How your customers can start their AI journey

AI is everywhere in the news on TV, social media and websites. As with most technologies going through a hype cycle, there is a lot of noise in the marketplace and it is hard to zero in on exactly what AI is and how government agencies could use it

In simple terms, AI is a set of statistical and mathematical algorithms that use vast amounts of historical data to learn and make predictions into the future.

AI has been amongst us for more than a decade embedded inside many things we use. Some popular use case examples include the amazingly accurate spam filters in Gmail, Microsoft Outlook’s segregation of emails into clutter, focused and other streams, Google assistant, Siri, Alexa, Netflix, Amazon, and Tesla autopilot (the latest versions use machine learning to sense and turn on wipers too).

But the growth of AI is dependent on good data. We need good quality data for training AI models. IT systems were a boon for companies and agencies alike, moving them from a world of paper and human-intensive tasks to more automated systems, self-serve and in general a higher level of customer and citizen services.

But as IT systems grown and evolved they have gone often in independent directions. Agencies do not have standardized data taxonomy for even simple fields like a person’s name or addresses. For example, the Health and Human Services procurement systems have 11 different terms for a contract “award”. On our IRS contract, we learned that in one system a field called State did not refer to what we might think – the 50 states in the country, but the state of something else. This is not unique to government agencies but is common across large private sector companies as well.

To get over the proliferation of stove-piped data, a common solution used to be the construction of big “data warehouses” which brought together information from multiple systems into one database system. Unfortunately, when data is moved, it can get mangled, corrupted, changed or irrevocably transformed in ways that are not intended making that data quite useless.

Across both transactional systems as well as data warehouses there are numerous data quality issues including duplication of data, incorrect data, corrupted data to name a few. When the input data is flawed, it results in the GIGO problem or Garbage In, Garbage Out.

Another serious problem agencies face is fragmentation of data across the enterprise and data living in silos that are often inaccessible. Most agencies systems were developed in independent directions, often focused on running the mission function of one part of the organization. For example, a taxpayer may be handled by multiple siloed systems that deal with wages and income information, audits, enforcement or criminal investigation. It is very hard to get a 360-degree view of the main entity in question: the taxpayer who exists across these systems. System owners also zealously guard data in their systems and getting access is often tedious and challenging.

These are serious problems. To get outcomes we can rely on, we need at least some amount of good data with reasonable quality. Does this mean that no AI is possible before first cleaning, standardizing and combining all the data we have?

All is not doomed; there are opportunities to get started:

AI to augment, not replace

AI techniques are not a panacea for all your problems. They are one more tool (like IT and the web were) that is available to help. Think of AI as a means to augment and boost what you currently do rather than replacing what you do. Like the “intel inside” slogan, you can embed AI to make your systems and processes more “intelligent” and effective.

AI to (semi) automate

The biggest opportunity for AI techniques is increasing the levels of automation. Until Terminator arrives, AI is going to be a dumb mutt for a long time. As such it is well suited to automate tasks freeing up humans to spend their time doing more valuable things. This is already happening with the likes of Google Assistant which regularly gives you inputs on traffic that might impact your drive home, how long it takes to get to work or new restaurants in your area that you may like (based on your visits and reviews).

Intelligent Change

The most significant challenge to implementing AI in government will be the workforce. In any organization (not just government) change management among the people has been the biggest challenge even in the roll-out of Information Technology. At Brillient, we see this even today in rolling new interfaces to systems, people prefer what they are used to. Top-down mandates may face some success and then run into the reef banks of failure. The imperative is for intelligent change. Start with some pilots to show employees how their lives can be made easier, how the agency’s ability to execute the mission is improved and how it makes citizen’s lives better. Gaining champions in a crawl, walk, run progression is key to success.

Commoditization of AI

Even as it goes through the hype curve AI tools are getting commoditized. Google, Amazon, Microsoft and open-source groups have already created tools you can download and start building, deploying and running AI models. Without knowing a whole lot about AI algorithms or python programming a reasonably technical person can implement a no-code AI model with some success. Many pre-built models already come trained to recognize faces or objects using a Convolutional Neural Network (CNN) for instance. This commoditization is both an advantage as well as a disadvantage. For somewhat straightforward problems like image recognition (as explained later) these models offer a quick path to a solution. Applying models to perform data analytics in the hands of an amateur can lead to disastrous results. AI tools are like the Integrated Development Environments (IDEs) of the IT programming age. They are better used in the hands of AI and data-savvy technical people.

There are opportunities and pitfalls ahead for artificial intelligence. For a deeper dive and more details, download this report by Sukumar Iyer.