Many businesses are in the midst of exploring key strategies and approaches for incorporating AI into their digital strategies, while making it explainable, ethical, and productive. Rick Madigan, UNRVLD Lead Strategist, shares UNRVLD’s thinking on the matter.
One of the key dangers is the inflated hype around AI. It can be incredibly useful if used in the right way, but lots of organisations are jumping in without understanding what, why and how to manage its superpowers with due care. This inevitably leads to errors, to mistrust and poor satisfaction around the technology. It’s not AI’s fault in any way. It’s the approach that’s taken.
To get the most from this revolutionary new technology, we need to step back and consider three perspectives: Productive, Explainable and Ethical.
1: Productive AI
- What is the problem itself?
- Why is the current process and / or technology failing?
- What is the data (quant and qual) saying?
If we discover there is a solution that AI can fix and there is a suitable off-the-shelf product available to do so, a lot of training will already be baked into the product. An example is Paradox for conversational AI on careers sites. In such instances, the parameters of what it can or should be doing will already be ringfenced so you could go straight to a full configuration and launch. To mitigate risk, however, we would usually recommend you opt for a soft, beta launch first.
Conversely, if our recommendation is for you to leverage a conversational model or an AI framework like ChatGPT or OpenAI, caution needs to be aired. In this instance, we will take the problem we have identified, then either develop an AI solution to address that problem, or, if the problem is too big, break it down into prioritised elements of the problem and create a solution to address the highest priority.
Despite the fact that you will be getting the base tools you need out of the box, you will essentially be handing your service to children who have been taught basic concepts but have no further training. They won’t have any understanding your business or purpose, or the nuances and formalities of communication and operation within your industry either.
Launching such solutions without due diligence is akin to a commercial grenade. To use these tools effectively you need to invest. And that can be pricey. So, as a strategy at UNRVLD, we are likely to advise you to opt for development of the solution as a series of test and learn innovation cycles and outputs. We then quickly train, develop and launch the tool to a set audience with set guidelines. On analysis, you may then choose to throw it away, amend it and re-test or decide to go into production. This mitigates risk and ensures you are focused on value.
This approach is critical to delivering Productive AI because we’re focused on creating something that actually both solves a problem and delivers value.
2: Explainable AI
If we are building you an AI solution from scratch, we adopt the same process as above, but the models or frameworks we create will be built within the overall solution design. Whether applying conversational AI or something else, applying explainable AI (XAI) processes and methods effectively via flow diagrams and visual representations will make it easier for your own staff to understand, and more importantly, trust the solution. This also helps them manage the results and output the solution kicks out. Accuracy, fairness, transparency, biases and expected impact all have to be assessed in the context of what we’re building.
The risk here is the XAI documentation produced can quickly become incredibly complex. There will always be technical people involved in your project who need full detail. But depending on the nature of your application, other audiences will only need simplified versions to gain a relevant level of understanding.
The added dimension to XAI here is the end user. Explaining AI to them entirely depends on what the AI solution is doing. If we’re thinking about solutions that help suggest the most relevant products (e.g., loans, schemes, courses) to a user, we need to be able to “show our working out”. If we’re simply retrieving information from a knowledge base or from a user’s account, there’s less “working out to show”. It ties back to the concept of Productive AI and being crystal clear on the problem we’re trying to solve, how we’re solving it (the solution design) and the mindset and expectations of the user.
In short, Explainable AI hinges on transparency, a cornerstone of our third perspective.
3: Ethical AI
All the usual ethical design principles stand when it comes to AI – inclusivity (build for needs; solve for one, deliver to many; reducing cognitive load; progressive enhancement over graceful degradation), sustainability, data, etc.
But we have extra ones to throw into the mix:
Transparency - it’s easy to be clear about what data you’re capturing and using on a website. People, to an extent, get it as websites have been around for a long time. But there’s a lot of misconceptions and a lack of knowledge on new technology such as AI. We need to be clear on what data is captured and used by the technology, what the technology is and the boundaries and limitations of that technology.
Cultural bias - can manifest in many ways, including accents, cultural sensitivities, religious beliefs, racial profiling and beyond. AI enables us to create richer and more engaging experiences as well as communicate in new and exciting ways. But, there have been many cases of AI demonstrating cultural bias and creating some highly damaging and offensive situations. Cultural bias has to be addressed immediately whenever AI is brought in.
Plan B – Not everyone will trust or want to use this technology, no matter how transparent you are, how user friendly you are or whether you are solving a problem. We always need a fallback to ensure we meet the needs for everyone, delivering an inclusive solution.
AI has the potential to create incredible digital experiences. By carefully assessing your challenge and considering the application of AI from all three perspectives, you can realise that potential.