Challenges of AI and Algorithms

While Artificial Intelligence and Algorithms have many benefits, they also come with considerable challenges. The governance and ethics of AI and Algorithms will dominate public discourse for several years to come. Some challenges related to these attributes include:

• Garbage in/garbage out. One of the oldest truths in computing is that the quality of the result depends upon the quality of the data supplied. Bad input data does not somehow magically turn into accurate, reliable, and actionable results.

• Not sure what it’s doing and how it’s working. AIs and Algorithms are often a “black box” about which the user knows next to nothing. As tools like ChatGPT become prevalent, decisions will be made and/or implemented with few clues about the underlying process. Thus, organizations risk using good data but getting bad results or, worse, seemingly good results that don’t actually meet their needs.

• Hallucinations. As ChatGPT and GPT4 have come online in 2023, the term “use but verify” and Hallucinations have come into vogue. Often, AI spits out inaccurate information. For example, Peter queried GPT4 about “the top 5 awards and recognitions he had received,” GPT4 delivered three that were correct and two that were false (although conceptually on point since they were in the space field).

• Lack of quantifiable effects. Artificial Intelligence can be very appealing to implement. It’s trendy, can be very efficient, and it looks good to customers and shareholders. But it can also be a black hole for investment. Hold off on AI and Algorithms until you can find definitive, empirically tracked positive effects on productivity and ROI.

• Ethics. Artificial intelligence is a computer-based technology. Its behavior is entirely defined by its programmers and/or the large data sets it was trained on, and its current algorithms make no allowance for human emotion, needs, or freedom. This limitation, combined with its ability to function at superhuman speeds, makes it a powerful, world-changing tool. And the same qualities can also make it intrusive, controlling, and even dangerous if left unmonitored. 

• Liability: Beyond the moral implications of AI, note that you and your enterprise—not your AI—will be held responsible for any damage that it incurs.

• Governance. AGIs communicating and working with each other to solve the world’s problems is a hugely appealing notion. But that scenario could also prove to be the stuff of science-fiction nightmares if these new systems are inadequately controlled, or their operating functions are misdirected. As we progress further into the Age of Artificial General Intelligence, these challenges will inevitably multiply.

• Deepfakes. These AI hoaxes use deep-learning technology, including neural networks, to swap faces on videos and other digital content for propaganda, entertainment, and criminal purposes. The widespread use of deepfakes has already begun to undermine the public’s trust in video imagery.

• Negative public perception. There’s a persistent public fear that robots or AI will replace workers, putting people out of jobs and driving poverty. Dystopian depictions of robots/AI in movies (think 2001: A Space Odyssey, Terminator, The Matrix) don’t help.

• Data availability and management. With exponentially increased amounts of data come exponential increases in data management. How will your organization access data (partnerships, open sources, etc.)? How will data be accounted for: as an asset similar to cash, plants, facilities, and customers? How will that data asset be managed? Will your organization focus first on finding good data over building more complex models? 

• Legal. Is your organization prepared to deal with privacy and liability issues? What risks are you willing to take while waiting for legislation to catch up with faster-moving technology? 

• Modeling. Can you use your customer data to train a model? If a customer opts out (as per the European Union’s general data protection regulations (GDPR)), do you need to re-train those models?