By now, you have become aware of the various effects that AI can have in different areas of our lives. The biggest recent success is TikTok: you’ve either been sucked into endless TikTok scroll sessions, or you’ve at least heard of this platform and its endless dance, prank, lip sync, or cooking videos. TikTok’s success lies in its algorithm of its “For You Page” – the algorithm is highly effective in retaining attention and because of this, TikTok’s reach is unmatched (19).
Naturally, businesses have been highly interested in understanding TikTok’s secrets, especially marketing companies. Of course, much of how the algorithm works is hidden – it wouldn’t be smart to expose the secret ingredients for success, just like the secret formula for Krabby Patties! What matters is that algorithms can be a formula for success. What works for a social media company can cross over to other businesses and markets. Businesses or whole markets can benefit from AI, as AI-based tools have been shown to improve effectiveness and efficiency.
In short, AI tools can support businesses in the automation of operations, informed decision-making, and enhancing productivity (20). Tasks that were traditionally done by people, such as recruitment and talent sourcing as well as customer-centric approaches can now be supported by or taken over by AI through improved tailoring. It’s no coincidence that the rise of AI systems has been tied to the Fourth Industrial Revolution, so a history lesson might be in order to understand what we see happening now. By learning about AI, you’re making history (21, 22)!
But, getting back to the here and now, AI systems and tools can help to reduce repetitive or manual tasks, increase the accuracy and speed of business practices, and give people the time and space to focus on other work that AI systems can’t do (yet…). There are many different ways to categorize AI tools for businesses, and one of the possible ways to do this is to look at how AI tools can support business activities:
Process automation: This first type is the least smart implementation of AI, as it isn’t programmed to learn and improve. Instead, it deals with automation of tasks that used to be done by people, such as data inputting and consuming information from multiple databases. This also makes it the easier type of AI to implement, because it can be relatively quick to adopt and less expensive because of its limited capabilities, leading to increased profits in the long run.
Cognitive insight: This type of AI deals with predictions, detections, personalizations, and modeling. It’s more complex and more costly to implement than the first type and can be used for fraud detection or automation or predicting consumer behavior. This type of AI can deal with larger datasets and more details, is trained on relevant data, and the underlying machine learning model can improve over time. This prioritizes AI-supported insights over traditional analytics.
Cognitive engagement: The third type of AI deals with interactions that were traditionally between people, such as customer support, internal communications, or healthcare, where customization is a key element for success. As customization can be intense and costly, AI can relieve much of that effort from people, who can then focus their efforts elsewhere (23).
These types of support through AI sound great, so why’s this not yet realized in all businesses? Well, unfortunately, or luckily, depending on your stance, the answer is that AI tools can’t fix every problem or AI tools aren’t the best or most realistic solution for a problem. This may be because of limited resources, lack of feasibility, or scaling challenges. Moreover, in some cases we have seen that AI-based decision-making actually increases bias or discrimination, showing that careless AI practices can result in significant problems.
Fortunately, you can also see how various players in the industry are trying to address these problems and create trustworthy AI. You’ve already learned about the key characteristics of trustworthy AI but, as you might have predicted, these requirements put additional expectations on businesses that are using or will use AI-based tools in their processes. You might have also thought about how some of the characteristics of trustworthy AI are required, at least partly, by legal mechanisms or laws. So why not focus on that?
Of course, it’s important for businesses to comply with legal requirements, but also any regulations that cover the markets in which businesses operate. For example, if a business provides an online service in Europe, that business will have to comply with the EU’s Digital Services Act and General Data Protection Regulation. If a business is based in the USA, but has European customers, they’ll have to comply with European regulations as well because these businesses have access to data from European citizens. With many more relevant regulations, and even more regulations to come, like the EU’s AI Act, you can see how complex the legal picture is turning out!
Now, in this course, you aren’t expected to learn all the ins and outs of different legal frameworks applicable to AI – but it’s important to be aware of legal requirements and the potential consequences of non-compliance. Instead, you’ll look beyond legal frameworks and requirements to get a broader understanding of practical consequences for businesses, people, and processes. There is more to trust and trustworthy AI than compliance!
