AI and NFPs

Artificial Intelligence (“AI”) has made amazing progress in recent years and has had  an incredible impact on the business world. Scanning resumes and making credit decisions are two examples of common AI utilizations that come immediately to mind. Even NFL teams are using it to evaluate players.  As time goes on, it will become more and more ubiquitous. Small NFP organizations  often lack the funds to deploy this technology, but as its cost comes down NFP entities will need to utilize AI to efficiently achieve their missions. However, before they do, management will need to understand that AI is not the panacea for all of its problems. The sad truth is that AIs often do make mistakes.  How can I prove this bold statement? Let’s look at the world of chess. 

The internet is ablaze with the story of AlphaZero, currently the best chess player in the world. AlphaZero is an AI owned by Google that trained itself to play chess.  It was given the rules of the game and then played 44 million games against itself to become the best player in the world. AlphaZero played two historic matches against Stockfish, a brute force chess engine that calculates millions and upon millions of moves  a second.  The first match was 100 games.  AlphaZero won 28 games and 72 games were drawn. This would be a staggering achievement for a human player. For technical reasons we don’t have to get into here,  many discounted this smashing victory.  The technical deficiencies were corrected in the second, longer match of 1,000 games. AlphaZero won 155 games, drew 839, and lost 6. This was another very lopsided match, but AlphaZero did make mistakes since it lost games.  

 AI can make mistakes. They are not infallible. (Of course, those of us from the Baby Boomer generation do remember Terminator and HAL 9000…) Granted, AlphaZero had a very limited number of losses but these were only the discernible mistakes.  There is a second type of mistake. AlphaZero may have made inferior moves earlier in drawn games it was able to compensate for later in the game.  Put another way, AlphaZero could have won the game but it only ended up with a draw. A third type of mistake could occur when AlphaZero played an inferior move and then had to fight for a draw. The latter two types of mistakes are often overlooked because we tend to focus on the games won and lost.  Drawn games are often insufficiently analyzed. 

What lessons can NFP management learn when they begin installing AI? NFP organizations need to understand their tolerance for errors and omissions mistakes and put appropriate internal controls in place to insure against them.  After all, many NFP organizations deal with human services.  Six mistakes out of 1,000 can still be devastating when dealing with human lives. 

In short, AIs are very efficient assistants, but they can make mistakes. Sometimes these mistakes can be hard to find. An organization employing AI needs to make sure it has sufficient safeguards and internal controls in place to make sure these errors are not dangerous to the organization’s mission.