AI’s role in modern business: An interview with Elin Hauge

In the second part of an interview with Jennet Siebrits, Head of UK Research at CBRE, and AI strategist Elin Hauge, Elin offers her insights for business leaders navigating the AI landscape. Continue reading as Elin delves into the pivotal importance of data, the alignment of human and machine collaboration, and the essential criteria for selecting AI vendors.

What are your key messages to business leaders looking to adopt AI?

The first thing is to remember AI is not a strategy, it’s a toolbox. This means that the first step has to be to define which problem to solve. With this problem in mind, look into the AI toolbox and find the tool that enables you to solve your problem. This may be a not-so-fancy but well-tested prediction model rather than generative AI. Don’t look for the shiniest hammer, that only turns all problems into shiny nails.

The second thing to remember is the importance of data. Artificial intelligence is about mathematical algorithms applied to large amounts of data. This means that data about your business operations is a strategic asset. These data are digital representations of your business operations. Hence AI is about utilising the powers of your strategic asset – the data. It’s not magic. And it’s not a silver bullet. You will have to go the tedious way through data engineering, data quality, and data integrity.

What is your key message for business leaders not looking to adopt AI?

Wake up! AI is the natural next step of digitalisation. You have data, you have digital systems and processes, and you have access to the required processing power. It makes sense to use mathematics applied to your data and processes to automate processes and decisions and optimise outputs. This is what AI as we know it today is about. It’s not going to disappear.

How can businesses align humans and machines for the best possible outcome?

First of all, we need to stop anthropomorphising AI. We have always done this. In earlier times, humans assigned divine powers to e.g. stones and mountains. Now we do it again – with computers – and it creates unnecessary friction around organisational adoption.

I keep coming back to the term ‘toolbox’. If we consider AI as a toolbox, albeit an advanced toolbox, it is much easier for the organisation to accept the role of these technologies as part of their everyday work. The last two years have also clearly demonstrated that the extremely hyped expectations for generative AI to take over almost any job is not realistic. On the other hand, many more jobs than we were even able to imagine a couple of years ago have now some component of AI in them. Just think about Microsoft’s Copilot.

At the hospitals in Vestre Viken Hospital Trust in Norway, they are now using AI for CT diagnostics. You might immediately think that this means replacing the radiologists. However, if you look at the process of getting a CT scan from the patient’s side, many CT scans are done to rule out a fracture. At Vestre Viken Hospital Trust, they use AI to make decisions on which patients do not have a fracture, allowing them to be sent home rather than waiting for a radiologist to analyse their scan. The radiologists can then prioritise their time on the patients who really need them. Ultimately, the patients spend less time waiting and the radiologists spend more time applying their expertise where it is most needed.

To maximise the value of AI, we need to stop fighting this immediate fear of replacement and rather explore the many ways that this powerful toolbox can enable us humans to do more, better, and in new ways. For top management, though, this means that they need to resist the urge to use AI as a a proxy for downsizing. If you need to downsize, do it for the right reasons with the right arguments, not under the cover of tech hype.

Another example of this is the current urge to replace human agents in call centres with agents based on generative AI. Claims are that customers are frequently happier with the AI agents than with human agents. However, what’s better than great customer service? No customer service! My point is that AI may also be used to identify and solve high-frequency problems before the customers experience the friction in the first place, leaving the human agents to spend time on the service calls where the human interaction really matters.

What should businesses look for when selecting AI vendors and technology partners?

Don’t buy hype. Buy solutions to real problems. To be able to do that, you as a buyer need to know the basics of what AI is and is not, and you need to have some level of understanding of its limitations.

AI technologies also have substantial environmental footprints, related to the data centre processing requirements. I would highly recommend vendors and partners that have a realistic understanding of these challenges, and that provide balanced perspectives.

You should also look for partners that are prepared to help you build your internal capabilities as a part of your collaboration. AI is the natural next step of digitalisation, and you will need both competence and capacity internally over time.

Does using AI open a business up to potential cybersecurity threats? How can these be mitigated?

The more digital and connected the world becomes, the more vulnerable businesses become to cybersecurity threats. AI brings, in particular, three new dimensions of cybersecurity threats to the table:

1. Deep fakes enable scamming and phishing at a whole new scale, and we will all be fooled at some point. In business, the closer you are to money or secrets, the more likely it is that you will become a victim.
2. Generative AI is vulnerable to malicious prompt engineering, such as prompts that trigger malicious attacks behind the perimeters of the business.
3. Data pollution causing AI models in operation to go haywire.

Cybersecurity is no longer an issue for the CISO and the IT department, it concerns everybody. Mitigation involves everybody in the organisation who has some form interaction with the company’s systems. Humans are always the weakest link in cybersecurity. This is even more true in the era of AI, as the established borders between humans and technology becomes blurry. There is no one single solution to the mitigation of this risk, but a core component is knowledge, all the way up to the board level. Without knowledge, you are blind to the risks.

What are the cost drivers of AI?

All tech tools come with a price tag, but there are several ways to make AI a particularly costly affair. First of all, building generative models is in itself an extremely expensive endeavour due to the immense computation requirements for the training of the models. This, in addition to the need for enormous amounts of training data, is the key reason for why there will only be a few LLM providers in the market. Secondly, using LLM-based enterprise applications may also demonstrate to be costly. The business models are still being tested, and right now, the cost of training and operating far surpasses the revenue streams from customer subscriptions. However the methods, algorithms, and architectures have improved to such an extent that we are going to see many more providers of SLMs, i.e. Small Language Models. These are cheaper to develop, and also require less computation power for operation. This brings me to an important point for leaders who “need” an LLM for their enterprise applications; you may not need a large language model. An SLM may be sufficient for your needs.

If we move our focus to prediction models of various kinds, i.e. non-generative AI, the cost drivers are most of all the human resources and the systems needed to deploy and operate these models. Should you build, buy, or a combination? That depends on what type of tool you need. In essence, the value of AI is not in the mathematical algorithms, but in the data used for training the model. AI requires access to data. Data engineering, the plumbing of your strategic assets (aka data), may cause a costly headache if you have ignored it up until now. However, continuing to ignore it will cost you more in the future. It’s a bit like ignoring the not watertight plumbing under your kitchen sink only because calling in someone to fix it feels expensive.

In simple words, all tools come with a price tag, and all purchase decisions need to be supported by a realistic business case. The essence, though, is whether you are ready to acknowledge that mathematics applied to your digital value chain may give valuable insights, optimisations, and automations.

Last but not least, regulative compliance requirements may also incur costs. However, if you keep your house in order and make sure you have proper control of your data plumbing and integrity, this is not likely to be an unaffordable issue.