Business technology changes constantly, and every solution we implement has a lifespan. You may have been a Microsoft shop for the past thirty years, but your company is not currently operating on Windows 95. To run a successful company, you need to find advantages anywhere possible.
Right now, some of the most significant productivity advantages, useful features, and cost savings can be secured by implementing programs based on artificial intelligence and machine learning concepts. These solutions can be directed to solve the precise problems that have afflicted your company. But you need to understand what you are buying when you implement AI/ML algorithms, including their shortcomings and risks.
This is not Buck Rogers technology. Large companies are incorporating it throughout their enterprises. Many code developers are capable of creating effective AI. Many vendors have arisen in this space – some to license the finalized technology to you, and some to help you develop it yourself.
The best current practical definition I have seen is from an IBM paper which quotes a Stanford study, stating, “Artificial Intelligence (AI) is a science and a set of computational technologies that are inspired by—but typically operate quite differently from—the ways people use their nervous systems and bodies to sense, learn, reason and take action.” The paper notes that many companies are already using machine learning programs like virtual assistants, customer chatbots or fraud detection AI. Your company may benefit in very practical ways from AI and Machine Learning. This article assumes a basic knowledge of what AI can do for your business.
Once you have explored AI and are comfortable with the type of sorting, distinguishing and categorization tasks that AI/ML excels at completing, you should look for very specific, practical business problem that AI can resolve in your company. It is important that AI projects be tied to concrete objectives and clear business value.
Decide whether to develop the tools yourself. If you have software and database developers in-house, it is likely that some of them have already use machine learning tools to help complete jobs. They also likely know how to pull open source AI tools from Github to develop the AI for solving business problems. If you are concerned about owning the results of the development, and owning the output of the AI once it is working in your business, then building it yourself can provide some comfort. Even using commercial tools with in-house developers can reach the same goal.
On the other end of the spectrum, you may want to start by licensing the AI from a specialist. You may not own the machine learning program in this scenario and it won’t be developed specifically to solve your company’s particular problems, but that may not matter for you. As an example, Salesforce.com developed a solution that works well for thousands of companies. It turns out that sales and marketing generates similar issues for nearly everyone who undertakes them, so a general solution that may be customized around the edges could be your best solution. Your business problem may fall into this category, so that a universal solution is effective for you. Just make sure you have all of the rights to your own data and a way to carry on the business functions should you stop using such a service.
The middle ground involves hiring a vendor or contractor to develop a custom (or at least customized) AI solution to solve your business problem. In this case, most customer companies would want to own/hold 1) all relevant rights in the machine learning algorithm that was created in the process and 2) all of the output of that algorithm. The customer is unlikely to care for owning either the software tools used to generate that algorithm, or the databases used to train or test that algorithm. So don’t get hung up on controlling the vendor’s tools or databases, when the results of the development process will provide value, and also the results of the AI’s work for your company.
Ownership is one of many potential issues and risks for developing and using AI programs. But you also need to understand the provenance of the data and the tools used to make that product. For example, a recent FTC consent agreement forced the target company to stop using AI trained on an ill-gotten set of biometric data. To the FTC, the suspect nature of the data used to train the AI rendered the final AI product rotten – like the fruit of a poisonous tree. This case is currently an outlier, but maybe only because other rulings haven’t addressed the question of what to do with an AI program trained with tainted data.
And how do you know if the data is clean, properly obtained and otherwise untainted? You could capture your own data and use it for this purpose. Given that your company will be designing a machine learning program to solve problems particular to your business, then you may already hold the best training and testing data. If not, make sure that the vendor can speak to where the training and testing data sets came from, how the information was captured, and whether permissions of data subjects were required or obtained. If the vendor is providing the training data and development tools, insist on an indemnity for third party claims in case there was anything wrong with either set of AI ingredients.
Remember that developing AI and using AI both swallow huge amounts of data. This means not only will you need enough data to manage the job, but enough free storage space as well. New storage capacity is one of the many items that should be addressed ahead of implementation
These are only a few preliminary concerns, but if you are careful about solving practical and substantive business problems, then developing or licensing an AI solution may be the best technology available. It may be best to start small and simple, even with a beta test to see if the technology suits your business strategy. Just make sure you have the right advice so that your AI solutions don’t spawn entirely different problems for your company.