processing unit (CPU) as a “16-core neural engine” instead of a chip and a “retina display” instead of high-definition. It’s not a keyboard; it’s a “magic keyboard.”
It’s not gray; it’s “space gray.”
The same bias applies to artificial intelligence, which has the unfortunate side effect of leading to overly complex projects. Even the term “artificial intelligence” is a symptom of complexity bias because it really just means “optimization” or “minimizing error with a composite function.” There’s nothing intelligent about it.
Many overcomplicate AI projects by thinking that they need a big, expensive team skilled in data engineering, data modeling, deployment and a host of tools, from Python to Kubernetes to PyTorch.
In reality, you don’t need any experience in AI or code. You can use no-code AI tools like Obviously AI or Intersect Labs to get models up and running in minutes.
Most organizational AI use cases revolve around optimizing and predicting a certain column in a data table — something like absenteeism, attrition, churn, conversions, traffic or fraud.
However, you’ll find that many people don’t understand how AI can be used. If you Google “we want to use AI to,” here are some of the results:
- • We want to use AI to bring in cognitive computing capabilities.
- • We want to use AI to reduce inequality.
- • We want to use AI to solve complex problems that we wouldn’t otherwise be able to solve.
- • We want to use AI to secure jobs and to raise the standard of living.
- • We want to use AI to make the world better.
These are all terrific intentions, but they’re all ambiguous. You need to be hyper-specific on exactly how you’re going to use AI to accomplish your goals.
At it’s simplest, you need to know what data column you want to optimize. If it’s a meaningful key performance indicator (KPI) for your organization, then you’ll also be more likely to succeed by following through with the rest of the implementation.
3. No Follow-Through
Suppose you’ve come up with a clear idea, like reducing churn, and built a model. The thing is, a model alone isn’t enough. You need implementation.
Even the most accurate model in the world won’t help you if it’s sitting on a server somewhere, not making decisions and improving the bottom line.
For example, personal loan apps could use a predictive application programming interface (API) from an AutoML tool to serve predictions to users. Don’t just make predictions. Act on them.
4. Lack Of Data-Driven Culture
If your company has a traditional culture based on making instinctual, gut-feeling decisions, then management probably won’t defer to the data.
This actually relates to the first mistake of overcomplexity. If the organization feels that AI is too difficult to pursue, that will be reflected in the company culture. […]
Read more: www.forbes.com