Key Challenges

      I opened this chapter with a Stephen Hawking quote, “Success in creating AI would be the biggest event in human history.” Hawking immediately followed that up with, “Unfortunately, it might also be the last, unless we learn how to avoid the risks.”

      AI isn’t without its challenges and risks. For one thing, there are potentially huge risks for society and human life as we know it (particularly when you consider some countries are racing to develop AI-enabled autonomous weapons). But let’s focus on the key challenges that everyday businesses will have to overcome if they’re to deploy AI successfully.

      Regulation

      There will no doubt be regulatory hurdles to negotiate as regulators begin (quite rightly and belatedly) to take a greater interest in the application of AI. Until now, some of the early adopters of AI have played a bit fast and loose with the technology (Facebook, for example, is facing legal action over its use of facial recognition technology for auto-tagging photos, without gaining user consent).¹⁷ That sort of behavior can’t continue, and business leaders will have to take an ethical, responsible approach to AI.

      Privacy Concerns

Part of using AI ethically means making sure you respect individuals’ privacy, gain consent to use their data for AI applications, and make it clear how you are using their data. Again, this is where some big players have fallen short in the past. Amazon, for example, faced consumer outrage over the news that contractors were listening to people’s Alexa requests. Individuals could not be identified by their audio, and Amazon stressed the practice was necessary to help develop Alexa’s capabilities, but the fact remains that most users had no idea that anyone would ever hear their private audio. Amazon has since introduced a “no human review” option to its Alexa settings, which allows users to opt out of their audio being manually reviewed.¹⁸

      Lack of Explainability

      Remember I said that AI can now solve a Rubik’s Cube in just 1.2 seconds? Interestingly, the researchers who built the puzzle-solving AI can’t quite tell how the system did it. This is known as the “blackbox problem” – which means, to put it bluntly, we can’t always tell how very complex AI systems arrive at their decisions.

      This raises some serious questions around accountability and trust. For example, if a doctor alters a patient’s treatment plan based on an AI prediction – when he or she has no idea how the system arrived at that prediction – then who is responsible if the AI turns out to be wrong? What’s more, under GDPR (the General Data Protection Regulation legislation brought in by the European Union), individuals have the right to obtain an explanation of how automated systems make decisions that affect them.¹⁹ But, with many AIs, we simply can’t explain how the system makes decisions.

      New approaches and tools are currently being developed that help to better understand how AIs make decisions but many of these are still in their infancy.

      Data Issues

Put simply, AI is only as good as the data it’s trained with. If that data is biased or unreliable, then the results will be biased or unreliable. For example, facial recognition technology was found to be generally better at identifying white males than women and people of color, because a leading data set used to train facial recognition systems was estimated to be more than 75% male and 80% white – something that programmers were able to correct by adding a more diverse range of faces to the training dataset.²⁰ This means companies will need to ensure their data is as unbiased, inclusive, and representative as possible if they’re to get the best results from AI.

      The AI Skills Gap

      Finally, one area in which many companies will struggle is finding the right AI talent. There’s a shortage of people who can develop these complex AI systems – and what talent there is tends to be scooped up by the Googles and IBMs of this world. AI-as-a-service (AIaaS) could be part of the solution. AIaaS offerings from companies like IBM and Amazon allow companies to make use of AI tools, without having to invest in expensive infrastructure or new hires, which makes AI much more accessible to businesses of all shapes and sizes.