Here is an excerpt from the insideBIGDATA article by Greg Council, published April 7, 2016. For a full view of the entire article click here.
Significant advances in computing technology can yield less significant, yet still more successful and practical applications for businesses. And yet, it’s easy to become enamored with technology for technology’s sake. Successfully applied artificial intelligence requires taking small steps to learn and adapt.
AI is not as smart as you think
Breathless headlines trumpeting cognitive computing, machine learning, and artificial intelligence are everywhere, but few articles actually discuss what it all really means. For all the hype about similarities to human thinking, artificial intelligence is not all that intelligent. Yet. Keeping this in mind is the key to successfully applying AI to any business. The underlying concepts of AI can be placed under the umbrella of “machine learning,” which combines a number of technologies aimed at automating tasks that typically require humans at the center or, at a minimum, human interaction.
Just like us, machine learning needs inputs. It then needs guidance on providing the right outputs. Machine learning algorithms don’t know how to experiment like we do. Left without any guidance on inputs, humans gradually figure things out through trial and error. AI needs tons of data and guidance on what that data means in order to produce useful output at anywhere near the performance level of humans. Google’s self-driving cars are a case in point on needing a lot of data to solve a complex problem. Google has spent years on the technology driving cars several million miles in the process to obtain that data and feedback. Most businesses do not have the resources to supply that level of data and feedback.
The Difference is in the Scope: Applied AI
Let’s talk about how applied AI works using an expense management scenario. A company typically establishes spending policies and requirements for employee reimbursement. Keeping the receipts involved is mandatory to show proof of the expenditure as well as to approve each expense. Most organizations cannot automatically apply policies. The process requires a lot of effort on the part of employees submitting expenses and the staff who review them.
Now let’s introduce applied AI. Almost every company has many samples of receipts along with expenses that were approved or rejected and their expense categories. The receipt samples and results for each expense can be input into machine learning algorithms to answer two questions:
- What expense category does each receipt belong to?
- What receipts are approved?
Because this information is already answered for each receipt (based on expenses already processed), the software locates data and identifies patterns in it to make inferences and answer each question. For instance, expenses for the category “dinner” above $50 are repeatedly not approved. The software identifies this pattern and automatically applies a rule for assigning an expense to the category of “dinner” and then determines if the expense can be approved. Going further, if the expenses for the category of “dinner” always have exceptions for alcohol, the software detects this pattern, identifies terms associated with those items, the amounts, and then automatically applies this outcome as well.
Here applied AI focuses on a specific problem and performs two functions: Assign expense categories and determine whether expenses are approved. This kind of problem is perfect for the level of AI that is commercially available and, while not creating earth-shattering changes, can have a dramatic effect on a time-consuming business processes.
Approaching the use of artificial intelligence requires identifying the problem, keeping the scope narrow, and then focusing on improvements that are achievable.