Some of the problems addressed before look like they can be solved by adopting trustworthy AI. However, trustworthy AI isn’t a catch-all solution and might even raise new problems. As a result, businesses need to transform their practices to not just comply with legal requirements, but also develop ethical AI practices. You have read it before, but AI is the hottest of topics currently. This is great news for AI developers and researchers as it gives them many opportunities. However, AI’s popularity is also risky as it can lead to careless adoption and increased risks.
Implementing AI tools is already a difficult process, and factoring in trustworthiness could add another layer of complexity. In other cases, it’s just a matter of understanding which model we choose as an approach or knowing the potential security risks in the planning process and working around that. For now, the main focus for many businesses working on trustworthiness is the concern for more fairness and avoiding or reducing biases (24). Aside from potential legal issues, businesses have to consider their reputation! No business would last long without a customer base, and customers could boycott any company if it becomes publicly known that their AI tools are discriminatory. You’d think there’s no such thing as bad publicity, but this isn’t always the case!
It’s no secret that businesses need to be profitable to continue to exist. Think of Blockbuster, Toys "R" Us, or your favorite local shop or restaurant…but we digress.
What we mean to say is that profits and market risks are top of mind for businesses. If businesses don’t adapt effectively to market trends, they might reduce profits because of poor alignment. Trends can be deduced using analytics, so this means that problems with data collection, processing, and interpretation might result in wrong predictions, bad decisions, or unhelpful customer-support solutions. As such tasks are increasingly supported by AI tools, it’s important to make sure that your business collects the right data, makes valid predictions, and doesn’t infringe on human rights. And, you guessed it, this is where trustworthy AI comes in! Focusing on trustworthy AI practices reduces a lot of potential for mistakes or risks to businesses, so there is a clear benefit to business practice to invest time and expertise to develop trustworthy AI.
What can businesses do to enhance the trustworthiness of AI and their business processes? Well, for a start, they can ask themselves the following set of questions.
1) AI or no AI, that is the question! Or what’s the problem we’re trying to solve and is it solvable by AI? Before adopting AI, businesses should consider the problem that AI can solve for them, and can the tool actually measure the appropriate items? Think back to the issue of validity. And, to follow up on that, is AI the most optimal solution considering the costs and benefits? In the end, businesses should consider feasibility, profits, and market risks for longevity.
2) Which people and/or departments should be involved? When selecting the right AI-based tools, businesses should develop strong interdepartmental relations among the Purchasing department, Legal department, and IT and Security departments. Together, these teams can make sure that newly acquired tools are in line with laws and regulations – and also if these tools fit within the IT ecosystem. Moreover, some services might require third-party collaboration, which needs to be properly reviewed if it meets various benchmarks, for example with regard to processing agreements, data security, or maintenance.
3) What legal regulations apply to my business case? The implementation of AI tools requires careful consideration. This course doesn’t provide a deeper analysis of legal frameworks, but that doesn’t mean that businesses should avoid reading up on legal requirements! (25, 26) For example, the EU’s GDPR states that EU citizens have the right to have all decisions made about them made by a person. Therefore, the implementation process should have space for people to be involved, collaborating closely with the AI tools.
4) Who will be working with the AI tools? Businesses should consider the people who’ll work with the adopted AI tools. How could AI tools potentially affect the people who work with it, for example in terms of privacy or bias? (27) Employees will need the right skills and knowledge to work with high-level digital tools. Topics such as data analytics and statistics become especially relevant. In addition, developing ethical data practices within the whole organization is highly important (26). This requires businesses to train their current employees and to consider adjusting existing job profiles for hiring new staff.
5) How will we audit and control AI tools and their use? Businesses need to think of checks and balances that help audit and control the systems they’ll implement (25). For instance, auditing mechanisms like measurable criteria and logs help to keep track of the validity of AI tools. Remember the validity problem we discussed earlier? Similar principles apply to auditing!
This isn’t an exhaustive list as there are many more considerations that businesses need to include when implementing AI tools. However, this list gives a summarized overview of the topics businesses should take into account when adopting trustworthy AI. It may seem like a lot of effort for businesses to take up, but it comes with a lot of benefits: investing in trustworthy AI will increase the robustness and resilience of AI tools, which in turn reduces maintenance costs. Moreover, investing in the privacy of AI tools will ensure the security of the users, improving the relationship between businesses and their customers. All in all, trustworthiness is a worthy goal for businesses to achieve and well worth the effort.
