10 Things Learned From Deploying AI in Human Environments
Deploying AI in human environments requires some finesse, unlike the pure environments you often encounter in school. Cameron Turner of Datorium is an expert at blending the world of business with AI deployment and is here to tell us ten things he’s learned so far.
[Related Article: How to Integrate AI: Digital Adoption Plus Human Resources]
1. The Best Model Doesn’t Always Win
Clients don’t always care about the model type — for example, a project done for a beverage retailer for employer fraud. The daily ranking on the employees’ probability of committing fraud and going through the process allowed him to learn a lot about the organization. In the end, they had a problem with deployment, so they didn’t follow the ‘best” model. Instead, they developed an easier to deploy the model for production purposes.
The lesson: Sometimes, the customer determines what model is used because your models don’t exist in a pure environment.
2. Black Box Solutions aren’t always the best option
Your models need explainability in the realm of business, and without that kind of communication, your model may not be the best fit. The approach of choice can be determined by what provides the most communication about the outcome rather than the sheer accuracy.
The lesson: Consider the communications needs of the business before deploying black box.
3. AI Implementers Must Be Educators
You’ll always have to perform some sort of education task, even with simple models. You must know where your stakeholders are in terms of understanding and continue to monitor their stake in the process.
If you’re an extremely talented developer, you must learn the people skills to teach those around you about your models, both the system and the implementation. You should get your stakeholders into the mindset of data science to help reduce silos and help people break out of their current boxes.
The lesson: Cultivate an educator mindset.
4. Humans Ruin Everything
Your awesome build is going to be ruined by human error. You must consider how real humans will use your models to scale. It doesn’t matter how foolproof your model is, someone will still mess up pieces of it as the monitor and deploy. You must find ways to control for that tendency and cultivate an educational mindset to help offset the human factor.
The lesson: Nothing exists in a vacuum. Human intervention will always alter what you’ve built.
5. Humans Need to Feel in Control
No one wants a robot overlord. We want to feel like we’re in control of the outcome when deploying AI. When we’re managing a system, we don’t really want to be hands-off. It causes worry and encourages distrust. Make sure when you’re building the system that you allow the operator to have something to do even if it’s small.
The lesson: Hands off should never mean locked out.
6. Perception Trumps Truth
Data can be manipulated easily, so what people think could be more important than what the truth is. Humans aren’t entirely rational, so relying too much on pure data isn’t going to work. Also, sometimes AI is wrong, so human intervention is still needed in times of feedback. It’s not going to work to present all the data without helping the viewer come to the true conclusion.
The truth: You need to understand how data visualization affects the perception of the truth.
7. Politics Trumps Passion
You must understand that you can’t predict how people will use data. When we think inside of organizations, becoming a data protectionist can be a human reaction, but the open nature of data should remain an option. However, with the changing nature of data and our desire for a competitive edge, you never know when previously open data won’t be there anymore — plan for this.
The lesson: Strive for open data but plan for obstacles.
8. You Need the SME
Humans are vital for AI and learning. Subject matter expertise is vital to true data science, so you can’t expect to do great work without it. When you show up to your first job, you need to listen to these subject matter experts to put you on the right path. You’re the data expert, but understanding the framework for how your business or field works is a vital part.
The lesson: You don’t know everything despite the data.
[Related Article: Managing Effective Data Science Teams]
9. You only succeed if your Stakeholders Do
Again, data doesn’t exist in a vacuum. There are many ways you can take data that could equal success, but if your goals don’t align with the business or organization, you aren’t going to get very far. Pay attention to what the real goals are for your business and use that as a metric for your success.
The lesson: It’s your stakeholder’s victory, not yours.
10. Do What You Love and Love What You Do
There is so much opportunity out there, so find something you love to do. Right now, your skillset is universally needed, and finding what you love can be the trigger that launches your career. Give your love priority.
The lesson: There are so many great problems out there, if you don’t love your current one, find another.
Overall, deploying AI in human environments is a tricky part of being a data scientist, but it can also be the most rewarding.
Read more data science articles on OpenDataScience.com, including tutorials and guides from beginner to advanced levels! Subscribe to our weekly newsletter here and receive the latest news every Thursday.
