Over the years, I’ve looked at business tools for my organization and have been asked to create solutions to bridge process gaps across the different technology tools. Today is much different than several years ago, however, eerily familiar. I find myself asking the same questions that I was being asked when I was in the field of software development yet looking at the problems with a different lens on every topic from implementation to impact.
Today, many applications are utilizing machine learning, but we don’t always know how or to what capacity. How do we integrate machine learning at our business? How do we get from A, not using any supplemental machine learning tools to B, full integration and adoption?
Although not widely realized or known, many tools we utilize every day already integrate many facets of machine learning, we just don’t see it. The algorithms behind intelligent search, suggestions on Netflix, Social Platform Censorship, and even Microsoft PowerPoint, are all powered by machine learning. Most of the time when you see a PowerPoint referencing A.I. What you really want to know about is the machine learning. So, what’s the difference between machine learning and AI?
Artificial Intelligence is more of a buzzword these days, you can buy a domain and claim to use AI in a presentation, but in all honestly, there’s probably a good amount of fluff mixed in. Many claim to use AI but have minimal achievements. In application development, especially with AI and Data Science, we can often put the cart before the horse. Most AI today is a pitch in a slide deck. More recently, we’ve seen models as a series of regressions or neural networks, attempting to solve some really complex decision-making issues. But at the enterprise business level these tools have a long way to go to reach unencumbered adoption.
The reality is that in order for AI to be true and robust, it would need to enable in-the-moment decision making with an influx of new data all the time. The entire artificial intelligence module must be embedded at every layer of the company’s ecosystem. To be truly effective, it would have to live in the Operating System (OS) of your computer, your e-mail, your devices, and company work product. It would have to integrate back with cloud services in real-time and analyze potentially terabytes of data in seconds. It would need a feedback system that tells it whether it did a good job or not, and some serious looks at how often re-training of the models is necessary.
Today, this is not only possible, it’s frequently happening at every level of the industry. However, many tools that sit on top of those big players simply don’t have the in-depth subject matter expertise to solve the problem. Maybe the model is too broad, or maybe the data isn’t internally consistent, it might not get you the answers you need when you need them. Don’t get me wrong it’s here already but more so for the bigger market players than the niche consulting or legal firms.
While looking at these products and tools there’s a few things we should especially look out for, and questions to ask:
- Vet the Executive Team: Evaluate startups and review backgrounds. Look for technical expertise and many years of statistics and data science background. Bonus points for specializations in certain fields and certifications that are specific to you.
- Is it internally consistent? Ask some basic technical questions, and some tough ones. There are AI products that are totally bogus out there. Talk to the vendors about where they get the data they use in the models, what models and techniques they use, and how they validate the model. What would you check in our data to ensure the models work the same?
- Consider the fit within the organization. Can all the people at my organization use the tool, or is it specific to a select few “power users”? It’s not enough to just ask someone in the tech team if this will solve our problems, you need a team if you want it to work for a team. Start slow, with a use case. Some early adopters and supporters who share in your frustration may help bring the solution to life and assist in implementation. You need allies.
- Will the product evolve with the organization?
Now more than ever the importance of explaining business outcomes hits home for me. Next time you see a company saying they do AI, ask them questions, make them prove it.
Ask about these processes to ensure that the data your next tool plans on integrating is being used in a way that makes sense to the business. Where does your data live? How will this tool extract it and how frequently?
Finally, are the models robust enough to actually get you an answer? Many models out there are broad in nature and only provide probabilities.